Update readme.md

function calling for python. already had ts.
Signed-off-by: Matt Williams <m@technovangelist.com>
2023-11-29 10:45:07 -08:00 · 2023-11-29 10:06:11 -08:00
139 changed files with 3017 additions and 7302 deletions
--- a/.dockerignore
+++ b/.dockerignore
@@ -2,7 +2,8 @@
 ollama
 app
 dist
-llm/llama.cpp
+scripts
+llm/llama.cpp/ggml
+llm/llama.cpp/gguf
 .env
 .cache
-test_data
--- a/.gitignore
+++ b/.gitignore
@@ -8,5 +8,4 @@ ollama
 ggml-metal.metal
 .cache
 *.exe
-.idea
-test_data
+.idea
--- a/.gitmodules
+++ b/.gitmodules
@@ -1,4 +1,10 @@
-[submodule "llama.cpp"]
-	path = llm/llama.cpp
-	url = https://github.com/ggerganov/llama.cpp.git
-	shallow = true
+[submodule "llm/llama.cpp/ggml"]
+    path = llm/llama.cpp/ggml
+    url = https://github.com/ggerganov/llama.cpp.git
+    ignore = dirty
+    shallow = true
+[submodule "llm/llama.cpp/gguf"]
+    path = llm/llama.cpp/gguf
+    url = https://github.com/ggerganov/llama.cpp.git
+    ignore = dirty
+    shallow = true
--- a/6
+++ b/6
@@ -19,11 +19,5 @@ RUN apt-get update && apt-get install -y ca-certificates
 COPY --from=0 /go/src/github.com/jmorganca/ollama/ollama /bin/ollama
 EXPOSE 11434
 ENV OLLAMA_HOST 0.0.0.0
-
-# set some environment variable for better NVIDIA compatibility
-ENV PATH=/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
-ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64
-ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
-
 ENTRYPOINT ["/bin/ollama"]
 CMD ["serve"]
--- a/Dockerfile.build
+++ b/Dockerfile.build
@@ -1,101 +1,31 @@
-ARG GOLANG_VERSION=1.21.3
-ARG CMAKE_VERSION=3.22.1
-ARG CUDA_VERSION=11.3.1
-ARG ROCM_VERSION=5.7.1
-
-FROM --platform=linux/amd64 nvidia/cuda:$CUDA_VERSION-devel-centos7 AS cuda-build-amd64
-
-ARG CMAKE_VERSION
-
-RUN yum install -y https://repo.ius.io/ius-release-el7.rpm centos-release-scl \
-    && yum update -y \
-    && yum install -y devtoolset-10-gcc devtoolset-10-gcc-c++ git236
+# centos7 amd64 dependencies
+FROM --platform=linux/amd64 nvidia/cuda:11.3.1-devel-centos7 AS base-amd64
+RUN yum install -y https://repo.ius.io/ius-release-el7.rpm centos-release-scl && \
+    yum update -y && \
+    yum install -y devtoolset-10-gcc devtoolset-10-gcc-c++ git236 wget
+RUN wget "https://github.com/Kitware/CMake/releases/download/v3.27.6/cmake-3.27.6-linux-x86_64.sh" -O cmake-installer.sh && chmod +x cmake-installer.sh && ./cmake-installer.sh --skip-license --prefix=/usr/local
 ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH

-ADD https://github.com/Kitware/CMake/releases/download/v$CMAKE_VERSION/cmake-$CMAKE_VERSION-linux-x86_64.tar.gz /tmp/cmake-$CMAKE_VERSION.tar.gz
-RUN tar -zx -C /usr --strip-components 1 </tmp/cmake-$CMAKE_VERSION.tar.gz
+# centos8 arm64 dependencies
+FROM --platform=linux/arm64 nvidia/cuda-arm64:11.3.1-devel-centos8 AS base-arm64
+RUN sed -i -e 's/mirrorlist/#mirrorlist/g' -e 's|#baseurl=http://mirror.centos.org|baseurl=http://vault.centos.org|g' /etc/yum.repos.d/CentOS-*
+RUN yum install -y git cmake

+FROM base-${TARGETARCH}
+ARG TARGETARCH
+ARG GOFLAGS="'-ldflags -w -s'"
+
+# install go
+ADD https://dl.google.com/go/go1.21.3.linux-$TARGETARCH.tar.gz /tmp/go1.21.3.tar.gz
+RUN mkdir -p /usr/local && tar xz -C /usr/local </tmp/go1.21.3.tar.gz
+
+# build the final binary
 WORKDIR /go/src/github.com/jmorganca/ollama
 COPY . .

-WORKDIR llm/generate
-RUN sh gen_linux.sh
+ENV GOOS=linux
+ENV GOARCH=$TARGETARCH
+ENV GOFLAGS=$GOFLAGS

-FROM --platform=linux/arm64 nvidia/cuda:$CUDA_VERSION-devel-rockylinux8 AS cuda-build-arm64
-
-ARG CMAKE_VERSION
-
-RUN dnf install -y git cmake
-
-WORKDIR /go/src/github.com/jmorganca/ollama
-COPY . .
-
-WORKDIR llm/generate
-RUN sh gen_linux.sh
-
-FROM --platform=linux/amd64 rocm/dev-centos-7:$ROCM_VERSION-complete AS rocm-build-amd64
-
-ARG CMAKE_VERSION
-
-RUN yum install -y https://repo.ius.io/ius-release-el7.rpm centos-release-scl \
-    && yum update -y \
-    && yum remove -y git \
-    && yum install -y devtoolset-10-gcc devtoolset-10-gcc-c++ git236
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-ENV LIBRARY_PATH /opt/amdgpu/lib64
-
-ADD https://github.com/Kitware/CMake/releases/download/v$CMAKE_VERSION/cmake-$CMAKE_VERSION-linux-x86_64.tar.gz /tmp/cmake-$CMAKE_VERSION.tar.gz
-RUN tar -zx -C /usr --strip-components 1 </tmp/cmake-$CMAKE_VERSION.tar.gz
-
-WORKDIR /go/src/github.com/jmorganca/ollama
-COPY . .
-
-WORKDIR llm/generate
-RUN sh gen_linux.sh
-
-FROM --platform=linux/amd64 centos:7 AS build-amd64
-ENV CGO_ENABLED 1
-
-ARG GOLANG_VERSION
-ARG GOFLAGS
-ARG CGO_FLAGS
-
-RUN yum install -y centos-release-scl \
-    && yum update -y \
-    && yum install -y devtoolset-10-gcc devtoolset-10-gcc-c++
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-
-ADD https://dl.google.com/go/go$GOLANG_VERSION.linux-amd64.tar.gz /tmp/go-$GOLANG_VERSION.tar.gz
-RUN mkdir -p /usr/local && tar xz -C /usr/local </tmp/go-$GOLANG_VERSION.tar.gz
-ENV PATH /usr/local/go/bin:$PATH
-
-WORKDIR /go/src/github.com/jmorganca/ollama
-COPY . .
-COPY --from=cuda-build-amd64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/cpu/lib llm/llama.cpp/build/linux/cpu/lib
-COPY --from=cuda-build-amd64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/cuda/lib llm/llama.cpp/build/linux/cuda/lib
-COPY --from=rocm-build-amd64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/rocm/lib llm/llama.cpp/build/linux/rocm/lib
-RUN go build .
-
-FROM --platform=linux/arm64 centos:7 AS build-arm64
-ENV CGO_ENABLED 1
-
-ARG GOLANG_VERSION
-ARG GOFLAGS
-ARG CGO_FLAGS
-
-RUN yum install -y centos-release-scl \
-    && yum update -y \
-    && yum install -y devtoolset-10-gcc devtoolset-10-gcc-c++
-ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
-
-ADD https://dl.google.com/go/go$GOLANG_VERSION.linux-arm64.tar.gz /tmp/go-$GOLANG_VERSION.tar.gz
-RUN mkdir -p /usr/local && tar xz -C /usr/local </tmp/go-$GOLANG_VERSION.tar.gz
-ENV PATH /usr/local/go/bin:$PATH
-
-WORKDIR /go/src/github.com/jmorganca/ollama
-COPY . .
-COPY --from=cuda-build-arm64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/cpu/lib llm/llama.cpp/build/linux/cpu/lib
-COPY --from=cuda-build-arm64 /go/src/github.com/jmorganca/ollama/llm/llama.cpp/build/linux/cuda/lib llm/llama.cpp/build/linux/cuda/lib
-RUN go build .
-
-FROM build-$TARGETARCH
+RUN /usr/local/go/bin/go generate ./... && \
+    /usr/local/go/bin/go build .
--- a/README.md
+++ b/README.md
@@ -17,7 +17,7 @@ Get up and running with large language models locally.

 ### Windows

-Coming soon! For now, you can install Ollama on Windows via WSL2.
+Coming soon!

 ### Linux & WSL2

@@ -47,23 +47,20 @@ Here are some example open-source models that can be downloaded:

 | Model              | Parameters | Size  | Download                       |
 | ------------------ | ---------- | ----- | ------------------------------ |
-| Llama 2            | 7B         | 3.8GB | `ollama run llama2`            |
-| Mistral            | 7B         | 4.1GB | `ollama run mistral`           |
-| Dolphin Phi        | 2.7B       | 1.6GB | `ollama run dolphin-phi`       |
-| Phi-2              | 2.7B       | 1.7GB | `ollama run phi`               |
 | Neural Chat        | 7B         | 4.1GB | `ollama run neural-chat`       |
 | Starling           | 7B         | 4.1GB | `ollama run starling-lm`       |
+| Mistral            | 7B         | 4.1GB | `ollama run mistral`           |
+| Llama 2            | 7B         | 3.8GB | `ollama run llama2`            |
 | Code Llama         | 7B         | 3.8GB | `ollama run codellama`         |
 | Llama 2 Uncensored | 7B         | 3.8GB | `ollama run llama2-uncensored` |
 | Llama 2 13B        | 13B        | 7.3GB | `ollama run llama2:13b`        |
 | Llama 2 70B        | 70B        | 39GB  | `ollama run llama2:70b`        |
 | Orca Mini          | 3B         | 1.9GB | `ollama run orca-mini`         |
 | Vicuna             | 7B         | 3.8GB | `ollama run vicuna`            |
-| LLaVA              | 7B         | 4.5GB | `ollama run llava`             |

-> Note: You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models.
+> Note: You should have at least 8 GB of RAM to run the 3B models, 16 GB to run the 7B models, and 32 GB to run the 13B models.

-## Customize a model
+## Customize your own model

 ### Import from GGUF

@@ -107,7 +104,7 @@ FROM llama2
 # set the temperature to 1 [higher is more creative, lower is more coherent]
 PARAMETER temperature 1

-# set the system message
+# set the system prompt
 SYSTEM """
 You are Mario from Super Mario Bros. Answer as Mario, the assistant, only.
 """
@@ -130,10 +127,6 @@ For more examples, see the [examples](examples) directory. For more information

 `ollama create` is used to create a model from a Modelfile.

-```
-ollama create mymodel -f ./Modelfile
-```
-
 ### Pull a model

 ```
@@ -165,13 +158,6 @@ For multiline input, you can wrap text with `"""`:
 I'm a basic program that prints the famous "Hello, world!" message to the console.
 ```

-### Multimodal models
-
-```
->>> What's in this image? /Users/jmorgan/Desktop/smile.png
-The image features a yellow smiley face, which is likely the central focus of the picture.
-```
-
 ### Pass in prompt as arguments

 ```
@@ -197,19 +183,13 @@ Install `cmake` and `go`:
 brew install cmake go
 ```

-Then generate dependencies:
+Then generate dependencies and build:
+
 ```
 go generate ./...
-```
-Then build the binary:
-```
 go build .
 ```

-More detailed instructions can be found in the [developer guide](https://github.com/jmorganca/ollama/blob/main/docs/development.md)
-
-
-### Running local builds
 Next, start the server:

 ```
@@ -225,8 +205,7 @@ Finally, in a separate shell, run a model:
 ## REST API

 Ollama has a REST API for running and managing models.
-
-### Generate a response
+For example, to generate text from a model:

 ```
 curl http://localhost:11434/api/generate -d '{
@@ -235,23 +214,16 @@ curl http://localhost:11434/api/generate -d '{
 }'
 ```

-### Chat with a model
-
-```
-curl http://localhost:11434/api/chat -d '{
-  "model": "mistral",
-  "messages": [
-    { "role": "user", "content": "why is the sky blue?" }
-  ]
-}'
-```
-
 See the [API documentation](./docs/api.md) for all endpoints.

 ## Community Integrations

+### Mobile
+
+- [Mobile Artificial Intelligence Distribution](https://github.com/MaidFoundation/Maid) (Maid)
+
 ### Web & Desktop
- [Bionic GPT](https://github.com/bionic-gpt/bionic-gpt)
+
 - [HTML UI](https://github.com/rtcfirefly/ollama-ui)
 - [Chatbot UI](https://github.com/ivanfioravanti/chatbot-ollama)
 - [Typescript UI](https://github.com/ollama-interface/Ollama-Gui?tab=readme-ov-file)
@@ -261,9 +233,6 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [big-AGI](https://github.com/enricoros/big-agi/blob/main/docs/config-ollama.md)
 - [Cheshire Cat assistant framework](https://github.com/cheshire-cat-ai/core)
 - [Amica](https://github.com/semperai/amica)
- [chatd](https://github.com/BruceMacD/chatd)
- [Ollama-SwiftUI](https://github.com/kghandour/Ollama-SwiftUI)
-

 ### Terminal

@@ -275,11 +244,6 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [ogpt.nvim](https://github.com/huynle/ogpt.nvim)
 - [gptel Emacs client](https://github.com/karthink/gptel)
 - [Oatmeal](https://github.com/dustinblackman/oatmeal)
- [cmdh](https://github.com/pgibler/cmdh)
-
-### Database
-
- [MindsDB](https://github.com/mindsdb/mindsdb/blob/staging/mindsdb/integrations/handlers/ollama_handler/README.md)

 ### Package managers

@@ -298,12 +262,10 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [OllamaKit for Swift](https://github.com/kevinhermawan/OllamaKit)
 - [Ollama for Dart](https://github.com/breitburg/dart-ollama)
 - [Ollama for Laravel](https://github.com/cloudstudio/ollama-laravel)
- [LangChainDart](https://github.com/davidmigloz/langchain_dart)

 ### Mobile

- [Enchanted](https://github.com/AugustDev/enchanted)
- [Maid](https://github.com/Mobile-Artificial-Intelligence/maid)
+- [Maid](https://github.com/danemadsen/Maid) (Mobile Artificial Intelligence Distribution)

 ### Extensions & Plugins

@@ -314,7 +276,6 @@ See the [API documentation](./docs/api.md) for all endpoints.
 - [Logseq Ollama plugin](https://github.com/omagdy7/ollama-logseq)
 - [Dagger Chatbot](https://github.com/samalba/dagger-chatbot)
 - [Discord AI Bot](https://github.com/mekb-turtle/discord-ai-bot)
- [Ollama Telegram Bot](https://github.com/ruecat/ollama-telegram)
 - [Hass Ollama Conversation](https://github.com/ej52/hass-ollama-conversation)
 - [Rivet plugin](https://github.com/abrenneke/rivet-plugin-ollama)
 - [Llama Coder](https://github.com/ex3ndr/llama-coder) (Copilot alternative using Ollama)
--- a/api/client.go
+++ b/api/client.go
@@ -221,19 +221,6 @@ func (c *Client) Generate(ctx context.Context, req *GenerateRequest, fn Generate
 	})
 }

-type ChatResponseFunc func(ChatResponse) error
-
-func (c *Client) Chat(ctx context.Context, req *ChatRequest, fn ChatResponseFunc) error {
-	return c.stream(ctx, http.MethodPost, "/api/chat", req, func(bts []byte) error {
-		var resp ChatResponse
-		if err := json.Unmarshal(bts, &resp); err != nil {
-			return err
-		}
-
-		return fn(resp)
-	})
-}
-
 type PullProgressFunc func(ProgressResponse) error

 func (c *Client) Pull(ctx context.Context, req *PullRequest, fn PullProgressFunc) error {
@@ -309,13 +296,6 @@ func (c *Client) Heartbeat(ctx context.Context) error {
 	}
 	return nil
 }
-func (c *Client) Embeddings(ctx context.Context, req *EmbeddingRequest) (*EmbeddingResponse, error) {
-	var resp EmbeddingResponse
-	if err := c.do(ctx, http.MethodPost, "/api/embeddings", req, &resp); err != nil {
-		return nil, err
-	}
-	return &resp, nil
-}

 func (c *Client) CreateBlob(ctx context.Context, digest string, r io.Reader) error {
 	if err := c.do(ctx, http.MethodHead, fmt.Sprintf("/api/blobs/%s", digest), nil, nil); err != nil {
@@ -331,15 +311,3 @@ func (c *Client) CreateBlob(ctx context.Context, digest string, r io.Reader) err

 	return nil
 }
-
-func (c *Client) Version(ctx context.Context) (string, error) {
-	var version struct {
-		Version string `json:"version"`
-	}
-
-	if err := c.do(ctx, http.MethodGet, "/api/version", nil, &version); err != nil {
-		return "", err
-	}
-
-	return version.Version, nil
-}
--- a/api/types.go
+++ b/api/types.go
@@ -6,7 +6,6 @@ import (
 	"math"
 	"os"
 	"reflect"
-	"strconv"
 	"strings"
 	"time"
 )
@@ -31,56 +30,19 @@ func (e StatusError) Error() string {
 	}
 }

-type ImageData []byte
-
 type GenerateRequest struct {
-	Model    string      `json:"model"`
-	Prompt   string      `json:"prompt"`
-	System   string      `json:"system"`
-	Template string      `json:"template"`
-	Context  []int       `json:"context,omitempty"`
-	Stream   *bool       `json:"stream,omitempty"`
-	Raw      bool        `json:"raw,omitempty"`
-	Format   string      `json:"format"`
-	Images   []ImageData `json:"images,omitempty"`
+	Model    string `json:"model"`
+	Prompt   string `json:"prompt"`
+	System   string `json:"system"`
+	Template string `json:"template"`
+	Context  []int  `json:"context,omitempty"`
+	Stream   *bool  `json:"stream,omitempty"`
+	Raw      bool   `json:"raw,omitempty"`
+	Format   string `json:"format"`

 	Options map[string]interface{} `json:"options"`
 }

-type ChatRequest struct {
-	Model    string    `json:"model"`
-	Messages []Message `json:"messages"`
-	Stream   *bool     `json:"stream,omitempty"`
-	Format   string    `json:"format"`
-
-	Options map[string]interface{} `json:"options"`
-}
-
-type Message struct {
-	Role    string      `json:"role"` // one of ["system", "user", "assistant"]
-	Content string      `json:"content"`
-	Images  []ImageData `json:"images,omitempty"`
-}
-
-type ChatResponse struct {
-	Model     string    `json:"model"`
-	CreatedAt time.Time `json:"created_at"`
-	Message   Message   `json:"message"`
-
-	Done bool `json:"done"`
-
-	Metrics
-}
-
-type Metrics struct {
-	TotalDuration      time.Duration `json:"total_duration,omitempty"`
-	LoadDuration       time.Duration `json:"load_duration,omitempty"`
-	PromptEvalCount    int           `json:"prompt_eval_count,omitempty"`
-	PromptEvalDuration time.Duration `json:"prompt_eval_duration,omitempty"`
-	EvalCount          int           `json:"eval_count,omitempty"`
-	EvalDuration       time.Duration `json:"eval_duration,omitempty"`
-}
-
 // Options specfied in GenerateRequest, if you add a new option here add it to the API docs also
 type Options struct {
 	Runner
@@ -148,21 +110,15 @@ type DeleteRequest struct {
 }

 type ShowRequest struct {
-	Name     string `json:"name"`
-	Model    string `json:"model"`
-	System   string `json:"system"`
-	Template string `json:"template"`
-
-	Options map[string]interface{} `json:"options"`
+	Name string `json:"name"`
 }

 type ShowResponse struct {
-	License    string       `json:"license,omitempty"`
-	Modelfile  string       `json:"modelfile,omitempty"`
-	Parameters string       `json:"parameters,omitempty"`
-	Template   string       `json:"template,omitempty"`
-	System     string       `json:"system,omitempty"`
-	Details    ModelDetails `json:"details,omitempty"`
+	License    string `json:"license,omitempty"`
+	Modelfile  string `json:"modelfile,omitempty"`
+	Parameters string `json:"parameters,omitempty"`
+	Template   string `json:"template,omitempty"`
+	System     string `json:"system,omitempty"`
 }

 type CopyRequest struct {
@@ -198,11 +154,10 @@ type ListResponse struct {
 }

 type ModelResponse struct {
-	Name       string       `json:"name"`
-	ModifiedAt time.Time    `json:"modified_at"`
-	Size       int64        `json:"size"`
-	Digest     string       `json:"digest"`
-	Details    ModelDetails `json:"details,omitempty"`
+	Name       string    `json:"name"`
+	ModifiedAt time.Time `json:"modified_at"`
+	Size       int64     `json:"size"`
+	Digest     string    `json:"digest"`
 }

 type TokenResponse struct {
@@ -217,42 +172,39 @@ type GenerateResponse struct {
 	Done    bool  `json:"done"`
 	Context []int `json:"context,omitempty"`

-	Metrics
+	TotalDuration      time.Duration `json:"total_duration,omitempty"`
+	LoadDuration       time.Duration `json:"load_duration,omitempty"`
+	PromptEvalCount    int           `json:"prompt_eval_count,omitempty"`
+	PromptEvalDuration time.Duration `json:"prompt_eval_duration,omitempty"`
+	EvalCount          int           `json:"eval_count,omitempty"`
+	EvalDuration       time.Duration `json:"eval_duration,omitempty"`
 }

-type ModelDetails struct {
-	Format            string   `json:"format"`
-	Family            string   `json:"family"`
-	Families          []string `json:"families"`
-	ParameterSize     string   `json:"parameter_size"`
-	QuantizationLevel string   `json:"quantization_level"`
-}
-
-func (m *Metrics) Summary() {
-	if m.TotalDuration > 0 {
-		fmt.Fprintf(os.Stderr, "total duration:       %v\n", m.TotalDuration)
+func (r *GenerateResponse) Summary() {
+	if r.TotalDuration > 0 {
+		fmt.Fprintf(os.Stderr, "total duration:       %v\n", r.TotalDuration)
 	}

-	if m.LoadDuration > 0 {
-		fmt.Fprintf(os.Stderr, "load duration:        %v\n", m.LoadDuration)
+	if r.LoadDuration > 0 {
+		fmt.Fprintf(os.Stderr, "load duration:        %v\n", r.LoadDuration)
 	}

-	if m.PromptEvalCount > 0 {
-		fmt.Fprintf(os.Stderr, "prompt eval count:    %d token(s)\n", m.PromptEvalCount)
+	if r.PromptEvalCount > 0 {
+		fmt.Fprintf(os.Stderr, "prompt eval count:    %d token(s)\n", r.PromptEvalCount)
 	}

-	if m.PromptEvalDuration > 0 {
-		fmt.Fprintf(os.Stderr, "prompt eval duration: %s\n", m.PromptEvalDuration)
-		fmt.Fprintf(os.Stderr, "prompt eval rate:     %.2f tokens/s\n", float64(m.PromptEvalCount)/m.PromptEvalDuration.Seconds())
+	if r.PromptEvalDuration > 0 {
+		fmt.Fprintf(os.Stderr, "prompt eval duration: %s\n", r.PromptEvalDuration)
+		fmt.Fprintf(os.Stderr, "prompt eval rate:     %.2f tokens/s\n", float64(r.PromptEvalCount)/r.PromptEvalDuration.Seconds())
 	}

-	if m.EvalCount > 0 {
-		fmt.Fprintf(os.Stderr, "eval count:           %d token(s)\n", m.EvalCount)
+	if r.EvalCount > 0 {
+		fmt.Fprintf(os.Stderr, "eval count:           %d token(s)\n", r.EvalCount)
 	}

-	if m.EvalDuration > 0 {
-		fmt.Fprintf(os.Stderr, "eval duration:        %s\n", m.EvalDuration)
-		fmt.Fprintf(os.Stderr, "eval rate:            %.2f tokens/s\n", float64(m.EvalCount)/m.EvalDuration.Seconds())
+	if r.EvalDuration > 0 {
+		fmt.Fprintf(os.Stderr, "eval duration:        %s\n", r.EvalDuration)
+		fmt.Fprintf(os.Stderr, "eval rate:            %.2f tokens/s\n", float64(r.EvalCount)/r.EvalDuration.Seconds())
 	}
 }

@@ -408,63 +360,3 @@ func (d *Duration) UnmarshalJSON(b []byte) (err error) {

 	return nil
 }
-
-// FormatParams converts specified parameter options to their correct types
-func FormatParams(params map[string][]string) (map[string]interface{}, error) {
-	opts := Options{}
-	valueOpts := reflect.ValueOf(&opts).Elem() // names of the fields in the options struct
-	typeOpts := reflect.TypeOf(opts)           // types of the fields in the options struct
-
-	// build map of json struct tags to their types
-	jsonOpts := make(map[string]reflect.StructField)
-	for _, field := range reflect.VisibleFields(typeOpts) {
-		jsonTag := strings.Split(field.Tag.Get("json"), ",")[0]
-		if jsonTag != "" {
-			jsonOpts[jsonTag] = field
-		}
-	}
-
-	out := make(map[string]interface{})
-	// iterate params and set values based on json struct tags
-	for key, vals := range params {
-		if opt, ok := jsonOpts[key]; !ok {
-			return nil, fmt.Errorf("unknown parameter '%s'", key)
-		} else {
-			field := valueOpts.FieldByName(opt.Name)
-			if field.IsValid() && field.CanSet() {
-				switch field.Kind() {
-				case reflect.Float32:
-					floatVal, err := strconv.ParseFloat(vals[0], 32)
-					if err != nil {
-						return nil, fmt.Errorf("invalid float value %s", vals)
-					}
-
-					out[key] = float32(floatVal)
-				case reflect.Int:
-					intVal, err := strconv.ParseInt(vals[0], 10, 64)
-					if err != nil {
-						return nil, fmt.Errorf("invalid int value %s", vals)
-					}
-
-					out[key] = intVal
-				case reflect.Bool:
-					boolVal, err := strconv.ParseBool(vals[0])
-					if err != nil {
-						return nil, fmt.Errorf("invalid bool value %s", vals)
-					}
-
-					out[key] = boolVal
-				case reflect.String:
-					out[key] = vals[0]
-				case reflect.Slice:
-					// TODO: only string slices are supported right now
-					out[key] = vals
-				default:
-					return nil, fmt.Errorf("unknown type %s for %s", field.Kind(), key)
-				}
-			}
-		}
-	}
-
-	return out, nil
-}
--- a/cmd/cmd.go
+++ b/cmd/cmd.go
@@ -31,12 +31,11 @@ import (
 	"github.com/jmorganca/ollama/format"
 	"github.com/jmorganca/ollama/parser"
 	"github.com/jmorganca/ollama/progress"
+	"github.com/jmorganca/ollama/readline"
 	"github.com/jmorganca/ollama/server"
 	"github.com/jmorganca/ollama/version"
 )

-type ImageData []byte
-
 func CreateHandler(cmd *cobra.Command, args []string) error {
 	filename, _ := cmd.Flags().GetString("file")
 	filename, err := filepath.Abs(filename)
@@ -134,7 +133,7 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 	}

 	request := api.CreateRequest{Name: args[0], Modelfile: string(modelfile)}
-	if err := client.Create(cmd.Context(), &request, fn); err != nil {
+	if err := client.Create(context.Background(), &request, fn); err != nil {
 		return err
 	}

@@ -149,29 +148,15 @@ func RunHandler(cmd *cobra.Command, args []string) error {

 	name := args[0]
 	// check if the model exists on the server
-	model, err := client.Show(cmd.Context(), &api.ShowRequest{Name: name})
+	_, err = client.Show(context.Background(), &api.ShowRequest{Name: name})
 	var statusError api.StatusError
 	switch {
 	case errors.As(err, &statusError) && statusError.StatusCode == http.StatusNotFound:
-		if err := PullHandler(cmd, []string{name}); err != nil {
+		if err := PullHandler(cmd, args); err != nil {
 			return err
 		}
 	case err != nil:
 		return err
-	default:
-		// the model was found, check if it is in the correct format
-		if model.Details.Format != "" && model.Details.Format != "gguf" {
-			// pull and retry to see if the model has been updated
-			parts := strings.Split(name, string(os.PathSeparator))
-			if len(parts) == 1 {
-				// this is a library model, log some info
-				fmt.Fprintln(os.Stderr, "This model is no longer compatible with Ollama. Pulling a new version...")
-			}
-			if err := PullHandler(cmd, []string{name}); err != nil {
-				fmt.Printf("Error: %s\n", err)
-				return fmt.Errorf("unsupported model, please update this model to gguf format") // relay the original error
-			}
-		}
 	}

 	return RunGenerate(cmd, args)
@@ -223,7 +208,7 @@ func PushHandler(cmd *cobra.Command, args []string) error {
 	}

 	request := api.PushRequest{Name: args[0], Insecure: insecure}
-	if err := client.Push(cmd.Context(), &request, fn); err != nil {
+	if err := client.Push(context.Background(), &request, fn); err != nil {
 		return err
 	}

@@ -237,7 +222,7 @@ func ListHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}

-	models, err := client.List(cmd.Context())
+	models, err := client.List(context.Background())
 	if err != nil {
 		return err
 	}
@@ -272,7 +257,7 @@ func DeleteHandler(cmd *cobra.Command, args []string) error {

 	for _, name := range args {
 		req := api.DeleteRequest{Name: name}
-		if err := client.Delete(cmd.Context(), &req); err != nil {
+		if err := client.Delete(context.Background(), &req); err != nil {
 			return err
 		}
 		fmt.Printf("deleted '%s'\n", name)
@@ -337,7 +322,7 @@ func ShowHandler(cmd *cobra.Command, args []string) error {
 	}

 	req := api.ShowRequest{Name: args[0]}
-	resp, err := client.Show(cmd.Context(), &req)
+	resp, err := client.Show(context.Background(), &req)
 	if err != nil {
 		return err
 	}
@@ -365,7 +350,7 @@ func CopyHandler(cmd *cobra.Command, args []string) error {
 	}

 	req := api.CopyRequest{Source: args[0], Destination: args[1]}
-	if err := client.Copy(cmd.Context(), &req); err != nil {
+	if err := client.Copy(context.Background(), &req); err != nil {
 		return err
 	}
 	fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
@@ -419,7 +404,7 @@ func PullHandler(cmd *cobra.Command, args []string) error {
 	}

 	request := api.PullRequest{Name: args[0], Insecure: insecure}
-	if err := client.Pull(cmd.Context(), &request, fn); err != nil {
+	if err := client.Pull(context.Background(), &request, fn); err != nil {
 		return err
 	}

@@ -427,22 +412,13 @@ func PullHandler(cmd *cobra.Command, args []string) error {
 }

 func RunGenerate(cmd *cobra.Command, args []string) error {
-	interactive := true
-
-	opts := generateOptions{
-		Model:    args[0],
-		WordWrap: os.Getenv("TERM") == "xterm-256color",
-		Options:  map[string]interface{}{},
-		Images:   []ImageData{},
-	}
-
 	format, err := cmd.Flags().GetString("format")
 	if err != nil {
 		return err
 	}
-	opts.Format = format

 	prompts := args[1:]
+
 	// prepend stdin to the prompt if provided
 	if !term.IsTerminal(int(os.Stdin.Fd())) {
 		in, err := io.ReadAll(os.Stdin)
@@ -451,41 +427,34 @@ func RunGenerate(cmd *cobra.Command, args []string) error {
 		}

 		prompts = append([]string{string(in)}, prompts...)
-		opts.WordWrap = false
-		interactive = false
 	}
-	opts.Prompt = strings.Join(prompts, " ")
-	if len(prompts) > 0 {
-		interactive = false
+
+	// output is being piped
+	if !term.IsTerminal(int(os.Stdout.Fd())) {
+		return generate(cmd, args[0], strings.Join(prompts, " "), false, format)
 	}

+	wordWrap := os.Getenv("TERM") == "xterm-256color"
+
 	nowrap, err := cmd.Flags().GetBool("nowordwrap")
 	if err != nil {
 		return err
 	}
-	opts.WordWrap = !nowrap
-
-	if !interactive {
-		return generate(cmd, opts)
+	if nowrap {
+		wordWrap = false
 	}

-	return generateInteractive(cmd, opts)
+	// prompts are provided via stdin or args so don't enter interactive mode
+	if len(prompts) > 0 {
+		return generate(cmd, args[0], strings.Join(prompts, " "), wordWrap, format)
+	}
+
+	return generateInteractive(cmd, args[0], wordWrap, format)
 }

 type generateContextKey string

-type generateOptions struct {
-	Model    string
-	Prompt   string
-	WordWrap bool
-	Format   string
-	System   string
-	Template string
-	Images   []ImageData
-	Options  map[string]interface{}
-}
-
-func generate(cmd *cobra.Command, opts generateOptions) error {
+func generate(cmd *cobra.Command, model, prompt string, wordWrap bool, format string) error {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
@@ -506,39 +475,34 @@ func generate(cmd *cobra.Command, opts generateOptions) error {

 	termWidth, _, err := term.GetSize(int(os.Stdout.Fd()))
 	if err != nil {
-		opts.WordWrap = false
+		wordWrap = false
 	}

-	ctx, cancel := context.WithCancel(cmd.Context())
+	cancelCtx, cancel := context.WithCancel(context.Background())
 	defer cancel()

 	sigChan := make(chan os.Signal, 1)
 	signal.Notify(sigChan, syscall.SIGINT)
+	var abort bool

 	go func() {
 		<-sigChan
 		cancel()
+		abort = true
 	}()

 	var currentLineLength int
 	var wordBuffer string

+	request := api.GenerateRequest{Model: model, Prompt: prompt, Context: generateContext, Format: format}
 	fn := func(response api.GenerateResponse) error {
 		p.StopAndClear()

 		latest = response

-		termWidth, _, _ = term.GetSize(int(os.Stdout.Fd()))
-		if opts.WordWrap && termWidth >= 10 {
+		if wordWrap {
 			for _, ch := range response.Response {
 				if currentLineLength+1 > termWidth-5 {
-					if len(wordBuffer) > termWidth-10 {
-						fmt.Printf("%s%c", wordBuffer, ch)
-						wordBuffer = ""
-						currentLineLength = 0
-						continue
-					}
-
 					// backtrack the length of the last word and clear to the end of the line
 					fmt.Printf("\x1b[%dD\x1b[K\n", len(wordBuffer))
 					fmt.Printf("%s%c", wordBuffer, ch)
@@ -558,44 +522,28 @@ func generate(cmd *cobra.Command, opts generateOptions) error {
 				}
 			}
 		} else {
-			fmt.Printf("%s%s", wordBuffer, response.Response)
-			if len(wordBuffer) > 0 {
-				wordBuffer = ""
-			}
+			fmt.Print(response.Response)
 		}

 		return nil
 	}

-	images := make([]api.ImageData, 0)
-	for _, i := range opts.Images {
-		images = append(images, api.ImageData(i))
-	}
-	request := api.GenerateRequest{
-		Model:    opts.Model,
-		Prompt:   opts.Prompt,
-		Context:  generateContext,
-		Format:   opts.Format,
-		System:   opts.System,
-		Template: opts.Template,
-		Options:  opts.Options,
-		Images:   images,
-	}
-
-	if err := client.Generate(ctx, &request, fn); err != nil {
-		if errors.Is(err, context.Canceled) {
+	if err := client.Generate(cancelCtx, &request, fn); err != nil {
+		if strings.Contains(err.Error(), "context canceled") && abort {
 			return nil
 		}
 		return err
 	}
-
-	if opts.Prompt != "" {
+	if prompt != "" {
 		fmt.Println()
 		fmt.Println()
 	}

 	if !latest.Done {
-		return nil
+		if abort {
+			return nil
+		}
+		return errors.New("unexpected end of response")
 	}

 	verbose, err := cmd.Flags().GetBool("verbose")
@@ -607,12 +555,227 @@ func generate(cmd *cobra.Command, opts generateOptions) error {
 		latest.Summary()
 	}

-	ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
+	ctx := cmd.Context()
+	ctx = context.WithValue(ctx, generateContextKey("context"), latest.Context)
 	cmd.SetContext(ctx)

 	return nil
 }

+func generateInteractive(cmd *cobra.Command, model string, wordWrap bool, format string) error {
+	// load the model
+	if err := generate(cmd, model, "", false, ""); err != nil {
+		return err
+	}
+
+	usage := func() {
+		fmt.Fprintln(os.Stderr, "Available Commands:")
+		fmt.Fprintln(os.Stderr, "  /set         Set session variables")
+		fmt.Fprintln(os.Stderr, "  /show        Show model information")
+		fmt.Fprintln(os.Stderr, "  /bye         Exit")
+		fmt.Fprintln(os.Stderr, "  /?, /help    Help for a command")
+		fmt.Fprintln(os.Stderr, "")
+		fmt.Fprintln(os.Stderr, "Use \"\"\" to begin a multi-line message.")
+		fmt.Fprintln(os.Stderr, "")
+	}
+
+	usageSet := func() {
+		fmt.Fprintln(os.Stderr, "Available Commands:")
+		fmt.Fprintln(os.Stderr, "  /set history      Enable history")
+		fmt.Fprintln(os.Stderr, "  /set nohistory    Disable history")
+		fmt.Fprintln(os.Stderr, "  /set wordwrap     Enable wordwrap")
+		fmt.Fprintln(os.Stderr, "  /set nowordwrap   Disable wordwrap")
+		fmt.Fprintln(os.Stderr, "  /set format json  Enable JSON mode")
+		fmt.Fprintln(os.Stderr, "  /set noformat     Disable formatting")
+		fmt.Fprintln(os.Stderr, "  /set verbose      Show LLM stats")
+		fmt.Fprintln(os.Stderr, "  /set quiet        Disable LLM stats")
+		fmt.Fprintln(os.Stderr, "")
+	}
+
+	usageShow := func() {
+		fmt.Fprintln(os.Stderr, "Available Commands:")
+		fmt.Fprintln(os.Stderr, "  /show license      Show model license")
+		fmt.Fprintln(os.Stderr, "  /show modelfile    Show Modelfile for this model")
+		fmt.Fprintln(os.Stderr, "  /show parameters   Show parameters for this model")
+		fmt.Fprintln(os.Stderr, "  /show system       Show system prompt")
+		fmt.Fprintln(os.Stderr, "  /show template     Show prompt template")
+		fmt.Fprintln(os.Stderr, "")
+	}
+
+	scanner, err := readline.New(readline.Prompt{
+		Prompt:         ">>> ",
+		AltPrompt:      "... ",
+		Placeholder:    "Send a message (/? for help)",
+		AltPlaceholder: `Use """ to end multi-line input`,
+	})
+	if err != nil {
+		return err
+	}
+
+	fmt.Print(readline.StartBracketedPaste)
+	defer fmt.Printf(readline.EndBracketedPaste)
+
+	var prompt string
+
+	for {
+		line, err := scanner.Readline()
+		switch {
+		case errors.Is(err, io.EOF):
+			fmt.Println()
+			return nil
+		case errors.Is(err, readline.ErrInterrupt):
+			if line == "" {
+				fmt.Println("\nUse Ctrl-D or /bye to exit.")
+			}
+
+			scanner.Prompt.UseAlt = false
+			prompt = ""
+
+			continue
+		case err != nil:
+			return err
+		}
+
+		switch {
+		case strings.HasPrefix(prompt, `"""`):
+			// if the prompt so far starts with """ then we're in multiline mode
+			// and we need to keep reading until we find a line that ends with """
+			cut, found := strings.CutSuffix(line, `"""`)
+			prompt += cut + "\n"
+
+			if !found {
+				continue
+			}
+
+			prompt = strings.TrimPrefix(prompt, `"""`)
+			scanner.Prompt.UseAlt = false
+		case strings.HasPrefix(line, `"""`) && len(prompt) == 0:
+			scanner.Prompt.UseAlt = true
+			prompt += line + "\n"
+			continue
+		case scanner.Pasting:
+			prompt += line + "\n"
+			continue
+		case strings.HasPrefix(line, "/list"):
+			args := strings.Fields(line)
+			if err := ListHandler(cmd, args[1:]); err != nil {
+				return err
+			}
+		case strings.HasPrefix(line, "/set"):
+			args := strings.Fields(line)
+			if len(args) > 1 {
+				switch args[1] {
+				case "history":
+					scanner.HistoryEnable()
+				case "nohistory":
+					scanner.HistoryDisable()
+				case "wordwrap":
+					wordWrap = true
+					fmt.Println("Set 'wordwrap' mode.")
+				case "nowordwrap":
+					wordWrap = false
+					fmt.Println("Set 'nowordwrap' mode.")
+				case "verbose":
+					cmd.Flags().Set("verbose", "true")
+					fmt.Println("Set 'verbose' mode.")
+				case "quiet":
+					cmd.Flags().Set("verbose", "false")
+					fmt.Println("Set 'quiet' mode.")
+				case "format":
+					if len(args) < 3 || args[2] != "json" {
+						fmt.Println("Invalid or missing format. For 'json' mode use '/set format json'")
+					} else {
+						format = args[2]
+						fmt.Printf("Set format to '%s' mode.\n", args[2])
+					}
+				case "noformat":
+					format = ""
+					fmt.Println("Disabled format.")
+				default:
+					fmt.Printf("Unknown command '/set %s'. Type /? for help\n", args[1])
+				}
+			} else {
+				usageSet()
+			}
+		case strings.HasPrefix(line, "/show"):
+			args := strings.Fields(line)
+			if len(args) > 1 {
+				client, err := api.ClientFromEnvironment()
+				if err != nil {
+					fmt.Println("error: couldn't connect to ollama server")
+					return err
+				}
+				resp, err := client.Show(cmd.Context(), &api.ShowRequest{Name: model})
+				if err != nil {
+					fmt.Println("error: couldn't get model")
+					return err
+				}
+
+				switch args[1] {
+				case "license":
+					if resp.License == "" {
+						fmt.Print("No license was specified for this model.\n\n")
+					} else {
+						fmt.Println(resp.License)
+					}
+				case "modelfile":
+					fmt.Println(resp.Modelfile)
+				case "parameters":
+					if resp.Parameters == "" {
+						fmt.Print("No parameters were specified for this model.\n\n")
+					} else {
+						fmt.Println(resp.Parameters)
+					}
+				case "system":
+					if resp.System == "" {
+						fmt.Print("No system prompt was specified for this model.\n\n")
+					} else {
+						fmt.Println(resp.System)
+					}
+				case "template":
+					if resp.Template == "" {
+						fmt.Print("No prompt template was specified for this model.\n\n")
+					} else {
+						fmt.Println(resp.Template)
+					}
+				default:
+					fmt.Printf("Unknown command '/show %s'. Type /? for help\n", args[1])
+				}
+			} else {
+				usageShow()
+			}
+		case strings.HasPrefix(line, "/help"), strings.HasPrefix(line, "/?"):
+			args := strings.Fields(line)
+			if len(args) > 1 {
+				switch args[1] {
+				case "set", "/set":
+					usageSet()
+				case "show", "/show":
+					usageShow()
+				}
+			} else {
+				usage()
+			}
+		case line == "/exit", line == "/bye":
+			return nil
+		case strings.HasPrefix(line, "/"):
+			args := strings.Fields(line)
+			fmt.Printf("Unknown command '%s'. Type /? for help\n", args[0])
+			continue
+		default:
+			prompt += line
+		}
+
+		if len(prompt) > 0 && prompt[0] != '/' {
+			if err := generate(cmd, model, prompt, wordWrap, format); err != nil {
+				return err
+			}
+
+			prompt = ""
+		}
+	}
+}
+
 func RunServer(cmd *cobra.Command, _ []string) error {
 	host, port, err := net.SplitHostPort(os.Getenv("OLLAMA_HOST"))
 	if err != nil {
@@ -631,7 +794,12 @@ func RunServer(cmd *cobra.Command, _ []string) error {
 		return err
 	}

-	return server.Serve(ln)
+	var origins []string
+	if o := os.Getenv("OLLAMA_ORIGINS"); o != "" {
+		origins = strings.Split(o, ",")
+	}
+
+	return server.Serve(ln, origins)
 }

 func initializeKeypair() error {
@@ -683,7 +851,7 @@ func initializeKeypair() error {
 	return nil
 }

-func startMacApp(ctx context.Context, client *api.Client) error {
+func startMacApp(client *api.Client) error {
 	exe, err := os.Executable()
 	if err != nil {
 		return err
@@ -707,24 +875,24 @@ func startMacApp(ctx context.Context, client *api.Client) error {
 		case <-timeout:
 			return errors.New("timed out waiting for server to start")
 		case <-tick:
-			if err := client.Heartbeat(ctx); err == nil {
+			if err := client.Heartbeat(context.Background()); err == nil {
 				return nil // server has started
 			}
 		}
 	}
 }

-func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
+func checkServerHeartbeat(_ *cobra.Command, _ []string) error {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
-	if err := client.Heartbeat(cmd.Context()); err != nil {
+	if err := client.Heartbeat(context.Background()); err != nil {
 		if !strings.Contains(err.Error(), "connection refused") {
 			return err
 		}
 		if runtime.GOOS == "darwin" {
-			if err := startMacApp(cmd.Context(), client); err != nil {
+			if err := startMacApp(client); err != nil {
 				return fmt.Errorf("could not connect to ollama app, is it running?")
 			}
 		} else {
@@ -734,29 +902,8 @@ func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
 	return nil
 }

-func versionHandler(cmd *cobra.Command, _ []string) {
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		return
-	}
-
-	serverVersion, err := client.Version(cmd.Context())
-	if err != nil {
-		fmt.Println("Warning: could not connect to a running Ollama instance")
-	}
-
-	if serverVersion != "" {
-		fmt.Printf("ollama version is %s\n", serverVersion)
-	}
-
-	if serverVersion != version.Version {
-		fmt.Printf("Warning: client version is %s\n", version.Version)
-	}
-}
-
 func NewCLI() *cobra.Command {
 	log.SetFlags(log.LstdFlags | log.Lshortfile)
-	cobra.EnableCommandSorting = false

 	rootCmd := &cobra.Command{
 		Use:           "ollama",
@@ -766,17 +913,10 @@ func NewCLI() *cobra.Command {
 		CompletionOptions: cobra.CompletionOptions{
 			DisableDefaultCmd: true,
 		},
-		Run: func(cmd *cobra.Command, args []string) {
-			if version, _ := cmd.Flags().GetBool("version"); version {
-				versionHandler(cmd, args)
-				return
-			}
-
-			cmd.Print(cmd.UsageString())
-		},
+		Version: version.Version,
 	}

-	rootCmd.Flags().BoolP("version", "v", false, "Show version information")
+	cobra.EnableCommandSorting = false

 	createCmd := &cobra.Command{
 		Use:     "create MODEL",
@@ -800,7 +940,7 @@ func NewCLI() *cobra.Command {
 	showCmd.Flags().Bool("modelfile", false, "Show Modelfile of a model")
 	showCmd.Flags().Bool("parameters", false, "Show parameters of a model")
 	showCmd.Flags().Bool("template", false, "Show template of a model")
-	showCmd.Flags().Bool("system", false, "Show system message of a model")
+	showCmd.Flags().Bool("system", false, "Show system prompt of a model")

 	runCmd := &cobra.Command{
 		Use:     "run MODEL [PROMPT]",
--- a/cmd/interactive.go
+++ b/cmd/interactive.go
@@ -1,532 +0,0 @@
-package cmd
-
-import (
-	"context"
-	"errors"
-	"fmt"
-	"io"
-	"net/http"
-	"os"
-	"regexp"
-	"strings"
-
-	"github.com/spf13/cobra"
-	"golang.org/x/exp/slices"
-
-	"github.com/jmorganca/ollama/api"
-	"github.com/jmorganca/ollama/readline"
-)
-
-type MultilineState int
-
-const (
-	MultilineNone MultilineState = iota
-	MultilinePrompt
-	MultilineSystem
-	MultilineTemplate
-)
-
-func modelIsMultiModal(cmd *cobra.Command, name string) bool {
-	// get model details
-	client, err := api.ClientFromEnvironment()
-	if err != nil {
-		fmt.Println("error: couldn't connect to ollama server")
-		return false
-	}
-
-	req := api.ShowRequest{Name: name}
-	resp, err := client.Show(cmd.Context(), &req)
-	if err != nil {
-		return false
-	}
-
-	return slices.Contains(resp.Details.Families, "clip")
-}
-
-func generateInteractive(cmd *cobra.Command, opts generateOptions) error {
-	multiModal := modelIsMultiModal(cmd, opts.Model)
-
-	// load the model
-	loadOpts := generateOptions{
-		Model:  opts.Model,
-		Prompt: "",
-		Images: []ImageData{},
-	}
-	if err := generate(cmd, loadOpts); err != nil {
-		return err
-	}
-
-	usage := func() {
-		fmt.Fprintln(os.Stderr, "Available Commands:")
-		fmt.Fprintln(os.Stderr, "  /set          Set session variables")
-		fmt.Fprintln(os.Stderr, "  /show         Show model information")
-		fmt.Fprintln(os.Stderr, "  /bye          Exit")
-		fmt.Fprintln(os.Stderr, "  /?, /help     Help for a command")
-		fmt.Fprintln(os.Stderr, "  /? shortcuts  Help for keyboard shortcuts")
-		fmt.Fprintln(os.Stderr, "")
-		fmt.Fprintln(os.Stderr, "Use \"\"\" to begin a multi-line message.")
-		fmt.Fprintln(os.Stderr, "")
-	}
-
-	usageSet := func() {
-		fmt.Fprintln(os.Stderr, "Available Commands:")
-		fmt.Fprintln(os.Stderr, "  /set parameter ...     Set a parameter")
-		fmt.Fprintln(os.Stderr, "  /set system <string>   Set system message")
-		fmt.Fprintln(os.Stderr, "  /set template <string> Set prompt template")
-		fmt.Fprintln(os.Stderr, "  /set history           Enable history")
-		fmt.Fprintln(os.Stderr, "  /set nohistory         Disable history")
-		fmt.Fprintln(os.Stderr, "  /set wordwrap          Enable wordwrap")
-		fmt.Fprintln(os.Stderr, "  /set nowordwrap        Disable wordwrap")
-		fmt.Fprintln(os.Stderr, "  /set format json       Enable JSON mode")
-		fmt.Fprintln(os.Stderr, "  /set noformat          Disable formatting")
-		fmt.Fprintln(os.Stderr, "  /set verbose           Show LLM stats")
-		fmt.Fprintln(os.Stderr, "  /set quiet             Disable LLM stats")
-		fmt.Fprintln(os.Stderr, "")
-	}
-
-	usageShortcuts := func() {
-		fmt.Fprintln(os.Stderr, "Available keyboard shortcuts:")
-		fmt.Fprintln(os.Stderr, "  Ctrl + a            Move to the beginning of the line (Home)")
-		fmt.Fprintln(os.Stderr, "  Ctrl + e            Move to the end of the line (End)")
-		fmt.Fprintln(os.Stderr, "   Alt + b            Move back (left) one word")
-		fmt.Fprintln(os.Stderr, "   Alt + f            Move forward (right) one word")
-		fmt.Fprintln(os.Stderr, "  Ctrl + k            Delete the sentence after the cursor")
-		fmt.Fprintln(os.Stderr, "  Ctrl + u            Delete the sentence before the cursor")
-		fmt.Fprintln(os.Stderr, "")
-		fmt.Fprintln(os.Stderr, "  Ctrl + l            Clear the screen")
-		fmt.Fprintln(os.Stderr, "  Ctrl + c            Stop the model from responding")
-		fmt.Fprintln(os.Stderr, "  Ctrl + d            Exit ollama (/bye)")
-		fmt.Fprintln(os.Stderr, "")
-	}
-
-	usageShow := func() {
-		fmt.Fprintln(os.Stderr, "Available Commands:")
-		fmt.Fprintln(os.Stderr, "  /show info         Show details for this model")
-		fmt.Fprintln(os.Stderr, "  /show license      Show model license")
-		fmt.Fprintln(os.Stderr, "  /show modelfile    Show Modelfile for this model")
-		fmt.Fprintln(os.Stderr, "  /show parameters   Show parameters for this model")
-		fmt.Fprintln(os.Stderr, "  /show system       Show system message")
-		fmt.Fprintln(os.Stderr, "  /show template     Show prompt template")
-		fmt.Fprintln(os.Stderr, "")
-	}
-
-	// only list out the most common parameters
-	usageParameters := func() {
-		fmt.Fprintln(os.Stderr, "Available Parameters:")
-		fmt.Fprintln(os.Stderr, "  /set parameter seed <int>             Random number seed")
-		fmt.Fprintln(os.Stderr, "  /set parameter num_predict <int>      Max number of tokens to predict")
-		fmt.Fprintln(os.Stderr, "  /set parameter top_k <int>            Pick from top k num of tokens")
-		fmt.Fprintln(os.Stderr, "  /set parameter top_p <float>          Pick token based on sum of probabilities")
-		fmt.Fprintln(os.Stderr, "  /set parameter num_ctx <int>          Set the context size")
-		fmt.Fprintln(os.Stderr, "  /set parameter temperature <float>    Set creativity level")
-		fmt.Fprintln(os.Stderr, "  /set parameter repeat_penalty <float> How strongly to penalize repetitions")
-		fmt.Fprintln(os.Stderr, "  /set parameter repeat_last_n <int>    Set how far back to look for repetitions")
-		fmt.Fprintln(os.Stderr, "  /set parameter num_gpu <int>          The number of layers to send to the GPU")
-		fmt.Fprintln(os.Stderr, "  /set parameter stop \"<string>\", ...   Set the stop parameters")
-		fmt.Fprintln(os.Stderr, "")
-	}
-
-	scanner, err := readline.New(readline.Prompt{
-		Prompt:         ">>> ",
-		AltPrompt:      "... ",
-		Placeholder:    "Send a message (/? for help)",
-		AltPlaceholder: `Use """ to end multi-line input`,
-	})
-	if err != nil {
-		return err
-	}
-
-	fmt.Print(readline.StartBracketedPaste)
-	defer fmt.Printf(readline.EndBracketedPaste)
-
-	var multiline MultilineState
-	var prompt string
-
-	for {
-		line, err := scanner.Readline()
-		switch {
-		case errors.Is(err, io.EOF):
-			fmt.Println()
-			return nil
-		case errors.Is(err, readline.ErrInterrupt):
-			if line == "" {
-				fmt.Println("\nUse Ctrl + d or /bye to exit.")
-			}
-
-			scanner.Prompt.UseAlt = false
-			prompt = ""
-
-			continue
-		case err != nil:
-			return err
-		}
-
-		switch {
-		case strings.HasPrefix(prompt, `"""`):
-			// if the prompt so far starts with """ then we're in multiline mode
-			// and we need to keep reading until we find a line that ends with """
-			cut, found := strings.CutSuffix(line, `"""`)
-			prompt += cut
-
-			if !found {
-				prompt += "\n"
-				continue
-			}
-
-			prompt = strings.TrimPrefix(prompt, `"""`)
-			scanner.Prompt.UseAlt = false
-
-			switch multiline {
-			case MultilineSystem:
-				opts.System = prompt
-				prompt = ""
-				fmt.Println("Set system message.")
-			case MultilineTemplate:
-				opts.Template = prompt
-				prompt = ""
-				fmt.Println("Set prompt template.")
-			}
-			multiline = MultilineNone
-		case strings.HasPrefix(line, `"""`) && len(prompt) == 0:
-			scanner.Prompt.UseAlt = true
-			multiline = MultilinePrompt
-			prompt += line + "\n"
-			continue
-		case scanner.Pasting:
-			prompt += line + "\n"
-			continue
-		case strings.HasPrefix(line, "/list"):
-			args := strings.Fields(line)
-			if err := ListHandler(cmd, args[1:]); err != nil {
-				return err
-			}
-		case strings.HasPrefix(line, "/set"):
-			args := strings.Fields(line)
-			if len(args) > 1 {
-				switch args[1] {
-				case "history":
-					scanner.HistoryEnable()
-				case "nohistory":
-					scanner.HistoryDisable()
-				case "wordwrap":
-					opts.WordWrap = true
-					fmt.Println("Set 'wordwrap' mode.")
-				case "nowordwrap":
-					opts.WordWrap = false
-					fmt.Println("Set 'nowordwrap' mode.")
-				case "verbose":
-					cmd.Flags().Set("verbose", "true")
-					fmt.Println("Set 'verbose' mode.")
-				case "quiet":
-					cmd.Flags().Set("verbose", "false")
-					fmt.Println("Set 'quiet' mode.")
-				case "format":
-					if len(args) < 3 || args[2] != "json" {
-						fmt.Println("Invalid or missing format. For 'json' mode use '/set format json'")
-					} else {
-						opts.Format = args[2]
-						fmt.Printf("Set format to '%s' mode.\n", args[2])
-					}
-				case "noformat":
-					opts.Format = ""
-					fmt.Println("Disabled format.")
-				case "parameter":
-					if len(args) < 4 {
-						usageParameters()
-						continue
-					}
-					var params []string
-					for _, p := range args[3:] {
-						params = append(params, p)
-					}
-					fp, err := api.FormatParams(map[string][]string{args[2]: params})
-					if err != nil {
-						fmt.Printf("Couldn't set parameter: %q\n\n", err)
-						continue
-					}
-					fmt.Printf("Set parameter '%s' to '%s'\n\n", args[2], strings.Join(params, ", "))
-					opts.Options[args[2]] = fp[args[2]]
-				case "system", "template":
-					if len(args) < 3 {
-						usageSet()
-						continue
-					}
-					line := strings.Join(args[2:], " ")
-					line = strings.TrimPrefix(line, `"""`)
-					if strings.HasPrefix(args[2], `"""`) {
-						cut, found := strings.CutSuffix(line, `"""`)
-						prompt += cut
-						if found {
-							if args[1] == "system" {
-								opts.System = prompt
-								fmt.Println("Set system message.")
-							} else {
-								opts.Template = prompt
-								fmt.Println("Set prompt template.")
-							}
-							prompt = ""
-						} else {
-							prompt = `"""` + prompt + "\n"
-							if args[1] == "system" {
-								multiline = MultilineSystem
-							} else {
-								multiline = MultilineTemplate
-							}
-							scanner.Prompt.UseAlt = true
-						}
-					} else {
-						opts.System = line
-						fmt.Println("Set system message.")
-					}
-				default:
-					fmt.Printf("Unknown command '/set %s'. Type /? for help\n", args[1])
-				}
-			} else {
-				usageSet()
-			}
-		case strings.HasPrefix(line, "/show"):
-			args := strings.Fields(line)
-			if len(args) > 1 {
-				client, err := api.ClientFromEnvironment()
-				if err != nil {
-					fmt.Println("error: couldn't connect to ollama server")
-					return err
-				}
-				req := &api.ShowRequest{
-					Name:     opts.Model,
-					System:   opts.System,
-					Template: opts.Template,
-					Options:  opts.Options,
-				}
-				resp, err := client.Show(cmd.Context(), req)
-				if err != nil {
-					fmt.Println("error: couldn't get model")
-					return err
-				}
-
-				switch args[1] {
-				case "info":
-					fmt.Println("Model details:")
-					if len(resp.Details.Families) > 0 {
-						fmt.Printf("Family              %s\n", strings.Join(resp.Details.Families, ", "))
-					} else if resp.Details.Family != "" {
-						fmt.Printf("Family              %s\n", resp.Details.Family)
-					}
-					fmt.Printf("Parameter Size      %s\n", resp.Details.ParameterSize)
-					fmt.Printf("Quantization Level  %s\n", resp.Details.QuantizationLevel)
-					fmt.Println("")
-				case "license":
-					if resp.License == "" {
-						fmt.Print("No license was specified for this model.\n\n")
-					} else {
-						fmt.Println(resp.License)
-					}
-				case "modelfile":
-					fmt.Println(resp.Modelfile)
-				case "parameters":
-					if resp.Parameters == "" {
-						fmt.Print("No parameters were specified for this model.\n\n")
-					} else {
-						if len(opts.Options) > 0 {
-							fmt.Println("User defined parameters:")
-							for k, v := range opts.Options {
-								fmt.Printf("%-*s %v\n", 30, k, v)
-							}
-							fmt.Println()
-						}
-						fmt.Println("Model defined parameters:")
-						fmt.Println(resp.Parameters)
-					}
-				case "system":
-					switch {
-					case opts.System != "":
-						fmt.Println(opts.System + "\n")
-					case resp.System != "":
-						fmt.Println(resp.System + "\n")
-					default:
-						fmt.Print("No system message was specified for this model.\n\n")
-					}
-				case "template":
-					switch {
-					case opts.Template != "":
-						fmt.Println(opts.Template + "\n")
-					case resp.Template != "":
-						fmt.Println(resp.Template)
-					default:
-						fmt.Print("No prompt template was specified for this model.\n\n")
-					}
-				default:
-					fmt.Printf("Unknown command '/show %s'. Type /? for help\n", args[1])
-				}
-			} else {
-				usageShow()
-			}
-		case strings.HasPrefix(line, "/help"), strings.HasPrefix(line, "/?"):
-			args := strings.Fields(line)
-			if len(args) > 1 {
-				switch args[1] {
-				case "set", "/set":
-					usageSet()
-				case "show", "/show":
-					usageShow()
-				case "shortcut", "shortcuts":
-					usageShortcuts()
-				}
-			} else {
-				usage()
-			}
-		case line == "/exit", line == "/bye":
-			return nil
-		case strings.HasPrefix(line, "/"):
-			args := strings.Fields(line)
-			isFile := false
-
-			if multiModal {
-				for _, f := range extractFileNames(line) {
-					if strings.HasPrefix(f, args[0]) {
-						isFile = true
-						break
-					}
-				}
-			}
-
-			if isFile {
-				prompt += line
-			} else {
-				fmt.Printf("Unknown command '%s'. Type /? for help\n", args[0])
-				continue
-			}
-		default:
-			prompt += line
-		}
-
-		if len(prompt) > 0 && multiline == MultilineNone {
-			opts.Prompt = prompt
-			if multiModal {
-				newPrompt, images, err := extractFileData(prompt)
-				if err != nil {
-					return err
-				}
-				opts.Prompt = newPrompt
-
-				// reset the context if we find another image
-				if len(images) > 0 {
-					opts.Images = images
-					ctx := cmd.Context()
-					ctx = context.WithValue(ctx, generateContextKey("context"), []int{})
-					cmd.SetContext(ctx)
-				}
-				if len(opts.Images) == 0 {
-					fmt.Println("This model requires you to add a jpeg, png, or svg image.")
-					fmt.Println()
-					prompt = ""
-					continue
-				}
-			}
-			if err := generate(cmd, opts); err != nil {
-				return err
-			}
-
-			prompt = ""
-		}
-	}
-}
-
-func normalizeFilePath(fp string) string {
-	// Define a map of escaped characters and their replacements
-	replacements := map[string]string{
-		"\\ ":  " ",  // Escaped space
-		"\\(":  "(",  // Escaped left parenthesis
-		"\\)":  ")",  // Escaped right parenthesis
-		"\\[":  "[",  // Escaped left square bracket
-		"\\]":  "]",  // Escaped right square bracket
-		"\\{":  "{",  // Escaped left curly brace
-		"\\}":  "}",  // Escaped right curly brace
-		"\\$":  "$",  // Escaped dollar sign
-		"\\&":  "&",  // Escaped ampersand
-		"\\;":  ";",  // Escaped semicolon
-		"\\'":  "'",  // Escaped single quote
-		"\\\\": "\\", // Escaped backslash
-		"\\*":  "*",  // Escaped asterisk
-		"\\?":  "?",  // Escaped question mark
-	}
-
-	for escaped, actual := range replacements {
-		fp = strings.ReplaceAll(fp, escaped, actual)
-	}
-	return fp
-}
-
-func extractFileNames(input string) []string {
-	// Regex to match file paths starting with / or ./ and include escaped spaces (\ or %20)
-	// and followed by more characters and a file extension
-	regexPattern := `(?:\./|/)[\S\\ ]+?\.(?i:jpg|jpeg|png|svg)\b`
-	re := regexp.MustCompile(regexPattern)
-
-	return re.FindAllString(input, -1)
-}
-
-func extractFileData(input string) (string, []ImageData, error) {
-	filePaths := extractFileNames(input)
-	var imgs []ImageData
-
-	for _, fp := range filePaths {
-		nfp := normalizeFilePath(fp)
-		data, err := getImageData(nfp)
-		if err != nil {
-			if os.IsNotExist(err) {
-				continue
-			}
-			fmt.Printf("Couldn't process image: %q\n", err)
-			return "", imgs, err
-		}
-		fmt.Printf("Added image '%s'\n", nfp)
-		input = strings.ReplaceAll(input, fp, "")
-		imgs = append(imgs, data)
-	}
-	return input, imgs, nil
-}
-
-func getImageData(filePath string) ([]byte, error) {
-	file, err := os.Open(filePath)
-	if err != nil {
-		return nil, err
-	}
-	defer file.Close()
-
-	buf := make([]byte, 512)
-	_, err = file.Read(buf)
-	if err != nil {
-		return nil, err
-	}
-
-	contentType := http.DetectContentType(buf)
-	allowedTypes := []string{"image/jpeg", "image/jpg", "image/svg+xml", "image/png"}
-	if !slices.Contains(allowedTypes, contentType) {
-		return nil, fmt.Errorf("invalid image type: %s", contentType)
-	}
-
-	info, err := file.Stat()
-	if err != nil {
-		return nil, err
-	}
-
-	// Check if the file size exceeds 100MB
-	var maxSize int64 = 100 * 1024 * 1024 // 100MB in bytes
-	if info.Size() > maxSize {
-		return nil, fmt.Errorf("file size exceeds maximum limit (100MB)")
-	}
-
-	buf = make([]byte, info.Size())
-	_, err = file.Seek(0, 0)
-	if err != nil {
-		return nil, err
-	}
-
-	_, err = io.ReadFull(file, buf)
-	if err != nil {
-		return nil, err
-	}
-
-	return buf, nil
-}
--- a/docs/README.md
+++ b/docs/README.md
@@ -1,25 +1,6 @@
 # Documentation

-To get started, see the project's **[quickstart](../README.md#quickstart)**.
-
-Ollama is a tool for running AI models on your hardware. Many users will choose to use the Command Line Interface (CLI) to work with Ollama. Learn more about all the commands in the CLI in the **[Main Readme](../README.md)**.
-
-Use the RESTful API using any language, including Python, JavaScript, Typescript, Go, Rust, and many more. Learn more about using the API in the **[API Documentation](./api.md)**.
-
-Create new models or modify models already in the library using the Modelfile. Learn more about the Modelfile syntax in the **[Modelfile Documentation](./modelfile.md)**.
-
-Import models using source model weights found on Hugging Face and similar sites by referring to the **[Import Documentation](./import.md)**.
-
-Installing on Linux in most cases is easy using the script on Ollama.ai. To get more detail about the install, including CUDA drivers, see the **[Linux Documentation](./linux.md)**.
-
-Many of our users like the flexibility of using our official Docker Image. Learn more about using Docker with Ollama using the **[Docker Documentation](https://hub.docker.com/r/ollama/ollama)**.
-
-It is easy to install on Linux and Mac, but many users will choose to build Ollama on their own. To do this, refer to the **[Development Documentation](./development.md)**.
-
-If encountering a problem with Ollama, the best place to start is the logs. Find more information about them here in the **[Troubleshooting Guide](./troubleshooting.md)**.
-
-Finally for all the questions that don't fit anywhere else, there is the **[FAQ](./faq.md)**
-
-[Tutorials](./tutorials.md) apply the documentation to tasks.
-
-For working code examples of using Ollama, see [Examples](../examples).
+- [Modelfile](./modelfile.md)
+- [How to develop Ollama](./development.md)
+- [API](./api.md)
+- [Tutorials](./tutorials.md)
--- a/docs/api.md
+++ b/docs/api.md
@@ -3,7 +3,6 @@
 ## Endpoints

 - [Generate a completion](#generate-a-completion)
- [Generate a chat completion](#generate-a-chat-completion)
 - [Create a Model](#create-a-model)
 - [List Local Models](#list-local-models)
 - [Show Model Information](#show-model-information)
@@ -17,7 +16,7 @@

 ### Model names

-Model names follow a `model:tag` format, where `model` can have an optional namespace such as `example/model`. Some examples are `orca-mini:3b-q4_1` and `llama2:70b`. The tag is optional and, if not provided, will default to `latest`. The tag is used to identify a specific version.
+Model names follow a `model:tag` format. Some examples are `orca-mini:3b-q4_1` and `llama2:70b`. The tag is optional and, if not provided, will default to `latest`. The tag is used to identify a specific version.

 ### Durations

@@ -25,7 +24,7 @@ All durations are returned in nanoseconds.

 ### Streaming responses

-Certain endpoints stream responses as JSON objects and can optional return non-streamed responses.
+Certain endpoints stream responses as JSON objects delineated with the newline (`\n`) character.

 ## Generate a completion

@@ -33,35 +32,32 @@ Certain endpoints stream responses as JSON objects and can optional return non-s
 POST /api/generate
 ```

-Generate a response for a given prompt with a provided model. This is a streaming endpoint, so there will be a series of responses. The final response object will include statistics and additional data from the request.
+Generate a response for a given prompt with a provided model. This is a streaming endpoint, so will be a series of responses. The final response object will include statistics and additional data from the request.

 ### Parameters

 - `model`: (required) the [model name](#model-names)
 - `prompt`: the prompt to generate a response for
- `images`: (optional) a list of base64-encoded images (for multimodal models such as `llava`)

 Advanced parameters (optional):

 - `format`: the format to return a response in. Currently the only accepted value is `json`
 - `options`: additional model parameters listed in the documentation for the [Modelfile](./modelfile.md#valid-parameters-and-values) such as `temperature`
- `system`: system message to (overrides what is defined in the `Modelfile`)
- `template`: the prompt template to use (overrides what is defined in the `Modelfile`)
+- `system`: system prompt to (overrides what is defined in the `Modelfile`)
+- `template`: the full prompt or prompt template (overrides what is defined in the `Modelfile`)
 - `context`: the context parameter returned from a previous request to `/generate`, this can be used to keep a short conversational memory
 - `stream`: if `false` the response will be returned as a single response object, rather than a stream of objects
- `raw`: if `true` no formatting will be applied to the prompt. You may choose to use the `raw` parameter if you are specifying a full templated prompt in your request to the API.
+- `raw`: if `true` no formatting will be applied to the prompt and no context will be returned. You may choose to use the `raw` parameter if you are specifying a full templated prompt in your request to the API, and are managing history yourself.

-#### JSON mode
+### JSON mode

-Enable JSON mode by setting the `format` parameter to `json`. This will structure the response as a valid JSON object. See the JSON mode [example](#generate-request-json-mode) below.
+Enable JSON mode by setting the `format` parameter to `json`. This will structure the response as valid JSON. See the JSON mode [example](#request-json-mode) below.

 > Note: it's important to instruct the model to use JSON in the `prompt`. Otherwise, the model may generate large amounts whitespace.

 ### Examples

-#### Generate request (Streaming)
-
-##### Request
+#### Request

 ```shell
 curl http://localhost:11434/api/generate -d '{
@@ -70,7 +66,7 @@ curl http://localhost:11434/api/generate -d '{
 }'
 ```

-##### Response
+#### Response

 A stream of JSON objects is returned:

@@ -87,6 +83,8 @@ The final response in the stream also includes additional data about the generat

 - `total_duration`: time spent generating the response
 - `load_duration`: time spent in nanoseconds loading the model
+- `sample_count`: number of samples generated
+- `sample_duration`: time spent generating samples
 - `prompt_eval_count`: number of tokens in the prompt
 - `prompt_eval_duration`: time spent in nanoseconds evaluating the prompt
 - `eval_count`: number of tokens the response
@@ -101,23 +99,21 @@ To calculate how fast the response is generated in tokens per second (token/s),
  "model": "llama2",
  "created_at": "2023-08-04T19:22:45.499127Z",
  "response": "",
-  "done": true,
  "context": [1, 2, 3],
-  "total_duration": 10706818083,
-  "load_duration": 6338219291,
-  "prompt_eval_count": 26,
-  "prompt_eval_duration": 130079000,
-  "eval_count": 259,
-  "eval_duration": 4232710000
+  "done": true,
+  "total_duration": 5589157167,
+  "load_duration": 3013701500,
+  "sample_count": 114,
+  "sample_duration": 81442000,
+  "prompt_eval_count": 46,
+  "prompt_eval_duration": 1160282000,
+  "eval_count": 113,
+  "eval_duration": 1325948000
 }
 ```

 #### Request (No streaming)

-##### Request
-
-A response can be received in one reply when streaming is off.
-
 ```shell
 curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
@@ -126,7 +122,7 @@ curl http://localhost:11434/api/generate -d '{
 }'
 ```

-##### Response
+#### Response

 If `stream` is set to `false`, the response will be a single JSON object:

@@ -135,23 +131,51 @@ If `stream` is set to `false`, the response will be a single JSON object:
  "model": "llama2",
  "created_at": "2023-08-04T19:22:45.499127Z",
  "response": "The sky is blue because it is the color of the sky.",
-  "done": true,
  "context": [1, 2, 3],
-  "total_duration": 5043500667,
-  "load_duration": 5025959,
-  "prompt_eval_count": 26,
-  "prompt_eval_duration": 325953000,
-  "eval_count": 290,
-  "eval_duration": 4709213000
+  "done": true,
+  "total_duration": 5589157167,
+  "load_duration": 3013701500,
+  "sample_count": 114,
+  "sample_duration": 81442000,
+  "prompt_eval_count": 46,
+  "prompt_eval_duration": 1160282000,
+  "eval_count": 13,
+  "eval_duration": 1325948000
+}
+```
+
+#### Request (Raw mode)
+
+In some cases you may wish to bypass the templating system and provide a full prompt. In this case, you can use the `raw` parameter to disable formatting and context.
+
+```shell
+curl http://localhost:11434/api/generate -d '{
+  "model": "mistral",
+  "prompt": "[INST] why is the sky blue? [/INST]",
+  "raw": true,
+  "stream": false
+}'
+```
+
+#### Response
+
+```json
+{
+  "model": "mistral",
+  "created_at": "2023-11-03T15:36:02.583064Z",
+  "response": " The sky appears blue because of a phenomenon called Rayleigh scattering.",
+  "done": true,
+  "total_duration": 14648695333,
+  "load_duration": 3302671417,
+  "prompt_eval_count": 14,
+  "prompt_eval_duration": 286243000,
+  "eval_count": 129,
+  "eval_duration": 10931424000
 }
 ```

 #### Request (JSON mode)

-> When `format` is set to `json`, the output will always be a well-formed JSON object. It's important to also instruct the model to respond in JSON.
-
-##### Request
-
 ```shell
 curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
@@ -161,7 +185,7 @@ curl http://localhost:11434/api/generate -d '{
 }'
 ```

-##### Response
+#### Response

 ```json
 {
@@ -169,13 +193,12 @@ curl http://localhost:11434/api/generate -d '{
  "created_at": "2023-11-09T21:07:55.186497Z",
  "response": "{\n\"morning\": {\n\"color\": \"blue\"\n},\n\"noon\": {\n\"color\": \"blue-gray\"\n},\n\"afternoon\": {\n\"color\": \"warm gray\"\n},\n\"evening\": {\n\"color\": \"orange\"\n}\n}\n",
  "done": true,
-  "context": [1, 2, 3],
-  "total_duration": 4648158584,
-  "load_duration": 4071084,
+  "total_duration": 4661289125,
+  "load_duration": 1714434500,
  "prompt_eval_count": 36,
-  "prompt_eval_duration": 439038000,
-  "eval_count": 180,
-  "eval_duration": 4196918000
+  "prompt_eval_duration": 264132000,
+  "eval_count": 75,
+  "eval_duration": 2112149000
 }
 ```

@@ -198,77 +221,10 @@ The value of `response` will be a string containing JSON similar to:
 }
 ```

-#### Request (with images)
-
-To submit images to multimodal models such as `llava` or `bakllava`, provide a list of base64-encoded `images`:
-
-#### Request
-
-```shell
-curl http://localhost:11434/api/generate -d '{
-  "model": "llava",
-  "prompt":"What is in this picture?",
-  "stream": false,
-  "images": ["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"]
-}'
-```
-
-#### Response
-
-```
-{
-  "model": "llava",
-  "created_at": "2023-11-03T15:36:02.583064Z",
-  "response": "A happy cartoon character, which is cute and cheerful.",
-  "done": true,
-  "context": [1, 2, 3],
-  "total_duration": 2938432250,
-  "load_duration": 2559292,
-  "prompt_eval_count": 1,
-  "prompt_eval_duration": 2195557000,
-  "eval_count": 44,
-  "eval_duration": 736432000
-}
-```
-
-#### Request (Raw Mode)
-
-In some cases, you may wish to bypass the templating system and provide a full prompt. In this case, you can use the `raw` parameter to disable templating. Also note that raw mode will not return a context.
-
-##### Request
-
-```shell
-curl http://localhost:11434/api/generate -d '{
-  "model": "mistral",
-  "prompt": "[INST] why is the sky blue? [/INST]",
-  "raw": true,
-  "stream": false
-}'
-```
-
-##### Response
-
-```json
-{
-  "model": "mistral",
-  "created_at": "2023-11-03T15:36:02.583064Z",
-  "response": " The sky appears blue because of a phenomenon called Rayleigh scattering.",
-  "done": true,
-  "total_duration": 8493852375,
-  "load_duration": 6589624375,
-  "prompt_eval_count": 14,
-  "prompt_eval_duration": 119039000,
-  "eval_count": 110,
-  "eval_duration": 1779061000
-}
-```
-
-#### Generate request (With options)
+#### Request (With options)

 If you want to set custom options for the model at runtime rather than in the Modelfile, you can do so with the `options` parameter. This example sets every available option, but you can set any of them individually and omit the ones you do not want to override.

-##### Request
-
 ```shell
 curl http://localhost:11434/api/generate -d '{
  "model": "llama2",
@@ -293,13 +249,14 @@ curl http://localhost:11434/api/generate -d '{
    "penalize_newline": true,
    "stop": ["\n", "user:"],
    "numa": false,
-    "num_ctx": 1024,
+    "num_ctx": 4,
    "num_batch": 2,
    "num_gqa": 1,
    "num_gpu": 1,
    "main_gpu": 0,
    "low_vram": false,
    "f16_kv": true,
+    "logits_all": false,
    "vocab_only": false,
    "use_mmap": true,
    "use_mlock": false,
@@ -307,264 +264,27 @@ curl http://localhost:11434/api/generate -d '{
    "rope_frequency_base": 1.1,
    "rope_frequency_scale": 0.8,
    "num_thread": 8
-  }
+    }
 }'
 ```

-##### Response
+#### Response

 ```json
 {
  "model": "llama2",
  "created_at": "2023-08-04T19:22:45.499127Z",
  "response": "The sky is blue because it is the color of the sky.",
-  "done": true,
  "context": [1, 2, 3],
-  "total_duration": 4935886791,
-  "load_duration": 534986708,
-  "prompt_eval_count": 26,
-  "prompt_eval_duration": 107345000,
-  "eval_count": 237,
-  "eval_duration": 4289432000
-}
-```
-
-#### Load a model
-
-If an empty prompt is provided, the model will be loaded into memory.
-
-##### Request
-
-```shell
-curl http://localhost:11434/api/generate -d '{
-  "model": "llama2"
-}'
-```
-
-##### Response
-
-A single JSON object is returned:
-
-```json
-{
-  "model": "llama2",
-  "created_at": "2023-12-18T19:52:07.071755Z",
-  "response": "",
-  "done": true
-}
-```
-
-## Generate a chat completion
-
-```shell
-POST /api/chat
-```
-
-Generate the next message in a chat with a provided model. This is a streaming endpoint, so there will be a series of responses. Streaming can be disabled using `"stream": false`. The final response object will include statistics and additional data from the request.
-
-### Parameters
-
- `model`: (required) the [model name](#model-names)
- `messages`: the messages of the chat, this can be used to keep a chat memory
-
-The `message` object has the following fields:
-
- `role`: the role of the message, either `system`, `user` or `assistant`
- `content`: the content of the message
- `images` (optional): a list of images to include in the message (for multimodal models such as `llava`)
-
-Advanced parameters (optional):
-
- `format`: the format to return a response in. Currently the only accepted value is `json`
- `options`: additional model parameters listed in the documentation for the [Modelfile](./modelfile.md#valid-parameters-and-values) such as `temperature`
- `template`: the prompt template to use (overrides what is defined in the `Modelfile`)
- `stream`: if `false` the response will be returned as a single response object, rather than a stream of objects
-
-### Examples
-
-#### Chat Request (Streaming)
-
-##### Request
-
-Send a chat message with a streaming response.
-
-```shell
-curl http://localhost:11434/api/chat -d '{
-  "model": "llama2",
-  "messages": [
-    {
-      "role": "user",
-      "content": "why is the sky blue?"
-    }
-  ]
-}'
-```
-
-##### Response
-
-A stream of JSON objects is returned:
-
-```json
-{
-  "model": "llama2",
-  "created_at": "2023-08-04T08:52:19.385406455-07:00",
-  "message": {
-    "role": "assisant",
-    "content": "The",
-    "images": null
-  },
-  "done": false
-}
-```
-
-Final response:
-
-```json
-{
-  "model": "llama2",
-  "created_at": "2023-08-04T19:22:45.499127Z",
  "done": true,
-  "total_duration": 4883583458,
-  "load_duration": 1334875,
-  "prompt_eval_count": 26,
-  "prompt_eval_duration": 342546000,
-  "eval_count": 282,
-  "eval_duration": 4535599000
-}
-```
-
-#### Chat request (No streaming)
-
-##### Request
-
-```shell
-curl http://localhost:11434/api/chat -d '{
-  "model": "llama2",
-  "messages": [
-    {
-      "role": "user",
-      "content": "why is the sky blue?"
-    }
-  ],
-  "stream": false
-}'
-```
-
-##### Response
-
-```json
-{
-  "model": "registry.ollama.ai/library/llama2:latest",
-  "created_at": "2023-12-12T14:13:43.416799Z",
-  "message": {
-    "role": "assistant",
-    "content": "Hello! How are you today?"
-  },
-  "done": true,
-  "total_duration": 5191566416,
-  "load_duration": 2154458,
-  "prompt_eval_count": 26,
-  "prompt_eval_duration": 383809000,
-  "eval_count": 298,
-  "eval_duration": 4799921000
-}
-```
-
-#### Chat request (With History)
-
-Send a chat message with a conversation history. You can use this same approach to start the conversation using multi-shot or chain-of-thought prompting.
-
-##### Request
-
-```shell
-curl http://localhost:11434/api/chat -d '{
-  "model": "llama2",
-  "messages": [
-    {
-      "role": "user",
-      "content": "why is the sky blue?"
-    },
-    {
-      "role": "assistant",
-      "content": "due to rayleigh scattering."
-    },
-    {
-      "role": "user",
-      "content": "how is that different than mie scattering?"
-    }
-  ]
-}'
-```
-
-##### Response
-
-A stream of JSON objects is returned:
-
-```json
-{
-  "model": "llama2",
-  "created_at": "2023-08-04T08:52:19.385406455-07:00",
-  "message": {
-    "role": "assisant",
-    "content": "The"
-  },
-  "done": false
-}
-```
-
-Final response:
-
-```json
-{
-  "model": "llama2",
-  "created_at": "2023-08-04T19:22:45.499127Z",
-  "done": true,
-  "total_duration": 8113331500,
-  "load_duration": 6396458,
-  "prompt_eval_count": 61,
-  "prompt_eval_duration": 398801000,
-  "eval_count": 468,
-  "eval_duration": 7701267000
-}
-```
-
-#### Chat request (with images)
-
-##### Request
-
-Send a chat message with a conversation history.
-
-```shell
-curl http://localhost:11434/api/chat -d '{
-  "model": "llava",
-  "messages": [
-    {
-      "role": "user",
-      "content": "what is in this image?",
-      "images": ["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"]
-    },
-  ]
-}'
-```
-
-##### Response
-
-```json
-{
-  "model": "llava",
-  "created_at": "2023-12-13T22:42:50.203334Z",
-  "message": {
-    "role": "assistant",
-    "content": " The image features a cute, little pig with an angry facial expression. It's wearing a heart on its shirt and is waving in the air. This scene appears to be part of a drawing or sketching project.",
-    "images": null
-  },
-  "done": true,
-  "total_duration": 1668506709,
-  "load_duration": 1986209,
-  "prompt_eval_count": 26,
-  "prompt_eval_duration": 359682000,
-  "eval_count": 83,
-  "eval_duration": 1303285000
+  "total_duration": 5589157167,
+  "load_duration": 3013701500,
+  "sample_count": 114,
+  "sample_duration": 81442000,
+  "prompt_eval_count": 46,
+  "prompt_eval_duration": 1160282000,
+  "eval_count": 13,
+  "eval_duration": 1325948000
 }
 ```

@@ -574,7 +294,7 @@ curl http://localhost:11434/api/chat -d '{
 POST /api/create
 ```

-Create a model from a [`Modelfile`](./modelfile.md). It is recommended to set `modelfile` to the content of the Modelfile rather than just set `path`. This is a requirement for remote create. Remote model creation must also create any file blobs, fields such as `FROM` and `ADAPTER`, explicitly with the server using [Create a Blob](#create-a-blob) and the value to the path indicated in the response.
+Create a model from a [`Modelfile`](./modelfile.md). It is recommended to set `modelfile` to the content of the Modelfile rather than just set `path`. This is a requirement for remote create. Remote model creation should also create any file blobs, fields such as `FROM` and `ADAPTER`, explicitly with the server using [Create a Blob](#create-a-blob) and the value to the path indicated in the response.

 ### Parameters

@@ -585,11 +305,7 @@ Create a model from a [`Modelfile`](./modelfile.md). It is recommended to set `m

 ### Examples

-#### Create a new model
-
-Create a new model from a `Modelfile`.
-
-##### Request
+#### Request

 ```shell
 curl http://localhost:11434/api/create -d '{
@@ -598,22 +314,14 @@ curl http://localhost:11434/api/create -d '{
 }'
 ```

-##### Response
+#### Response

-A stream of JSON objects. Notice that the final JSON object shows a `"status": "success"`.
+A stream of JSON objects. When finished, `status` is `success`.

 ```json
-{"status":"reading model metadata"}
-{"status":"creating system layer"}
-{"status":"using already created layer sha256:22f7f8ef5f4c791c1b03d7eb414399294764d7cc82c7e94aa81a1feb80a983a2"}
-{"status":"using already created layer sha256:8c17c2ebb0ea011be9981cc3922db8ca8fa61e828c5d3f44cb6ae342bf80460b"}
-{"status":"using already created layer sha256:7c23fb36d80141c4ab8cdbb61ee4790102ebd2bf7aeff414453177d4f2110e5d"}
-{"status":"using already created layer sha256:2e0493f67d0c8c9c68a8aeacdf6a38a2151cb3c4c1d42accf296e19810527988"}
-{"status":"using already created layer sha256:2759286baa875dc22de5394b4a925701b1896a7e3f8e53275c36f75a877a82c9"}
-{"status":"writing layer sha256:df30045fe90f0d750db82a058109cecd6d4de9c90a3d75b19c09e5f64580bb42"}
-{"status":"writing layer sha256:f18a68eb09bf925bb1b669490407c1b1251c5db98dc4d3d81f3088498ea55690"}
-{"status":"writing manifest"}
-{"status":"success"}
+{
+  "status": "parsing modelfile"
+}
 ```

 ### Check if a Blob Exists
@@ -622,7 +330,7 @@ A stream of JSON objects. Notice that the final JSON object shows a `"status": "
 HEAD /api/blobs/:digest
 ```

-Ensures that the file blob used for a FROM or ADAPTER field exists on the server. This is checking your Ollama server and not Ollama.ai.
+Check if a blob is known to the server.

 #### Query Parameters

@@ -646,7 +354,7 @@ Return 200 OK if the blob exists, 404 Not Found if it does not.
 POST /api/blobs/:digest
 ```

-Create a blob from a file on the server. Returns the server file path.
+Create a blob from a file. Returns the server file path.

 #### Query Parameters

@@ -662,7 +370,7 @@ curl -T model.bin -X POST http://localhost:11434/api/blobs/sha256:29fdb92e57cf08

 ##### Response

-Return 201 Created if the blob was successfully created, 400 Bad Request if the digest used is not expected.
+Return 201 Created if the blob was successfully created.

 ## List Local Models

@@ -688,30 +396,14 @@ A single JSON object will be returned.
 {
  "models": [
    {
-      "name": "codellama:13b",
-      "modified_at": "2023-11-04T14:56:49.277302595-07:00",
-      "size": 7365960935,
-      "digest": "9f438cb9cd581fc025612d27f7c1a6669ff83a8bb0ed86c94fcf4c5440555697",
-      "details": {
-        "format": "gguf",
-        "family": "llama",
-        "families": null,
-        "parameter_size": "13B",
-        "quantization_level": "Q4_0"
-      }
+      "name": "llama2",
+      "modified_at": "2023-08-02T17:02:23.713454393-07:00",
+      "size": 3791730596
    },
    {
-      "name": "llama2:latest",
-      "modified_at": "2023-12-07T09:32:18.757212583-08:00",
-      "size": 3825819519,
-      "digest": "fe938a131f40e6f6d40083c9f0f430a515233eb2edaa6d72eb85c50d64f2300e",
-      "details": {
-        "format": "gguf",
-        "family": "llama",
-        "families": null,
-        "parameter_size": "7B",
-        "quantization_level": "Q4_0"
-      }
+      "name": "llama2:13b",
+      "modified_at": "2023-08-08T12:08:38.093596297-07:00",
+      "size": 7323310500
    }
  ]
 }
@@ -723,7 +415,7 @@ A single JSON object will be returned.
 POST /api/show
 ```

-Show information about a model including details, modelfile, template, parameters, license, and system prompt.
+Show details about a model including modelfile, template, parameters, license, and system prompt.

 ### Parameters

@@ -743,16 +435,10 @@ curl http://localhost:11434/api/show -d '{

 ```json
 {
-  "modelfile": "# Modelfile generated by \"ollama show\"\n# To build a new Modelfile based on this one, replace the FROM line with:\n# FROM llava:latest\n\nFROM /Users/matt/.ollama/models/blobs/sha256:200765e1283640ffbd013184bf496e261032fa75b99498a9613be4e94d63ad52\nTEMPLATE \"\"\"{{ .System }}\nUSER: {{ .Prompt }}\nASSSISTANT: \"\"\"\nPARAMETER num_ctx 4096\nPARAMETER stop \"\u003c/s\u003e\"\nPARAMETER stop \"USER:\"\nPARAMETER stop \"ASSSISTANT:\"",
-  "parameters": "num_ctx                        4096\nstop                           \u003c/s\u003e\nstop                           USER:\nstop                           ASSSISTANT:",
-  "template": "{{ .System }}\nUSER: {{ .Prompt }}\nASSSISTANT: ",
-  "details": {
-    "format": "gguf",
-    "family": "llama",
-    "families": ["llama", "clip"],
-    "parameter_size": "7B",
-    "quantization_level": "Q4_0"
-  }
+  "license": "<contents of license block>",
+  "modelfile": "# Modelfile generated by \"ollama show\"\n# To build a new Modelfile based on this one, replace the FROM line with:\n# FROM llama2:latest\n\nFROM /Users/username/.ollama/models/blobs/sha256:8daa9615cce30c259a9555b1cc250d461d1bc69980a274b44d7eda0be78076d8\nTEMPLATE \"\"\"[INST] {{ if and .First .System }}<<SYS>>{{ .System }}<</SYS>>\n\n{{ end }}{{ .Prompt }} [/INST] \"\"\"\nSYSTEM \"\"\"\"\"\"\nPARAMETER stop [INST]\nPARAMETER stop [/INST]\nPARAMETER stop <<SYS>>\nPARAMETER stop <</SYS>>\n",
+  "parameters": "stop                           [INST]\nstop                           [/INST]\nstop                           <<SYS>>\nstop                           <</SYS>>",
+  "template": "[INST] {{ if and .First .System }}<<SYS>>{{ .System }}<</SYS>>\n\n{{ end }}{{ .Prompt }} [/INST] "
 }
 ```

@@ -777,7 +463,7 @@ curl http://localhost:11434/api/copy -d '{

 #### Response

-Returns a 200 OK if successful, or a 404 Not Found if the source model doesn't exist.
+The only response is a 200 OK if successful.

 ## Delete a Model

@@ -803,7 +489,7 @@ curl -X DELETE http://localhost:11434/api/delete -d '{

 #### Response

-Returns a 200 OK if successful, 404 Not Found if the model to be deleted doesn't exist.
+If successful, the only response is a 200 OK.

 ## Pull a Model

--- a/docs/development.md
+++ b/docs/development.md
@@ -14,13 +14,7 @@ Install required tools:
 brew install go cmake gcc
 ```

-Optionally enable debugging and more verbose logging:
-
-```bash
-export CGO_CFLAGS="-g"
-```
-
-Get the required libraries and build the native LLM code:
+Get the required libraries:

 ```bash
 go generate ./...
@@ -40,35 +34,6 @@ Now you can run `ollama`:

 ## Building on Linux with GPU support

-
-### Linux/Windows CUDA (NVIDIA)
-*Your operating system distribution may already have packages for NVIDIA CUDA. Distro packages are often preferable, but instructions are distro-specific. Please consult distro-specific docs for dependencies if available!*
-
-Install `cmake` and `golang` as well as [NVIDIA CUDA](https://developer.nvidia.com/cuda-downloads) development and runtime packages.
-Then generate dependencies:
-```
-go generate ./...
-```
-Then build the binary:
-```
-go build .
-```
-
-### Linux ROCm (AMD)
-*Your operating system distribution may already have packages for AMD ROCm and CLBlast. Distro packages are often preferable, but instructions are distro-specific. Please consult distro-specific docs for dependencies if available!*
-
-Install [CLBlast](https://github.com/CNugteren/CLBlast/blob/master/doc/installation.md) and [ROCm](https://rocm.docs.amd.com/en/latest/deploy/linux/quick_start.html) developement packages first, as well as `cmake` and `golang`.
-Adjust the paths below (correct for Arch) as appropriate for your distributions install locations and generate dependencies:
-```
-CLBlast_DIR=/usr/lib/cmake/CLBlast ROCM_PATH=/opt/rocm go generate ./...
-```
-Then build the binary:
-```
-go build .
-```
-
-ROCm requires elevated privileges to access the GPU at runtime.  On most distros you can add your user account to the `render` group, or run as root.
-
-## Containerized Build
-
-If you have Docker available, you can build linux binaries with `./scripts/build_linux.sh` which has the CUDA and ROCm dependencies included.
+- Install cmake and nvidia-cuda-toolkit
+- run `go generate ./...`
+- run `go build .`
--- a/docs/faq.md
+++ b/docs/faq.md
@@ -1,88 +1,138 @@
 # FAQ

-## How can I upgrade Ollama?
-
-To upgrade Ollama, run the installation process again. On the Mac, click the Ollama icon in the menubar and choose the restart option if an update is available.
-
 ## How can I view the logs?

-Review the [Troubleshooting](./troubleshooting.md) docs for more about using logs.
+On macOS:

-## How do I use Ollama server environment variables on Mac
+```
+cat ~/.ollama/logs/server.log
+```

-On macOS, Ollama runs in the background and is managed by the menubar app. If adding environment variables, Ollama will need to be run manually.
+On Linux:

-1. Click the menubar icon for Ollama and choose **Quit Ollama**.
-2. Open a new terminal window and run the following command (this example uses `OLLAMA_HOST` with an IP address of `123.1.1.1`):
+```
+journalctl -u ollama
+```

-   ```bash
-   OLLAMA_HOST=123.1.1.1 ollama serve
-   ```
-
-## How do I use Ollama server environment variables on Linux?
-
-If Ollama is installed with the install script, a systemd service was created, running as the Ollama user. To add an environment variable, such as OLLAMA_HOST, follow these steps:
-
-1. Create a `systemd` drop-in directory and add a config file. This is only needed once.
-
-   ```bash
-   mkdir -p /etc/systemd/system/ollama.service.d
-   echo '[Service]' >>/etc/systemd/system/ollama.service.d/environment.conf
-   ```
-
-2. For each environment variable, add it to the config file:
-
-   ```bash
-   echo 'Environment="OLLAMA_HOST=0.0.0.0:11434"' >>/etc/systemd/system/ollama.service.d/environment.conf
-   ```
-
-3. Reload `systemd` and restart Ollama:
-
-   ```bash
-   systemctl daemon-reload
-   systemctl restart ollama
-   ```
+If you're running `ollama serve` directly, the logs will be printed to the console.

 ## How can I expose Ollama on my network?

-Ollama binds to 127.0.0.1 port 11434 by default. Change the bind address with the `OLLAMA_HOST` environment variable. Refer to the section above for how to use environment variables on your platform.
+Ollama binds to 127.0.0.1 port 11434 by default. Change the bind address with the `OLLAMA_HOST` environment variable.
+
+On macOS:
+
+```bash
+OLLAMA_HOST=0.0.0.0:11435 ollama serve
+```
+
+On Linux:
+
+Create a `systemd` drop-in directory and set `Environment=OLLAMA_HOST`
+
+```bash
+mkdir -p /etc/systemd/system/ollama.service.d
+echo '[Service]' >>/etc/systemd/system/ollama.service.d/environment.conf
+```
+
+```bash
+echo 'Environment="OLLAMA_HOST=0.0.0.0:11434"' >>/etc/systemd/system/ollama.service.d/environment.conf
+```
+
+Reload `systemd` and restart Ollama:
+
+```bash
+systemctl daemon-reload
+systemctl restart ollama
+```

 ## How can I allow additional web origins to access Ollama?

-Ollama allows cross-origin requests from `127.0.0.1` and `0.0.0.0` by default. Add additional origins with the `OLLAMA_ORIGINS` environment variable. For example, to add all ports on 192.168.1.1 and https://example.com, use:
+Ollama allows cross origin requests from `127.0.0.1` and `0.0.0.0` by default. Add additional origins with the `OLLAMA_ORIGINS` environment variable:

-```shell
-OLLAMA_ORIGINS=http://192.168.1.1:*,https://example.com
+On macOS:
+
+```bash
+OLLAMA_ORIGINS=http://192.168.1.1:*,https://example.com ollama serve
 ```

-Refer to the section above for how to use environment variables on your platform.
+On Linux:
+
+```bash
+echo 'Environment="OLLAMA_ORIGINS=http://129.168.1.1:*,https://example.com"' >>/etc/systemd/system/ollama.service.d/environment.conf
+```
+
+Reload `systemd` and restart Ollama:
+
+```bash
+systemctl daemon-reload
+systemctl restart ollama
+```

 ## Where are models stored?

- macOS: `~/.ollama/models`.
- Linux: `/usr/share/ollama/.ollama/models`
+- macOS: Raw model data is stored under `~/.ollama/models`.
+- Linux: Raw model data is stored under `/usr/share/ollama/.ollama/models`

-## How do I set them to a different location?
+Below the models directory you will find a structure similar to the following:

-If a different directory needs to be used, set the environment variable `OLLAMA_MODELS` to the chosen directory. Refer to the section above for how to use environment variables on your platform.
+```shell
+.
+├── blobs
+└── manifests
+   └── registry.ollama.ai
+      ├── f0rodo
+      ├── library
+      ├── mattw
+      └── saikatkumardey
+```
+
+There is a `manifests/registry.ollama.ai/namespace` path. In example above, the user has downloaded models from the official `library`, `f0rodo`, `mattw`, and `saikatkumardey` namespaces. Within each of those directories, you will find directories for each of the models downloaded. And in there you will find a file name representing each tag. Each tag file is the manifest for the model.  
+
+The manifest lists all the layers used in this model. You will see a `media type` for each layer, along with a digest. That digest corresponds with a file in the `models/blobs directory`.
+
+### How can I change where Ollama stores models?
+
+To modify where models are stored, you can use the `OLLAMA_MODELS` environment variable. Note that on Linux this means defining `OLLAMA_MODELS` in a drop-in `/etc/systemd/system/ollama.service.d` service file, reloading systemd, and restarting the ollama service.

 ## Does Ollama send my prompts and answers back to Ollama.ai to use in any way?

-No, Ollama runs entirely locally, and conversation data will never leave your machine.
+No. Anything you do with Ollama, such as generate a response from the model, stays with you. We don't collect any data about how you use the model. You are always in control of your own data.

 ## How can I use Ollama in Visual Studio Code?

-There is already a large collection of plugins available for VSCode as well as other editors that leverage Ollama. See the list of [extensions & plugins](https://github.com/jmorganca/ollama#extensions--plugins) at the bottom of the main repository readme.
+There is already a large collection of plugins available for VSCode as well as other editors that leverage Ollama. You can see the list of [extensions & plugins](https://github.com/jmorganca/ollama#extensions--plugins) at the bottom of the main repository readme.

 ## How do I use Ollama behind a proxy?

-Ollama is compatible with proxy servers if `HTTP_PROXY` or `HTTPS_PROXY` are configured. When using either variables, ensure it is set where `ollama serve` can access the values. When using `HTTPS_PROXY`, ensure the proxy certificate is installed as a system certificate. Refer to the section above for how to use environment variables on your platform.
+Ollama is compatible with proxy servers if `HTTP_PROXY` or `HTTPS_PROXY` are configured. When using either variables, ensure it is set where `ollama serve` can access the values.
+
+When using `HTTPS_PROXY`, ensure the proxy certificate is installed as a system certificate.
+
+On macOS:
+
+```bash
+HTTPS_PROXY=http://proxy.example.com ollama serve
+```
+
+On Linux:
+
+```bash
+echo 'Environment="HTTPS_PROXY=https://proxy.example.com"' >>/etc/systemd/system/ollama.service.d/environment.conf
+```
+
+Reload `systemd` and restart Ollama:
+
+```bash
+systemctl daemon-reload
+systemctl restart ollama
+```

 ### How do I use Ollama behind a proxy in Docker?

 The Ollama Docker container image can be configured to use a proxy by passing `-e HTTPS_PROXY=https://proxy.example.com` when starting the container.

-Alternatively, the Docker daemon can be configured to use a proxy. Instructions are available for Docker Desktop on [macOS](https://docs.docker.com/desktop/settings/mac/#proxies), [Windows](https://docs.docker.com/desktop/settings/windows/#proxies), and [Linux](https://docs.docker.com/desktop/settings/linux/#proxies), and Docker [daemon with systemd](https://docs.docker.com/config/daemon/systemd/#httphttps-proxy).
+Alternatively, Docker daemon can be configured to use a proxy. Instructions are available for Docker Desktop on [macOS](https://docs.docker.com/desktop/settings/mac/#proxies), [Windows](https://docs.docker.com/desktop/settings/windows/#proxies), and [Linux](https://docs.docker.com/desktop/settings/linux/#proxies), and Docker [daemon with systemd](https://docs.docker.com/config/daemon/systemd/#httphttps-proxy).

 Ensure the certificate is installed as a system certificate when using HTTPS. This may require a new Docker image when using a self-signed certificate.

@@ -104,11 +154,3 @@ docker run -d -e HTTPS_PROXY=https://my.proxy.example.com -p 11434:11434 ollama-
 The Ollama Docker container can be configured with GPU acceleration in Linux or Windows (with WSL2). This requires the [nvidia-container-toolkit](https://github.com/NVIDIA/nvidia-container-toolkit). See [ollama/ollama](https://hub.docker.com/r/ollama/ollama) for more details.

 GPU acceleration is not available for Docker Desktop in macOS due to the lack of GPU passthrough and emulation.
-
-## Why is networking slow in WSL2 on Windows 10?
-
-This can impact both installing Ollama, as well as downloading models.
-
-Open `Control Panel > Networking and Internet > View network status and tasks` and click on `Change adapter settings` on the left panel. Find the `vEthernel (WSL)` adapter, right click and select `Properties`.
-Click on `Configure` and open the `Advanced` tab. Search through each of the properties until you find `Large Send Offload Version 2 (IPv4)` and `Large Send Offload Version 2 (IPv6)`. *Disable* both of these
-properties.
--- a/docs/import.md
+++ b/docs/import.md
@@ -43,6 +43,7 @@ Ollama supports a set of model architectures, with support for more coming soon:

 - Llama & Mistral
 - Falcon & RW
+- GPT-NeoX
 - BigCode

 To view a model's architecture, check the `config.json` file in its HuggingFace repo. You should see an entry under `architectures` (e.g. `LlamaForCausalLM`).
@@ -72,7 +73,7 @@ docker run --rm -v .:/model ollama/quantize -q q4_0 /model
 This will output two files into the directory:

 - `f16.bin`: the model converted to GGUF
- `q4_0.bin` the model quantized to a 4-bit quantization (Ollama will use this file to create the Ollama model)
+- `q4_0.bin` the model quantized to a 4-bit quantization (we will use this file to create the Ollama model)

 ### Step 3: Write a `Modelfile`

@@ -148,7 +149,6 @@ The quantization options are as follow (from highest highest to lowest levels of
 - `q5_K_M`
 - `q6_K`
 - `q8_0`
- `f16`

 ## Manually converting & quantizing models

@@ -184,6 +184,9 @@ python convert.py <path to model directory>
 # FalconForCausalLM
 python convert-falcon-hf-to-gguf.py <path to model directory>

+# GPTNeoXForCausalLM
+python convert-gptneox-hf-to-gguf.py <path to model directory>
+
 # GPTBigCodeForCausalLM
 python convert-starcoder-hf-to-gguf.py <path to model directory>
 ```
--- a/docs/modelfile.md
+++ b/docs/modelfile.md
@@ -1,6 +1,6 @@
 # Ollama Model File

-> Note: `Modelfile` syntax is in development
+> Note: this `Modelfile` syntax is in development

 A model file is the blueprint to create and share models with Ollama.

@@ -30,14 +30,14 @@ The format of the `Modelfile`:
 INSTRUCTION arguments
 ```

-| Instruction                         | Description                                                    |
-| ----------------------------------- | -------------------------------------------------------------- |
-| [`FROM`](#from-required) (required) | Defines the base model to use.                                 |
-| [`PARAMETER`](#parameter)           | Sets the parameters for how Ollama will run the model.         |
-| [`TEMPLATE`](#template)             | The full prompt template to be sent to the model.              |
-| [`SYSTEM`](#system)                 | Specifies the system message that will be set in the template. |
-| [`ADAPTER`](#adapter)               | Defines the (Q)LoRA adapters to apply to the model.            |
-| [`LICENSE`](#license)               | Specifies the legal license.                                   |
+| Instruction                         | Description                                                   |
+| ----------------------------------- | ------------------------------------------------------------- |
+| [`FROM`](#from-required) (required) | Defines the base model to use.                                |
+| [`PARAMETER`](#parameter)           | Sets the parameters for how Ollama will run the model.        |
+| [`TEMPLATE`](#template)             | The full prompt template to be sent to the model.             |
+| [`SYSTEM`](#system)                 | Specifies the system prompt that will be set in the template. |
+| [`ADAPTER`](#adapter)               | Defines the (Q)LoRA adapters to apply to the model.           |
+| [`LICENSE`](#license)               | Specifies the legal license.                                  |

 ## Examples

@@ -52,7 +52,7 @@ PARAMETER temperature 1
 # sets the context window size to 4096, this controls how many tokens the LLM can use as context to generate the next token
 PARAMETER num_ctx 4096

-# sets a custom system message to specify the behavior of the chat assistant
+# sets a custom system prompt to specify the behavior of the chat assistant
 SYSTEM You are Mario from super mario bros, acting as an assistant.
 ```

@@ -70,12 +70,12 @@ More examples are available in the [examples directory](../examples).
 There are two ways to view `Modelfile`s underlying the models in [ollama.ai/library][1]:

 - Option 1: view a details page from a model's tags page:
-  1.  Go to a particular model's tags (e.g. https://ollama.ai/library/llama2/tags)
-  2.  Click on a tag (e.g. https://ollama.ai/library/llama2:13b)
-  3.  Scroll down to "Layers"
+   1. Go to a particular model's tags (e.g. https://ollama.ai/library/llama2/tags)
+   2. Click on a tag (e.g. https://ollama.ai/library/llama2:13b)
+   3. Scroll down to "Layers"
      - Note: if the [`FROM` instruction](#from-required) is not present,
        it means the model was created from a local file
- Option 2: use `ollama show` to print the `Modelfile` for any local models like so:
+- Option 2: use `ollama show` to print the `Modelfile` like so:

  ```bash
  > ollama show --modelfile llama2:13b
@@ -152,15 +152,15 @@ PARAMETER <parameter> <parametervalue>

 ### TEMPLATE

-`TEMPLATE` of the full prompt template to be passed into the model. It may include (optionally) a system message and a user's prompt. This is used to create a full custom prompt, and syntax may be model specific. You can usually find the template for a given model in the readme for that model.
+`TEMPLATE` of the full prompt template to be passed into the model. It may include (optionally) a system prompt and a user's prompt. This is used to create a full custom prompt, and syntax may be model specific. You can usually find the template for a given model in the readme for that model.

 #### Template Variables

-| Variable        | Description                                                                                                   |
-| --------------- | ------------------------------------------------------------------------------------------------------------- |
-| `{{ .System }}` | The system message used to specify custom behavior, this must also be set in the Modelfile as an instruction. |
-| `{{ .Prompt }}` | The incoming prompt, this is not specified in the model file and will be set based on input.                  |
-| `{{ .First }}`  | A boolean value used to render specific template information for the first generation of a session.           |
+| Variable        | Description                                                                                                  |
+| --------------- | ------------------------------------------------------------------------------------------------------------ |
+| `{{ .System }}` | The system prompt used to specify custom behavior, this must also be set in the Modelfile as an instruction. |
+| `{{ .Prompt }}` | The incoming prompt, this is not specified in the model file and will be set based on input.                 |
+| `{{ .First }}`  | A boolean value used to render specific template information for the first generation of a session.          |

 ```modelfile
 TEMPLATE """
@@ -180,7 +180,7 @@ SYSTEM """<system message>"""

 ### SYSTEM

-The `SYSTEM` instruction specifies the system message to be used in the template, if applicable.
+The `SYSTEM` instruction specifies the system prompt to be used in the template, if applicable.

 ```modelfile
 SYSTEM """<system message>"""
@@ -206,7 +206,7 @@ LICENSE """

 ## Notes

- the **`Modelfile` is not case sensitive**. In the examples, uppercase instructions are used to make it easier to distinguish it from arguments.
- Instructions can be in any order. In the examples, the `FROM` instruction is first to keep it easily readable.
+- the **`Modelfile` is not case sensitive**. In the examples, we use uppercase for instructions to make it easier to distinguish it from arguments.
+- Instructions can be in any order. In the examples, we start with FROM instruction to keep it easily readable.

 [1]: https://ollama.ai/library
--- a/docs/troubleshooting.md
+++ b/docs/troubleshooting.md
@@ -1,22 +0,0 @@
-# How to troubleshoot issues
-
-Sometimes Ollama may not perform as expected. One of the best ways to figure out what happened is to take a look at the logs. Find the logs on Mac by running the command:
-
-```shell
-cat ~/.ollama/logs/server.log
-```
-
-On Linux systems with systemd, the logs can be found with this command:
-
-```shell
-journalctl -u ollama
-```
-
-If manually running `ollama serve` in a terminal, the logs will be on that terminal.
-
-Join the [Discord](https://discord.gg/ollama) for help interpreting the logs.
-
-## Known issues
-
-
-* `signal: illegal instruction (core dumped)`: Ollama requires AVX support from the CPU. This was introduced in 2011 and CPUs started offering it in 2012. CPUs from before that and some lower end CPUs after that may not have AVX support and thus are not supported by Ollama. Some users have had luck with building Ollama on their machines disabling the need for AVX.
--- a/docs/tutorials/fly-gpu.md
+++ b/docs/tutorials/fly-gpu.md
@@ -1,83 +0,0 @@
-# Running Ollama on Fly.io GPU Instances
-
-Ollama runs with little to no configuration on [Fly.io GPU instances](https://fly.io/docs/gpus/gpu-quickstart/). If you don't have access to GPUs yet, you'll need to [apply for access](https://fly.io/gpu/) on the waitlist. Once you're accepted, you'll get an email with instructions on how to get started.
-
-Create a new app with `fly apps create`:
-
-```bash
-fly apps create
-```
-
-Then create a `fly.toml` file in a new folder that looks like this:
-
-```toml
-app = "sparkling-violet-709"
-primary_region = "ord"
-vm.size = "a100-40gb" # see https://fly.io/docs/gpus/gpu-quickstart/ for more info
-
-[build]
-  image = "ollama/ollama"
-
-[http_service]
-  internal_port = 11434
-  force_https = false
-  auto_stop_machines = true
-  auto_start_machines = true
-  min_machines_running = 0
-  processes = ["app"]
-
-[mounts]
-  source = "models"
-  destination = "/root/.ollama"
-  initial_size = "100gb"
-```
-
-Then create a [new private IPv6 address](https://fly.io/docs/reference/private-networking/#flycast-private-load-balancing) for your app:
-
-```bash
-fly ips allocate-v6 --private
-```
-
-Then deploy your app:
-
-```bash
-fly deploy
-```
-
-And finally you can access it interactively with a new Fly.io Machine:
-
-```
-fly machine run -e OLLAMA_HOST=http://your-app-name.flycast --shell ollama/ollama
-```
-
-```bash
-$ ollama run openchat:7b-v3.5-fp16
->>> How do I bake chocolate chip cookies?
- To bake chocolate chip cookies, follow these steps:
-
-1. Preheat the oven to 375°F (190°C) and line a baking sheet with parchment paper or silicone baking mat.
-
-2. In a large bowl, mix together 1 cup of unsalted butter (softened), 3/4 cup granulated sugar, and 3/4
-cup packed brown sugar until light and fluffy.
-
-3. Add 2 large eggs, one at a time, to the butter mixture, beating well after each addition. Stir in 1
-teaspoon of pure vanilla extract.
-
-4. In a separate bowl, whisk together 2 cups all-purpose flour, 1/2 teaspoon baking soda, and 1/2 teaspoon
-salt. Gradually add the dry ingredients to the wet ingredients, stirring until just combined.
-
-5. Fold in 2 cups of chocolate chips (or chunks) into the dough.
-
-6. Drop rounded tablespoons of dough onto the prepared baking sheet, spacing them about 2 inches apart.
-
-7. Bake for 10-12 minutes, or until the edges are golden brown. The centers should still be slightly soft.
-
-8. Allow the cookies to cool on the baking sheet for a few minutes before transferring them to a wire rack
-to cool completely.
-
-Enjoy your homemade chocolate chip cookies!
-```
-
-When you set it up like this, it will automatically turn off when you're done using it. Then when you access it again, it will automatically turn back on. This is a great way to save money on GPU instances when you're not using them. If you want a persistent wake-on-use connection to your Ollama instance, you can set up a [connection to your Fly network using WireGuard](https://fly.io/docs/reference/private-networking/#discovering-apps-through-dns-on-a-wireguard-connection). Then you can access your Ollama instance at `http://your-app-name.flycast`.
-
-And that's it!
--- a/docs/tutorials/langchainpy.md
+++ b/docs/tutorials/langchainpy.md
@@ -42,13 +42,12 @@ text_splitter=RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=0)
 all_splits = text_splitter.split_documents(data)
 ```

-It's split up, but we have to find the relevant splits and then submit those to the model. We can do this by creating embeddings and storing them in a vector database. We can use Ollama directly to instantiate an embedding model. We will use ChromaDB in this example for a vector database. `pip install GPT4All chromadb`
+It's split up, but we have to find the relevant splits and then submit those to the model. We can do this by creating embeddings and storing them in a vector database. For now, we don't have embeddings built in to Ollama, though we will be adding that soon, so for now, we can use the GPT4All library for that. We will use ChromaDB in this example for a vector database. `pip install GPT4All chromadb`

 ```python
-from langchain.embeddings import OllamaEmbeddings
+from langchain.embeddings import GPT4AllEmbeddings
 from langchain.vectorstores import Chroma
-oembed = OllamaEmbeddings(base_url="http://localhost:11434", model="llama2")
-vectorstore = Chroma.from_documents(documents=all_splits, embedding=oembed)
+vectorstore = Chroma.from_documents(documents=all_splits, embedding=GPT4AllEmbeddings())
 ```

 Now let's ask a question from the document. **Who was Neleus, and who is in his family?** Neleus is a character in the Odyssey, and the answer can be found in our text.
--- a/examples/.gitignore
+++ b/examples/.gitignore
@@ -1,10 +1,7 @@
 node_modules
-bun.lockb
-.vscode
 # OSX
 .DS_STORE

-
 # Models
 models/

--- a/examples/golang-simplegenerate/main.go
+++ b/examples/golang-simplegenerate/main.go
@@ -18,8 +18,6 @@ func main() {
 		os.Exit(1)
 	}

-	defer resp.Body.Close()
-	
 	responseData, err := io.ReadAll(resp.Body)
 	if err != nil {
 		log.Fatal(err)
--- a/examples/kubernetes/gpu.yaml
+++ b/examples/kubernetes/gpu.yaml
@@ -25,11 +25,9 @@ spec:
        image: ollama/ollama:latest
        env:
        - name: PATH
-          value: /usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
+          value: /usr/local/nvidia/bin:/usr/local/nvidia/lib64:/usr/bin:/usr/sbin:/bin:/sbin
        - name: LD_LIBRARY_PATH
-          value: /usr/local/nvidia/lib:/usr/local/nvidia/lib64
-        - name: NVIDIA_DRIVER_CAPABILITIES
-          value: compute,utility
+          value: /usr/local/nvidia/lib64
        ports:
        - name: http
          containerPort: 11434
--- a/examples/langchain-python-rag-websummary/README.md
+++ b/examples/langchain-python-rag-websummary/README.md
@@ -1,23 +1,15 @@
 # LangChain Web Summarization

-This example summarizes the website, [https://ollama.ai/blog/run-llama2-uncensored-locally](https://ollama.ai/blog/run-llama2-uncensored-locally)
+This example summarizes a website

-## Running the Example
+## Setup

-1. Ensure you have the `llama2` model installed:
+```
+pip install -r requirements.txt
+```

-   ```bash
-   ollama pull llama2
-   ```
+## Run

-2. Install the Python Requirements.
-
-   ```bash
-   pip install -r requirements.txt
-   ```
-
-3. Run the example:
-
-   ```bash
-   python main.py
-   ```
+```
+python main.py
+```
--- a/examples/langchain-python-rag-websummary/requirements.txt
+++ b/examples/langchain-python-rag-websummary/requirements.txt
@@ -1 +1,2 @@
 langchain==0.0.259
+bs4==0.0.1
--- a/examples/langchain-python-simple/README.md
+++ b/examples/langchain-python-simple/README.md
@@ -2,23 +2,20 @@

 This example is a basic "hello world" of using LangChain with Ollama.

-## Running the Example
+## Setup

-1. Ensure you have the `llama2` model installed:
+```
+pip install -r requirements.txt
+```

-   ```bash
-   ollama pull llama2
-   ```
+## Run

-2. Install the Python Requirements.
+```
+python main.py
+```

-   ```bash
-   pip install -r requirements.txt
-   ```
+Running this example will print the response for "hello":

-3. Run the example:
-
-   ```bash
-   python main.py
-   ```
-  
+```
+Hello! It's nice to meet you. hopefully you are having a great day! Is there something I can help you with or would you like to chat?
+```
--- a/examples/langchain-python-simple/main.py
+++ b/examples/langchain-python-simple/main.py
@@ -1,6 +1,4 @@
 from langchain.llms import Ollama
-
-input = input("What is your question?")
 llm = Ollama(model="llama2")
-res = llm.predict(input)
+res = llm.predict("hello")
 print (res)
--- a/examples/langchain-typescript-simple/README.md
+++ b/examples/langchain-typescript-simple/README.md
@@ -2,22 +2,20 @@

 This example is a basic "hello world" of using LangChain with Ollama using Node.js and Typescript.

-## Running the Example
+## Setup

-1. Install the prerequisites:
+```shell
+npm install
+```

-   ```bash
-   npm install
-   ```
+## Run

-2. Ensure the `mistral` model is available:
+```shell
+ts-node main.ts
+```

-   ```bash
-   ollama pull mistral
-   ```
+Running this example will print the response for "hello":

-3. Run the example:
-
-   ```bash
-   npm start
-   ```
+```plaintext
+Hello! It's nice to meet you. hopefully you are having a great day! Is there something I can help you with or would you like to chat?
+```
--- a/examples/langchain-typescript-simple/main.ts
+++ b/examples/langchain-typescript-simple/main.ts
@@ -1,25 +1,15 @@
-import { Ollama } from 'langchain/llms/ollama';
-import * as readline from "readline";
+import { Ollama} from 'langchain/llms/ollama';

 async function main() {
  const ollama = new Ollama({
    model: 'mistral'    
    // other parameters can be found at https://js.langchain.com/docs/api/llms_ollama/classes/Ollama
-  });
-
-  const rl = readline.createInterface({
-    input: process.stdin,
-    output: process.stdout,
-  });
-
-  rl.question("What is your question: \n", async (user_input) => {
-    const stream = await ollama.stream(user_input);
-  
-    for await (const chunk of stream) {
-      process.stdout.write(chunk);
-    }
-    rl.close();
  })
+  const stream = await ollama.stream("Hello");
+
+  for await (const chunk of stream) {
+    process.stdout.write(chunk);
+  }
 }

 main();
--- a/examples/langchain-typescript-simple/package-lock.json
+++ b/examples/langchain-typescript-simple/package-lock.json
@@ -1,5 +1,5 @@
 {
-  "name": "langchain-typescript-simple",
+  "name": "with-langchain-typescript-simplegenerate",
  "lockfileVersion": 3,
  "requires": true,
  "packages": {
--- a/examples/langchain-typescript-simple/package.json
+++ b/examples/langchain-typescript-simple/package.json
@@ -1,13 +1,8 @@
 {
-  "scripts": {
-    "start": "tsx main.ts"
-  },
  "devDependencies": {
-    "tsx": "^4.6.2",
-    "typescript": "^5.3.3"
+    "typescript": "^5.2.2"
  },
  "dependencies": {
-    "langchain": "^0.0.165",
-    "readline": "^1.3.0"
+    "langchain": "^0.0.165"
  }
 }
--- a/examples/modelfile-10tweets/Modelfile
+++ b/examples/modelfile-10tweets/Modelfile
@@ -0,0 +1,7 @@
+# Modelfile for creating a list of ten tweets from a topic
+# Run `ollama create 10tweets -f ./Modelfile` and then `ollama run 10tweets` and enter a topic
+
+FROM llama2
+SYSTEM """
+You are a content marketer who needs to come up with 10 short but succinct tweets. The answer should be a list of ten tweets. Each tweet can have a maximum of 280 characters and should include hashtags. Each user input will be a subject and you should expand it in ten creative ways. Never stop after just one tweet. Always include ten. 
+"""
--- a/examples/modelfile-10tweets/README.md
+++ b/examples/modelfile-10tweets/README.md
@@ -0,0 +1,23 @@
+# Ten Tweets Modelfile
+
+This is a simple modelfile that generates ten tweets based off any topic.
+
+```bash
+ollama create tentweets
+
+ollama run tentweets
+>>> underwater basketweaving
+ Great! Here are ten creative tweets about underwater basketweaving:
+
+1. "Just discovered the ultimate stress-reliever: Underwater basketweaving! 🌊🧵 #UnderwaterBasketweaving #StressRelief"
+2. "Who needs meditation when you can do underwater basketweaving? 😴👀 #PeacefulDistraction #UnderwaterBasketweaving"
+3. "Just spent an hour in the pool and still managed to knot my basket. Goal: untangle it before next session. 💪🏽 #ChallengeAccepted #UnderwaterBasketweaving"
+4. "When life gives you lemons, make underwater basketweaving! 🍋🧵 #LemonadeLife #UnderwaterBasketweaving"
+5. "Just realized my underwater basketweaving skills could come in handy during a zombie apocalypse. 😂🧡 #SurvivalTips #UnderwaterBasketweaving"
+6. "I'm not lazy, I'm just conserving energy for my next underwater basketweaving session. 😴💤 #LazyDay #UnderwaterBasketweaving"
+7. "Just found my inner peace while doing underwater basketweaving. It's like meditation, but with knots! 🙏🧵 #Mindfulness #UnderwaterBasketweaving"
+8. "Why study for exams when you can do underwater basketweaving and forget all your worries? 😜🧵 #ProcrastinationStation #UnderwaterBasketweaving"
+9. "Just had to cut my underwater basketweaving session short due to a sudden urge to breathe. 🤯🌊 #AquaticAdventures #UnderwaterBasketweaving"
+10. "I'm not sure what's more impressive: my underwater basketweaving skills or the fact that I didn't drown trying to make this tweet. 😅🧵 #Accomplishment 
+#UnderwaterBasketweaving"
+```
--- a/examples/modelfile-devopsengineer/Modelfile
+++ b/examples/modelfile-devopsengineer/Modelfile
@@ -0,0 +1,8 @@
+# Modelfile for creating a devops engineer assistant
+# Run `ollama create devops-engineer -f ./Modelfile` and then `ollama run devops-engineer` and enter a topic
+
+FROM mistral
+PARAMETER temperature 1
+SYSTEM """
+You are a senior devops engineer, acting as an assistant. You offer help with cloud technologies like: Terraform, AWS, kubernetes, python. You answer with code examples when possible
+"""
--- a/examples/modelfile-devopsengineer/README.md
+++ b/examples/modelfile-devopsengineer/README.md
@@ -0,0 +1,40 @@
+# DevOps Engineer Modelfile
+
+This is a simple modelfile that takes will respond to any questions that a senior DevOps engineer should know how to solve.
+
+````text
+ollama create devopsengineer
+
+ollama run devopsengineer
+>>> how can i proxy all connections from inside the office through a single server that caches large files. Provide a config file for nginx to achieve this?
+
+Here is an example configuration file for Nginx that can be used to proxy all connections from inside the office network to a centralized cache server:
+```perl
+http {
+    upstream cache_server {
+        server <cache_server_ip>;
+        server <cache_server_ip>;
+        server <cache_server_ip>;
+    }
+    
+    server {
+        listen 80;
+        server_name localhost;
+        
+        location / {
+            proxy_pass http://cache_server;
+            proxy_set_header Host $host;
+            proxy_set_header X-Real-IP $remote_addr;
+            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+        }
+    }
+}
+```
+In this configuration, we define an `upstream` block that lists the IP addresses of the cache servers. We then define a server block that listens on port 80 and routes all 
+traffic to the `/` location to the cache servers using the `proxy_pass` directive. The `proxy_set_header` directives are used to preserve the source IP address of the client
+request when forwarding it to the cache server.
+
+To use this configuration, you would need to replace the placeholder `<cache_server_ip>` with the actual IP addresses of your cache servers. You would also need to make sure
+that the cache servers are configured to accept incoming connections from the Nginx server and handle requests for files.
+
+````
--- a/examples/modelfile-midjourney/Modelfile
+++ b/examples/modelfile-midjourney/Modelfile
@@ -0,0 +1,11 @@
+# Modelfile for creating a Midjourney prompts from a topic
+# This prompt was adapted from the original at https://www.greataiprompts.com/guide/midjourney/best-chatgpt-prompt-for-midjourney/
+# Run `ollama create mj -f ./Modelfile` and then `ollama run mj` and enter a topic
+
+FROM zephyr
+PARAMETER temperature 0.8
+PARAMETER top_k 500
+PARAMETER top_p 0.9
+SYSTEM """
+Embrace your role as a creative illustrator. Based on a concept provided, you must produce a single paragraph with a multifaceted description of an image, ensuring significant details of the concept and more is represented in your instructions. You do not need to write complete sentences but rather short concepts with the following information: the level of detail that should be represented, an artistic style and maybe a specific name of a painter or illustrator, the ideal color pallete, lighting, mood, perspective, the setting, time of day, weather, the season, the time period, location, materials, the textures, patterns, lines, brushstrokes, techniques, the medium, the genre, the rendering style. Don't include everything and keep the description length under 250 words. 
+"""
--- a/examples/modelfile-midjourney/README.md
+++ b/examples/modelfile-midjourney/README.md
@@ -0,0 +1,11 @@
+# Midjourney Prompt Generator Modelfile
+
+This simple modelfile will help create a prompt to feed to Midjourney.
+
+```text
+ollama create midjourney
+
+ollama run midjourney
+>>> a sports car in the mountains. 
+A sleek, high-performance automobile cuts through a serpentine mountain landscape. The concept is a classic illustration of speed and power, depicted in the style of pop art by Andy Warhol. The color palette is dominated by bold, primary hues of red, blue, and yellow, with striking accent colors of white, black, and metallic shades. The lighting is bright and focused, casting sharp shadows on the rugged terrain. A sense of excitement and anticipation permeates throughout the scene, as the car navigates a treacherous course through the winding road. The perspective is low, allowing for a full view of the vehicle's sleek lines and intricate details. The setting takes place in the afternoon during a sunny day in autumn, as evidenced by the vibrant foliage on the mountainside. The time period is modern, with nods to classic car design. The materials are primarily digital, allowing for smooth curves and sharp contrasts. The textures are sleek and polished, with meticulously detailed lines and brushstrokes that accentuate the car's aerodynamic design. The patterns consist of geometric shapes and bold stripes, adding to the car's dynamic appeal. The genre is modern realism, with a focus on precision and detail. The rendering style is highly technical, capturing the nuances and subtleties of the vehicle and its surroundings in breathtaking detail.
+```
--- a/examples/modelfile-recipemaker/Modelfile
+++ b/examples/modelfile-recipemaker/Modelfile
@@ -0,0 +1,6 @@
+# Modelfile for creating a recipe from a list of ingredients
+# Run `ollama create recipemaker -f ./Modelfile` and then `ollama run recipemaker` and feed it lists of ingredients to create recipes around.
+FROM nous-hermes
+SYSTEM """
+The instruction will be a list of ingredients. You should generate a recipe that can be made in less than an hour. You can also include ingredients that most people will find in their pantry every day. The recipe should be 4 people and you should include a description of what the meal will taste like
+"""
--- a/examples/modelfile-recipemaker/README.md
+++ b/examples/modelfile-recipemaker/README.md
@@ -0,0 +1,20 @@
+# Recipe Maker Modelfile 
+
+Simple modelfile to generate a recipe from a short list of ingredients.
+
+```
+ollama create recipemaker
+
+ollama run recipemaker
+>>> chilli pepper, white chocolate, kale
+ Ingredients:
+- 1 small chili pepper
+- 4 squares of white chocolate
+- handful of kale leaves
+
+Instructions:
+1. In a blender or food processor, puree the chilies and white chocolate until smooth.
+2. Add the chopped kale leaves to the blender and pulse until well combined.
+3. Serve immediately as a dip for crackers or use it as an ingredient in your favorite recipe. The mixture of spicy chili pepper with sweet white chocolate and nutritious 
+kale will make your taste buds dance with delight!
+```
--- a/examples/modelfile-sentiments/Modelfile
+++ b/examples/modelfile-sentiments/Modelfile
@@ -0,0 +1,28 @@
+# Modelfile for creating a sentiment analyzer. 
+# Run `ollama create sentiments -f pathtofile` and then `ollama run sentiments` and enter a topic
+
+FROM orca
+TEMPLATE """
+{{- if .First }}
+### System:
+{{ .System }}
+{{- end }}
+### User: 
+I hate it when my phone dies
+### Response: 
+NEGATIVE
+### User: 
+He is awesome
+### Response: 
+POSITIVE
+### User: 
+This is the link to the article
+### Response: 
+NEUTRAL
+### User:
+{{ .Prompt }}
+
+### Response:
+"""
+
+SYSTEM """You are a sentiment analyzer. You will receive text and output only one word, either POSITIVE or NEGATIVE or NEUTRAL, depending on the sentiment of the text."""
--- a/examples/modelfile-sentiments/Readme.md
+++ b/examples/modelfile-sentiments/Readme.md
@@ -0,0 +1,25 @@
+# Sentiments Modelfile
+
+This is a simple sentiments analyzer using the Orca model. When you pull Orca from the registry, it has a Template already defined that looks like this:
+
+```Modelfile
+{{- if .First }}
+### System:
+{{ .System }}
+{{- end }}
+
+### User:
+{{ .Prompt }}
+
+### Response:
+```
+
+If we just wanted to have the text:
+
+```Plaintext
+You are a sentiment analyzer. You will receive text and output only one word, either POSITIVE or NEGATIVE or NEUTRAL, depending on the sentiment of the text.
+```
+
+then we could have put this in a SYSTEM block. But we want to provide examples which require updating the full Template. Any Modelfile you create will inherit all the settings from the source model. But in this example, we are overriding the Template.
+
+When providing examples for the input and output, you should include the way the model usually provides information. Since the Orca model expects a user prompt to appear after ### User: and the response is after ### Response, we should format our examples like that as well. If we were using the Llama 2 model, the format would be a bit different.
--- a/examples/modelfile-tweetwriter/Modelfile
+++ b/examples/modelfile-tweetwriter/Modelfile
@@ -0,0 +1,7 @@
+# Modelfile for creating a tweet from a topic
+# Run `ollama create tweetwriter -f ./Modelfile` and then `ollama run tweetwriter` and enter a topic
+
+FROM nous-hermes
+SYSTEM """
+You are a content marketer who needs to come up with a short but succinct tweet. Make sure to include the appropriate hashtags and links. Sometimes when appropriate, describe a meme that can be included as well. All answers should be in the form of a tweet which has a max size of 280 characters. Every instruction will be the topic to create a tweet about.
+"""
--- a/examples/modelfile-tweetwriter/readme.md
+++ b/examples/modelfile-tweetwriter/readme.md
@@ -1,23 +0,0 @@
-# Example Modelfile - Tweetwriter
-
-This simple examples shows what you can do without any code, simply relying on a Modelfile. The file has two instructions:
-
-1. FROM - The From instructions defines the parent model to use for this one. If you choose a model from the library, you can enter just the model name. For all other models, you need to specify the namespace as well. You could also use a local file. Just include the relative path to the converted, quantized model weights file. To learn more about creating that file, see the `import.md` file in the docs folder of this repository.
-2. SYSTEM - This defines the system prompt for the model and overrides the system prompt from the parent model.
-
-## Running the Example
-
-1. Create the model:
-
-   ```bash
-   ollama create tweetwriter
-   ```
-
-2. Enter a topic to generate a tweet about.
-3. Show the Modelfile in the REPL.
-
-   ```bash
-   /show modelfile
-   ```
-
-   Notice that the FROM and SYSTEM match what was in the file. But there is also a TEMPLATE and PARAMETER. These are inherited from the parent model.
--- a/examples/python-dockerit/README.md
+++ b/examples/python-dockerit/README.md
@@ -1,31 +1,15 @@
 # DockerIt

-DockerIt is a tool to help you build and run your application in a Docker container. It consists of a model that defines the system prompt and model weights to use, along with a python script to then build the container and run the image automatically.
-
-## Running the Example
-
-1. Ensure you have the `mattw/dockerit` model installed:
-
-   ```bash
-   ollama pull mattw/dockerit
-   ```
-
-2. Make sure Docker is running on your machine.
-
-3. Install the Python Requirements.
-
-   ```bash
-   pip install -r requirements.txt
-   ```
-
-4. Run the example:
-
-   ```bash
-   python dockerit.py "simple postgres server with admin password set to 123"
-   ```
-
-5. Enter the name you would like to use for your container image.
+DockerIt is a tool to help you build and run your application in a Docker container. It consists of a model that defines the system prompt and model weights to use, along with a python script to then build the container and run the image automatically. 

 ## Caveats

-This is a simple example. It's assuming the Dockerfile content generated is going to work. In many cases, even with simple web servers, it fails when trying to copy files that don't exist. It's simply an example of what you could possibly do.
+This is an simple example. It's assuming the Dockerfile content generated is going to work. In many cases, even with simple web servers, it fails when trying to copy files that don't exist. It's simply an example of what you could possibly do.
+
+## Example Usage
+
+```bash
+> python3 ./dockerit.py "simple postgres server with admin password set to 123"
+Enter the name of the image: matttest
+Container named happy_keller  started with id:  7c201bb6c30f02b356ddbc8e2a5af9d7d7d7b8c228519c9a501d15c0bd9d6b3e
+```
--- a/examples/python-functioncalling/emails.txt
+++ b/examples/python-functioncalling/emails.txt
@@ -0,0 +1,17 @@
+---
+Hi matt, 
+
+thanks for letting me know that you are going to come today, November 16, for my tea party. My address is 123 Falk St on Bainbridge Island. I live in the house with the red door. I will be home all day so just come by whenever you want.
+
+Fred
+
+---
+Great, send the check to our office at 1917 1st St, Seattle, WA 98101. I will let you know when we receive it.
+
+Mark Richardson
+Big Corp
+---
+We are looking forward to seeing you at our Local AI Meetup. It will be held on December 3. It will be at the offices of Enormous Co. Our address is 344 1st Ave, Seattle, WA 98101. We will be meeting in the conference room on the 3rd floor.
+
+Barbara Reilly
+Enormous Co.
--- a/examples/python-functioncalling/extractemail.py
+++ b/examples/python-functioncalling/extractemail.py
@@ -0,0 +1,108 @@
+import requests
+import json
+
+model = "openchat"
+
+
+def reportEvents(name, date, location):
+    nameString = name if name else "an event"
+    dateString = f" on {date}" if date else ""
+    locationString = f" at {location}" if location else ""
+    print(f"You have an event: {nameString}{dateString}{locationString}")
+
+
+def reportAddresses(address):
+    for field in address:
+        if field == "city":
+            city = address["city"]
+            state = f", {address['state']}" if address["state"] else ""
+            zip = f" {address['zip']}"
+            print(f"{city}{state}{zip}\n")
+            break
+        else:
+            print(address[field])
+
+
+systemPrompt = "You will be given a text along with a prompt and a schema. You will have to extract the information requested in the prompt from the text and generate output in JSON observing the schema provided. If the schema shows a type of integer or number, you must only show a integer for that field. A string should always be a valid string. If a value is unknown, leave it empty. Output the JSON with extra spaces to ensure that it pretty prints."
+
+schema = {
+    "eventsQuantity": {
+        "type": "integer",
+        "description": "The number of events in the source text",
+    },
+    "addressesQuantity": {
+        "type": "integer",
+        "description": "The number of addresses in the source text",
+    },
+    "events": [
+        {
+            "name": {"type": "string", "description": "Name of the event"},
+            "date": {"type": "string", "description": "Date of the event"},
+            "location": {"type": "string", "description": "Location of the event"},
+            "extraInfo": {
+                "type": "string",
+                "description": "Any extra information that is provided about the event.",
+            },
+        }
+    ],
+    "people": [
+        {
+            "name": {"type": "string", "description": "Name of the person"},
+            "company": {
+                "type": "string",
+                "description": "Name of the company where they work",
+            },
+            "street": {
+                "type": "string",
+                "description": "Street address of the person or company. This is only the street name and the numerical address. Do not include city, state, or zip of the address in this field.",
+            },
+            "city": {
+                "type": "string",
+                "description": "City portion of the address of the person or company",
+            },
+            "state": {
+                "type": "string",
+                "description": "State portion of the address of the person or company",
+            },
+            "zip": {
+                "type": "string",
+                "description": "Zip code of the person or company",
+            },
+            "extraInfo": {
+                "type": "string",
+                "description": "Any extra information that is provided about the location.",
+            },
+        }
+    ],
+}
+
+with open("emails.txt") as f:
+    content=f.read()
+
+prompt = f"The source text is a series of emails that have been put into a single file. They are separated by three dashes. Review the source text and determine the full address of the person sending each of the emails as well as any events that we need to track. If they provide a company address use that. If any extra info is provided, such as a description of the place, or a floor, add it to extraInfo. The first field in the address JSON is quantity of events and should be set to the number of events tracked and the second field should be set to the number of addresses tracked in the file. Don't stuff an event into the output that isn't an event. Only add data to the mostly appropriate field. Don't make up fields that aren't in the schema. If there isn't a value for a field, use null. Output should be in JSON.\n\nSchema: \n{schema}\n\nSource Text:\n{content}"
+
+
+r = requests.post(
+    "http://localhost:11434/api/generate",
+    json={
+        "model": model,
+        "system": systemPrompt,
+        "prompt": prompt,
+        "format": "json",
+        "stream": False,
+    },
+)
+
+j = json.loads(r.text)
+
+output = json.loads(j["response"])
+events = output["events"]
+addresses = output["people"]
+
+print(f"Here are your {output['eventsQuantity']} events:")
+for event in events:
+    reportEvents(event["name"], event["date"], event["location"])
+
+print(f"\n\nHere are your {output['addressesQuantity']} addresses")
+for address in addresses:
+    reportAddresses(address)
--- a/examples/python-functioncalling/extractwp.py
+++ b/examples/python-functioncalling/extractwp.py
@@ -0,0 +1,52 @@
+import requests
+import json
+
+model = "orca2"
+
+systemprompt = "You will be given a text along with a prompt and a schema. You will have to extract the information requested in the prompt from the text and generate output in JSON observing the schema provided. If the schema shows a type of integer or number, you must only show a integer for that field. A string should always be a valid string. If a value is unknown, leave it empty. Output the JSON with extra spaces to ensure that it pretty prints."
+
+schema = {
+    "people": [
+        {
+            "name": {"type": "string", "description": "Name of the person"},
+            "title": {"type": "string", "description": "Title of the person"},
+        }
+    ],
+}
+
+# Read the content from the file
+words = []
+with open("wp.txt") as f:
+    maxwords = 2000
+    count = 0
+    lines = f.readlines()
+    for line in lines:
+        for word in line.split(" "):
+            count += 1
+            if count > maxwords:
+                break
+            words.append(word)
+content = ' '.join(words)
+
+# Use the text and schema to set the prompt
+prompt = f"Review the source text and determine 10 the most important people to focus on. Then extract the name and title for those people. Output should be in JSON.\n\nSchema: {schema}\n\nSource Text:\n{content}"
+
+
+# Make the actual request to the model
+r = requests.post(
+    "http://localhost:11434/api/generate",
+    json={
+        "model": model,
+        "system": systemprompt,
+        "prompt": prompt,
+        "format": "json",
+        "stream": False
+    },
+)
+
+# Get the response as JSON.
+j = json.loads(r.text)
+
+# Return the result.
+print(j["response"])
+
--- a/examples/python-functioncalling/readme.md
+++ b/examples/python-functioncalling/readme.md
@@ -0,0 +1,28 @@
+# Function calling
+
+![function calling 2023-11-16 16_12_58](https://github.com/jmorganca/ollama/assets/633681/a0acc247-9746-45ab-b325-b65dfbbee4fb)
+
+Function calling in the context of LLM's simply means that the output of the model is formatted in JSON, using a preconfigured schema, and uses the expected types. Then your code can use the output of the model and call functions with it. Using the JSON format in Ollama, you can use any model for function calling. 
+
+The two examples provided can extract information out of the provided texts. The first example uses the first couple of chapters from War and Peace by Lev Nikolayevich Tolstoy, and extracts the names and titles of the characters introduced in the story. The second example uses a more complicated schema to pull out addresses and event information from a series of emails.
+
+## Running the examples
+
+1. Clone this repo and navigate to the `examples/python-functioncalling` directory.
+2. Install the dependencies with `pip install -r requirements.txt`.
+3. Review the `wp.txt` file.
+4. Run `python extractwp.py`.
+5. Review the `info.txt` file.
+6. Run `python extractemail.py`.
+
+## Review the Code
+
+Both examples do roughly the same thing with different source material. They both use the same system prompt, which tells the model to expect some instructions and a schema. Then we inject the schema into the prompt and generate an answer.
+
+The first example, `extractwp.py`, outputs the resulting JSON to the console, listing the characters introduced at the start of War and Peace. The second example, `extractemail.py`, is a bit more complicated, extracting two different types of information: addresses and events. It outputs the results to a JSON blob, then the addresses are handed off to one function called `reportAddresses` and the events are handed off to another function called `reportEvents`.
+
+Notice that both examples are using the model from Intel called `openchat`. This is not a model tuned for function calling, yet it performs very well at this task.
+
+## Next Steps
+
+Try exporting some of your real emails to the input file and seeing how well the model does. Try pointing the first example at other books. You could even have it cycle through all the sections and maybe add up the number of times any character is seen throughout the book, determining the most important characters. You can also try out different models.
--- a/examples/python-functioncalling/wp.txt
+++ b/examples/python-functioncalling/wp.txt
@@ -0,0 +1,183 @@
+"Well, Prince, so Genoa and Lucca are now just family estates of the Buonapartes. But I warn you, if you don't tell me that this means war, if you still try to defend the infamies and horrors perpetrated by that Antichrist - I really believe he is Antichrist - I will have nothing more to do with you and you are no longer my friend, no longer my 'faithful slave,' as you call yourself! But how do you do? I see I have frightened you - sit down and tell me all the news."
+
+It was in July, 1805, and the speaker was the well-known Anna Pavlovna Scherer, maid of honor and favorite of the Empress Marya Fedorovna. With these words she greeted Prince Vasili Kuragin, a man of high rank and importance, who was the first to arrive at her reception. Anna Pavlovna had had a cough for some days. She was, as she said, suffering from la grippe; grippe being then a new word in St. Petersburg, used only by the elite.
+
+All her invitations without exception, written in French, and delivered by a scarlet-liveried footman that morning, ran as follows:
+
+"If you have nothing better to do, Count (or Prince), and if the prospect of spending an evening with a poor invalid is not too terrible, I shall be very charmed to see you tonight between 7 and 10 - Annette Scherer."
+
+"Heavens! what a virulent attack!" replied the prince, not in the least disconcerted by this reception. He had just entered, wearing an embroidered court uniform, knee breeches, and shoes, and had stars on his breast and a serene expression on his flat face. He spoke in that refined French in which our grandfathers not only spoke but thought, and with the gentle, patronizing intonation natural to a man of importance who had grown old in society and at court. He went up to Anna Pavlovna, kissed her hand, presenting to her his bald, scented, and shining head, and complacently seated himself on the sofa.
+
+"First of all, dear friend, tell me how you are. Set your friend's mind at rest," said he without altering his tone, beneath the politeness and affected sympathy of which indifference and even irony could be discerned.
+
+"Can one be well while suffering morally? Can one be calm in times like these if one has any feeling?" said Anna Pavlovna. "You are staying the whole evening, I hope?"
+
+"And the fete at the English ambassador's? Today is Wednesday. I must put in an appearance there," said the prince. "My daughter is coming for me to take me there."
+
+"I thought today's fete had been canceled. I confess all these festivities and fireworks are becoming wearisome."
+
+"If they had known that you wished it, the entertainment would have been put off," said the prince, who, like a wound-up clock, by force of habit said things he did not even wish to be believed.
+
+"Don't tease! Well, and what has been decided about Novosiltsev's dispatch? You know everything."
+
+"What can one say about it?" replied the prince in a cold, listless tone. "What has been decided? They have decided that Buonaparte has burnt his boats, and I believe that we are ready to burn ours."
+
+Prince Vasili always spoke languidly, like an actor repeating a stale part. Anna Pavlovna Scherer on the contrary, despite her forty years, overflowed with animation and impulsiveness. To be an enthusiast had become her social vocation and, sometimes even when she did not feel like it, she became enthusiastic in order not to disappoint the expectations of those who knew her. The subdued smile which, though it did not suit her faded features, always played round her lips expressed, as in a spoiled child, a continual consciousness of her charming defect, which she neither wished, nor could, nor considered it necessary, to correct.
+
+In the midst of a conversation on political matters Anna Pavlovna burst out:
+
+"Oh, don't speak to me of Austria. Perhaps I don't understand things, but Austria never has wished, and does not wish, for war. She is betraying us! Russia alone must save Europe. Our gracious sovereign recognizes his high vocation and will be true to it. That is the one thing I have faith in! Our good and wonderful sovereign has to perform the noblest role on earth, and he is so virtuous and noble that God will not forsake him. He will fulfill his vocation and crush the hydra of revolution, which has become more terrible than ever in the person of this murderer and villain! We alone must avenge the blood of the just one.... Whom, I ask you, can we rely on?... England with her commercial spirit will not and cannot understand the Emperor Alexander's loftiness of soul. She has refused to evacuate Malta. She wanted to find, and still seeks, some secret motive in our actions. What answer did Novosiltsev get? None. The English have not understood and cannot understand the self-abnegation of our Emperor who wants nothing for himself, but only desires the good of mankind. And what have they promised? Nothing! And what little they have promised they will not perform! Prussia has always declared that Buonaparte is invincible, and that all Europe is powerless before him.... And I don't believe a word that Hardenburg says, or Haugwitz either. This famous Prussian neutrality is just a trap. I have faith only in God and the lofty destiny of our adored monarch. He will save Europe!"
+
+She suddenly paused, smiling at her own impetuosity.
+
+"I think," said the prince with a smile, "that if you had been sent instead of our dear Wintzingerode you would have captured the King of Prussia's consent by assault. You are so eloquent. Will you give me a cup of tea?"
+
+"In a moment. A propos," she added, becoming calm again, "I am expecting two very interesting men tonight, le Vicomte de Mortemart, who is connected with the Montmorencys through the Rohans, one of the best French families. He is one of the genuine emigres, the good ones. And also the Abbe Morio. Do you know that profound thinker? He has been received by the Emperor. Had you heard?"
+
+"I shall be delighted to meet them," said the prince. "But tell me," he added with studied carelessness as if it had only just occurred to him, though the question he was about to ask was the chief motive of his visit, "is it true that the Dowager Empress wants Baron Funke to be appointed first secretary at Vienna? The baron by all accounts is a poor creature."
+
+Prince Vasili wished to obtain this post for his son, but others were trying through the Dowager Empress Marya Fedorovna to secure it for the baron.
+
+Anna Pavlovna almost closed her eyes to indicate that neither she nor anyone else had a right to criticize what the Empress desired or was pleased with.
+
+"Baron Funke has been recommended to the Dowager Empress by her sister," was all she said, in a dry and mournful tone.
+
+As she named the Empress, Anna Pavlovna's face suddenly assumed an expression of profound and sincere devotion and respect mingled with sadness, and this occurred every time she mentioned her illustrious patroness. She added that Her Majesty had deigned to show Baron Funke beaucoup d'estime, and again her face clouded over with sadness.
+
+The prince was silent and looked indifferent. But, with the womanly and courtierlike quickness and tact habitual to her, Anna Pavlovna wished both to rebuke him (for daring to speak as he had done of a man recommended to the Empress) and at the same time to console him, so she said:
+
+"Now about your family. Do you know that since your daughter came out everyone has been enraptured by her? They say she is amazingly beautiful."
+
+The prince bowed to signify his respect and gratitude.
+
+"I often think," she continued after a short pause, drawing nearer to the prince and smiling amiably at him as if to show that political and social topics were ended and the time had come for intimate conversation - "I often think how unfairly sometimes the joys of life are distributed. Why has fate given you two such splendid children? I don't speak of Anatole, your youngest. I don't like him," she added in a tone admitting of no rejoinder and raising her eyebrows. "Two such charming children. And really you appreciate them less than anyone, and so you don't deserve to have them."
+
+And she smiled her ecstatic smile.
+
+"I can't help it," said the prince. "Lavater would have said I lack the bump of paternity."
+
+"Don't joke; I mean to have a serious talk with you. Do you know I am dissatisfied with your younger son? Between ourselves" (and her face assumed its melancholy expression), "he was mentioned at Her Majesty's and you were pitied...."
+
+The prince answered nothing, but she looked at him significantly, awaiting a reply. He frowned.
+
+"What would you have me do?" he said at last. "You know I did all a father could for their education, and they have both turned out fools. Hippolyte is at least a quiet fool, but Anatole is an active one. That is the only difference between them." He said this smiling in a way more natural and animated than usual, so that the wrinkles round his mouth very clearly revealed something unexpectedly coarse and unpleasant.
+
+"And why are children born to such men as you? If you were not a father there would be nothing I could reproach you with," said Anna Pavlovna, looking up pensively.
+
+"I am your faithful slave and to you alone I can confess that my children are the bane of my life. It is the cross I have to bear. That is how I explain it to myself. It can't be helped!"
+
+He said no more, but expressed his resignation to cruel fate by a gesture. Anna Pavlovna meditated.
+
+"Have you never thought of marrying your prodigal son Anatole?" she asked. "They say old maids have a mania for matchmaking, and though I don't feel that weakness in myself as yet, I know a little person who is very unhappy with her father. She is a relation of yours, Princess Mary Bolkonskaya."
+
+Prince Vasili did not reply, though, with the quickness of memory and perception befitting a man of the world, he indicated by a movement of the head that he was considering this information.
+
+"Do you know," he said at last, evidently unable to check the sad current of his thoughts, "that Anatole is costing me forty thousand rubles a year? And," he went on after a pause, "what will it be in five years, if he goes on like this?" Presently he added: "That's what we fathers have to put up with.... Is this princess of yours rich?"
+
+"Her father is very rich and stingy. He lives in the country. He is the well-known Prince Bolkonski who had to retire from the army under the late Emperor, and was nicknamed 'the King of Prussia.' He is very clever but eccentric, and a bore. The poor girl is very unhappy. She has a brother; I think you know him, he married Lise Meinen lately. He is an aide-de-camp of Kutuzov's and will be here tonight."
+
+"Listen, dear Annette," said the prince, suddenly taking Anna Pavlovna's hand and for some reason drawing it downwards. "Arrange that affair for me and I shall always be your most devoted slave-slafe with an f, as a village elder of mine writes in his reports. She is rich and of good family and that's all I want."
+
+And with the familiarity and easy grace peculiar to him, he raised the maid of honor's hand to his lips, kissed it, and swung it to and fro as he lay back in his armchair, looking in another direction.
+
+"Attendez," said Anna Pavlovna, reflecting, "I'll speak to Lise, young Bolkonski's wife, this very evening, and perhaps the thing can be arranged. It shall be on your family's behalf that I'll start my apprenticeship as old maid."
+
+Anna Pavlovna's drawing room was gradually filling. The highest Petersburg society was assembled there: people differing widely in age and character but alike in the social circle to which they belonged. Prince Vasili's daughter, the beautiful Helene, came to take her father to the ambassador's entertainment; she wore a ball dress and her badge as maid of honor. The youthful little Princess Bolkonskaya, known as la femme la plus seduisante de Petersbourg, * was also there. She had been married during the previous winter, and being pregnant did not go to any large gatherings, but only to small receptions. Prince Vasili's son, Hippolyte, had come with Mortemart, whom he introduced. The Abbe Morio and many others had also come.
+
+* The most fascinating woman in Petersburg.
+
+To each new arrival Anna Pavlovna said, "You have not yet seen my aunt," or "You do not know my aunt?" and very gravely conducted him or her to a little old lady, wearing large bows of ribbon in her cap, who had come sailing in from another room as soon as the guests began to arrive; and slowly turning her eyes from the visitor to her aunt, Anna Pavlovna mentioned each one's name and then left them.
+
+Each visitor performed the ceremony of greeting this old aunt whom not one of them knew, not one of them wanted to know, and not one of them cared about; Anna Pavlovna observed these greetings with mournful and solemn interest and silent approval. The aunt spoke to each of them in the same words, about their health and her own, and the health of Her Majesty, "who, thank God, was better today." And each visitor, though politeness prevented his showing impatience, left the old woman with a sense of relief at having performed a vexatious duty and did not return to her the whole evening.
+
+The young Princess Bolkonskaya had brought some work in a gold-embroidered velvet bag. Her pretty little upper lip, on which a delicate dark down was just perceptible, was too short for her teeth, but it lifted all the more sweetly, and was especially charming when she occasionally drew it down to meet the lower lip. As is always the case with a thoroughly attractive woman, her defect - the shortness of her upper lip and her half-open mouth - seemed to be her own special and peculiar form of beauty. Everyone brightened at the sight of this pretty young woman, so soon to become a mother, so full of life and health, and carrying her burden so lightly. Old men and dull dispirited young ones who looked at her, after being in her company and talking to her a little while, felt as if they too were becoming, like her, full of life and health. All who talked to her, and at each word saw her bright smile and the constant gleam of her white teeth, thought that they were in a specially amiable mood that day.
+
+The little princess went round the table with quick, short, swaying steps, her workbag on her arm, and gaily spreading out her dress sat down on a sofa near the silver samovar, as if all she was doing was a pleasure to herself and to all around her. "I have brought my work," said she in French, displaying her bag and addressing all present. "Mind, Annette, I hope you have not played a wicked trick on me," she added, turning to her hostess. "You wrote that it was to be quite a small reception, and just see how badly I am dressed." And she spread out her arms to show her short-waisted, lace-trimmed, dainty gray dress, girdled with a broad ribbon just below the breast.
+
+"Soyez tranquille, Lise, you will always be prettier than anyone else," replied Anna Pavlovna.
+
+"You know," said the princess in the same tone of voice and still in French, turning to a general, "my husband is deserting me? He is going to get himself killed. Tell me what this wretched war is for?" she added, addressing Prince Vasili, and without waiting for an answer she turned to speak to his daughter, the beautiful Helene.
+
+"What a delightful woman this little princess is!" said Prince Vasili to Anna Pavlovna.
+
+One of the next arrivals was a stout, heavily built young man with close-cropped hair, spectacles, the light-colored breeches fashionable at that time, a very high ruffle, and a brown dress coat. This stout young man was an illegitimate son of Count Bezukhov, a well-known grandee of Catherine's time who now lay dying in Moscow. The young man had not yet entered either the military or civil service, as he had only just returned from abroad where he had been educated, and this was his first appearance in society. Anna Pavlovna greeted him with the nod she accorded to the lowest hierarchy in her drawing room. But in spite of this lowest-grade greeting, a look of anxiety and fear, as at the sight of something too large and unsuited to the place, came over her face when she saw Pierre enter. Though he was certainly rather bigger than the other men in the room, her anxiety could only have reference to the clever though shy, but observant and natural, expression which distinguished him from everyone else in that drawing room.
+
+"It is very good of you, Monsieur Pierre, to come and visit a poor invalid," said Anna Pavlovna, exchanging an alarmed glance with her aunt as she conducted him to her.
+
+Pierre murmured something unintelligible, and continued to look round as if in search of something. On his way to the aunt he bowed to the little princess with a pleased smile, as to an intimate acquaintance.
+
+Anna Pavlovna's alarm was justified, for Pierre turned away from the aunt without waiting to hear her speech about Her Majesty's health. Anna Pavlovna in dismay detained him with the words: "Do you know the Abbe Morio? He is a most interesting man."
+
+"Yes, I have heard of his scheme for perpetual peace, and it is very interesting but hardly feasible."
+
+"You think so?" rejoined Anna Pavlovna in order to say something and get away to attend to her duties as hostess. But Pierre now committed a reverse act of impoliteness. First he had left a lady before she had finished speaking to him, and now he continued to speak to another who wished to get away. With his head bent, and his big feet spread apart, he began explaining his reasons for thinking the abbe's plan chimerical.
+
+"We will talk of it later," said Anna Pavlovna with a smile.
+
+And having got rid of this young man who did not know how to behave, she resumed her duties as hostess and continued to listen and watch, ready to help at any point where the conversation might happen to flag. As the foreman of a spinning mill, when he has set the hands to work, goes round and notices here a spindle that has stopped or there one that creaks or makes more noise than it should, and hastens to check the machine or set it in proper motion, so Anna Pavlovna moved about her drawing room, approaching now a silent, now a too-noisy group, and by a word or slight rearrangement kept the conversational machine in steady, proper, and regular motion. But amid these cares her anxiety about Pierre was evident. She kept an anxious watch on him when he approached the group round Mortemart to listen to what was being said there, and again when he passed to another group whose center was the abbe.
+
+Pierre had been educated abroad, and this reception at Anna Pavlovna's was the first he had attended in Russia. He knew that all the intellectual lights of Petersburg were gathered there and, like a child in a toyshop, did not know which way to look, afraid of missing any clever conversation that was to be heard. Seeing the self-confident and refined expression on the faces of those present he was always expecting to hear something very profound. At last he came up to Morio. Here the conversation seemed interesting and he stood waiting for an opportunity to express his own views, as young people are fond of doing.
+
+CHAPTER III
+Anna Pavlovna's reception was in full swing. The spindles hummed steadily and ceaselessly on all sides. With the exception of the aunt, beside whom sat only one elderly lady, who with her thin careworn face was rather out of place in this brilliant society, the whole company had settled into three groups. One, chiefly masculine, had formed round the abbe. Another, of young people, was grouped round the beautiful Princess Helene, Prince Vasili's daughter, and the little Princess Bolkonskaya, very pretty and rosy, though rather too plump for her age. The third group was gathered round Mortemart and Anna Pavlovna.
+
+The vicomte was a nice-looking young man with soft features and polished manners, who evidently considered himself a celebrity but out of politeness modestly placed himself at the disposal of the circle in which he found himself. Anna Pavlovna was obviously serving him up as a treat to her guests. As a clever maitre d'hotel serves up as a specially choice delicacy a piece of meat that no one who had seen it in the kitchen would have cared to eat, so Anna Pavlovna served up to her guests, first the vicomte and then the abbe, as peculiarly choice morsels. The group about Mortemart immediately began discussing the murder of the Duc d'Enghien. The vicomte said that the Duc d'Enghien had perished by his own magnanimity, and that there were particular reasons for Buonaparte's hatred of him.
+
+"Ah, yes! Do tell us all about it, Vicomte," said Anna Pavlovna, with a pleasant feeling that there was something A la Louis XV in the sound of that sentence: "Contez nous cela, Vicomte."
+
+The vicomte bowed and smiled courteously in token of his willingness to comply. Anna Pavlovna arranged a group round him, inviting everyone to listen to his tale.
+
+"The vicomte knew the duc personally," whispered Anna Pavlovna to one of the guests. "The vicomte is a wonderful raconteur," said she to another. "How evidently he belongs to the best society," said she to a third; and the vicomte was served up to the company in the choicest and most advantageous style, like a well-garnished joint of roast beef on a hot dish.
+
+The vicomte wished to begin his story and gave a subtle smile.
+
+"Come over here, Helene, dear," said Anna Pavlovna to the beautiful young princess who was sitting some way off, the center of another group.
+
+The princess smiled. She rose with the same unchanging smile with which she had first entered the room - the smile of a perfectly beautiful woman. With a slight rustle of her white dress trimmed with moss and ivy, with a gleam of white shoulders, glossy hair, and sparkling diamonds, she passed between the men who made way for her, not looking at any of them but smiling on all, as if graciously allowing each the privilege of admiring her beautiful figure and shapely shoulders, back, and bosom - which in the fashion of those days were very much exposed - and she seemed to bring the glamour of a ballroom with her as she moved toward Anna Pavlovna. Helene was so lovely that not only did she not show any trace of coquetry, but on the contrary she even appeared shy of her unquestionable and all too victorious beauty. She seemed to wish, but to be unable, to diminish its effect.
+
+"How lovely!" said everyone who saw her; and the vicomte lifted his shoulders and dropped his eyes as if startled by something extraordinary when she took her seat opposite and beamed upon him also with her unchanging smile.
+
+"Madame, I doubt my ability before such an audience," said he, smilingly inclining his head.
+
+The princess rested her bare round arm on a little table and considered a reply unnecessary. She smilingly waited. All the time the story was being told she sat upright, glancing now at her beautiful round arm, altered in shape by its pressure on the table, now at her still more beautiful bosom, on which she readjusted a diamond necklace. From time to time she smoothed the folds of her dress, and whenever the story produced an effect she glanced at Anna Pavlovna, at once adopted just the expression she saw on the maid of honor's face, and again relapsed into her radiant smile.
+
+The little princess had also left the tea table and followed Helene.
+
+"Wait a moment, I'll get my work.... Now then, what are you thinking of?" she went on, turning to Prince Hippolyte. "Fetch me my workbag."
+
+There was a general movement as the princess, smiling and talking merrily to everyone at once, sat down and gaily arranged herself in her seat.
+
+"Now I am all right," she said, and asking the vicomte to begin, she took up her work.
+
+Prince Hippolyte, having brought the workbag, joined the circle and moving a chair close to hers seated himself beside her.
+
+Le charmant Hippolyte was surprising by his extraordinary resemblance to his beautiful sister, but yet more by the fact that in spite of this resemblance he was exceedingly ugly. His features were like his sister's, but while in her case everything was lit up by a joyous, self-satisfied, youthful, and constant smile of animation, and by the wonderful classic beauty of her figure, his face on the contrary was dulled by imbecility and a constant expression of sullen self-confidence, while his body was thin and weak. His eyes, nose, and mouth all seemed puckered into a vacant, wearied grimace, and his arms and legs always fell into unnatural positions.
+
+"It's not going to be a ghost story?" said he, sitting down beside the princess and hastily adjusting his lorgnette, as if without this instrument he could not begin to speak.
+
+"Why no, my dear fellow," said the astonished narrator, shrugging his shoulders.
+
+"Because I hate ghost stories," said Prince Hippolyte in a tone which showed that he only understood the meaning of his words after he had uttered them.
+
+He spoke with such self-confidence that his hearers could not be sure whether what he said was very witty or very stupid. He was dressed in a dark-green dress coat, knee breeches of the color of cuisse de nymphe effrayee, as he called it, shoes, and silk stockings.
+
+The vicomte told his tale very neatly. It was an anecdote, then current, to the effect that the Duc d'Enghien had gone secretly to Paris to visit Mademoiselle George; that at her house he came upon Bonaparte, who also enjoyed the famous actress' favors, and that in his presence Napoleon happened to fall into one of the fainting fits to which he was subject, and was thus at the duc's mercy. The latter spared him, and this magnanimity Bonaparte subsequently repaid by death.
+
+The story was very pretty and interesting, especially at the point where the rivals suddenly recognized one another; and the ladies looked agitated.
+
+"Charming!" said Anna Pavlovna with an inquiring glance at the little princess.
+
+"Charming!" whispered the little princess, sticking the needle into her work as if to testify that the interest and fascination of the story prevented her from going on with it.
+
+The vicomte appreciated this silent praise and smiling gratefully prepared to continue, but just then Anna Pavlovna, who had kept a watchful eye on the young man who so alarmed her, noticed that he was talking too loudly and vehemently with the abbe, so she hurried to the rescue. Pierre had managed to start a conversation with the abbe about the balance of power, and the latter, evidently interested by the young man's simple-minded eagerness, was explaining his pet theory. Both were talking and listening too eagerly and too naturally, which was why Anna Pavlovna disapproved.
+
+"The means are ... the balance of power in Europe and the rights of the people," the abbe was saying. "It is only necessary for one powerful nation like Russia - barbaric as she is said to be - to place herself disinterestedly at the head of an alliance having for its object the maintenance of the balance of power of Europe, and it would save the world!"
+
+"But how are you to get that balance?" Pierre was beginning.
+
+At that moment Anna Pavlovna came up and, looking severely at Pierre, asked the Italian how he stood Russian climate. The Italian's face instantly changed and assumed an offensively affected, sugary expression, evidently habitual to him when conversing with women.
+
+"I am so enchanted by the brilliancy of the wit and culture of the society, more especially of the feminine society, in which I have had the honor of being received, that I have not yet had time to think of the climate," said he.
+
+Not letting the abbe and Pierre escape, Anna Pavlovna, the more conveniently to keep them under observation, brought them into the larger circle.
+
--- a/examples/python-json-datagenerator/readme.md
+++ b/examples/python-json-datagenerator/readme.md
@@ -4,32 +4,6 @@

 There are two python scripts in this example. `randomaddresses.py` generates random addresses from different countries. `predefinedschema.py` sets a template for the model to fill in.

-## Running the Example
-
-1. Ensure you have the `llama2` model installed:
-
-   ```bash
-   ollama pull llama2
-   ```
-
-2. Install the Python Requirements.
-
-   ```bash
-   pip install -r requirements.txt
-   ```
-
-3. Run the Random Addresses example:
-
-   ```bash
-   python randomaddresses.py
-   ```
-
-4. Run the Predefined Schema example:
-
-   ```bash
-   python predefinedschema.py
-   ```
-
 ## Review the Code

 Both programs are basically the same, with a different prompt for each, demonstrating two different ideas. The key part of getting JSON out of a model is to state in the prompt or system prompt that it should respond using JSON, and specifying the `format` as `json` in the data body.
--- a/examples/python-loganalysis/loganalysis.py
+++ b/examples/python-loganalysis/loganalysis.py
@@ -16,12 +16,12 @@ def find_errors_in_log_file():
  with open(log_file_path, 'r') as log_file:
    log_lines = log_file.readlines()

-  error_logs = []
-  for i, line in enumerate(log_lines):
-      if "error" in line.lower():
-          start_index = max(0, i - prelines)
-          end_index = min(len(log_lines), i + postlines + 1)
-          error_logs.extend(log_lines[start_index:end_index])
+error_logs = []
+    for i, line in enumerate(log_lines):
+        if "error" in line.lower():
+            start_index = max(0, i - prelines)
+            end_index = min(len(log_lines), i + postlines + 1)
+            error_logs.extend(log_lines[start_index:end_index])

  return error_logs

@@ -32,6 +32,7 @@ data = {
  "model": "mattw/loganalyzer"
 }

+
 response = requests.post("http://localhost:11434/api/generate", json=data, stream=True)
 for line in response.iter_lines():
  if line:
--- a/examples/python-loganalysis/readme.md
+++ b/examples/python-loganalysis/readme.md
@@ -2,34 +2,12 @@

 ![loganalyzer 2023-11-10 08_53_29](https://github.com/jmorganca/ollama/assets/633681/ad30f1fc-321f-4953-8914-e30e24db9921)

-This example shows one possible way to create a log file analyzer. It uses the model **mattw/loganalyzer** which is based on **codebooga**, a 34b parameter model.
-
-To use it, run:
+This example shows one possible way to create a log file analyzer. To use it, run:

 `python loganalysis.py <logfile>`

 You can try this with the `logtest.logfile` file included in this directory.

-## Running the Example
-
-1. Ensure you have the `mattw/loganalyzer` model installed:
-
-   ```bash
-   ollama pull mattw/loganalyzer
-   ```
-
-2. Install the Python Requirements.
-
-   ```bash
-   pip install -r requirements.txt
-   ```
-
-3. Run the example:
-
-   ```bash
-   python loganalysis.py logtest.logfile
-   ```
-
 ## Review the code

 The first part of this example is a Modelfile that takes `codebooga` and applies a new System Prompt:
@@ -67,4 +45,4 @@ for line in response.iter_lines():

 There is a lot more that can be done here. This is a simple way to detect errors, looking for the word error. Perhaps it would be interesting to find anomalous activity in the logs. It could be interesting to create embeddings for each line and compare them, looking for similar lines. Or look into applying Levenshtein Distance algorithms to find similar lines to help identify the anomalous lines.

-Try different models and different prompts to analyze the data. You could consider adding retrieval augmented generation (RAG) to this to help understand newer log formats.
+Also try different models and different prompts to analyze the data. You could consider adding retrieval augmented generation (RAG) to this to help understand newer log formats.
--- a/examples/python-rag-newssummary/README.md
+++ b/examples/python-rag-newssummary/README.md
@@ -14,22 +14,9 @@ This example goes through a series of steps:

 This example lets you pick from a few different topic areas, then summarize the most recent x articles for that topic. It then creates chunks of sentences from each article and then generates embeddings for each of those chunks.

-## Running the Example
+You can run the example like this:

-1. Ensure you have the `mistral-openorca` model installed:
-
-   ```bash
-   ollama pull mistral-openorca
-   ```
-
-2. Install the Python Requirements.
-
-   ```bash
-   pip install -r requirements.txt
-   ```
-
-3. Run the example:
-
-   ```bash
-   python summ.py
-   ```
+```bash
+pip install -r requirements.txt
+python summ.py
+```
--- a/examples/python-simplechat/client.py
+++ b/examples/python-simplechat/client.py
@@ -1,47 +0,0 @@
-import json
-import requests
-
-# NOTE: ollama must be running for this to work, start the ollama app or run `ollama serve`
-model = "llama2"  # TODO: update this for whatever model you wish to use
-
-
-def chat(messages):
-    r = requests.post(
-        "http://0.0.0.0:11434/api/chat",
-        json={"model": model, "messages": messages, "stream": True},
-    )
-    r.raise_for_status()
-    output = ""
-
-    for line in r.iter_lines():
-        body = json.loads(line)
-        if "error" in body:
-            raise Exception(body["error"])
-        if body.get("done") is False:
-            message = body.get("message", "")
-            content = message.get("content", "")
-            output += content
-            # the response streams one token at a time, print that as we receive it
-            print(content, end="", flush=True)
-
-        if body.get("done", False):
-            message["content"] = output
-            return message
-
-
-def main():
-    messages = []
-
-    while True:
-        user_input = input("Enter a prompt: ")
-        if not user_input:
-            exit()
-        print()
-        messages.append({"role": "user", "content": user_input})
-        message = chat(messages)
-        messages.append(message)
-        print("\n\n")
-
-
-if __name__ == "__main__":
-    main()
--- a/examples/python-simplechat/readme.md
+++ b/examples/python-simplechat/readme.md
@@ -1,44 +0,0 @@
-# Simple Chat Example
-
-The **chat** endpoint is one of two ways to generate text from an LLM with Ollama, and is introduced in version 0.1.14. At a high level, you provide the endpoint an array of objects with a role and content specified. Then with each output and prompt, you add more of those role/content objects, which builds up the history.
-
-## Running the Example
-
-1. Ensure you have the `llama2` model installed:
-
-   ```bash
-   ollama pull llama2
-   ```
-
-2. Install the Python Requirements.
-
-   ```bash
-   pip install -r requirements.txt
-   ```
-
-3. Run the example:
-
-   ```bash
-   python client.py
-   ```
-
-## Review the Code
-
-You can see in the **chat** function that actually calling the endpoint is done simply with:
-
-```python
-r = requests.post(
-  "http://0.0.0.0:11434/api/chat",
-  json={"model": model, "messages": messages, "stream": True},
-)
-```
-
-With the **generate** endpoint, you need to provide a `prompt`. But with **chat**, you provide `messages`. And the resulting stream of responses includes a `message` object with a `content` field.
-
-The final JSON object doesn't provide the full content, so you will need to build the content yourself.
-
-In the **main** function, we collect `user_input` and add it as a message to our messages and that is passed to the chat function. When the LLM is done responding the output is added as another message.
-
-## Next Steps
-
-In this example, all generations are kept. You might want to experiment with summarizing everything older than 10 conversations to enable longer history with less context being used.
--- a/examples/python-simplechat/requirements.txt
+++ b/examples/python-simplechat/requirements.txt
@@ -1 +0,0 @@
-Requests==2.31.0
--- a/examples/python-simplegenerate/README.md
+++ b/examples/python-simplegenerate/README.md
@@ -1,29 +0,0 @@
-# Simple Generate Example
-
-This is a simple example using the **Generate** endpoint.
-
-## Running the Example
-
-1. Ensure you have the `stablelm-zephyr` model installed:
-
-   ```bash
-   ollama pull stablelm-zephyr
-   ```
-
-2. Install the Python Requirements.
-
-   ```bash
-   pip install -r requirements.txt
-   ```
-
-3. Run the example:
-
-   ```bash
-   python client.py
-   ```
-
-## Review the Code
-
-The **main** function simply asks for input, then passes that to the generate function. The output from generate is then passed back to generate on the next run.
-
-The **generate** function uses `requests.post` to call `/api/generate`, passing the model, prompt, and context. The `generate` endpoint returns a stream of JSON blobs that are then iterated through, looking for the response values. That is then printed out. The final JSON object includes the full context of the conversation so far, and that is the return value from the function.
--- a/examples/python-simplegenerate/client.py
+++ b/examples/python-simplegenerate/client.py
@@ -2,7 +2,7 @@ import json
 import requests

 # NOTE: ollama must be running for this to work, start the ollama app or run `ollama serve`
-model = 'stablelm-zephyr' # TODO: update this for whatever model you wish to use
+model = 'llama2' # TODO: update this for whatever model you wish to use

 def generate(prompt, context):
    r = requests.post('http://localhost:11434/api/generate',
@@ -30,8 +30,6 @@ def main():
    context = [] # the context stores a conversation history, you can use this to make the model more context aware
    while True:
        user_input = input("Enter a prompt: ")
-        if not user_input:
-            exit()
        print()
        context = generate(user_input, context)
        print()
--- a/examples/python-simplegenerate/requirements.txt
+++ b/examples/python-simplegenerate/requirements.txt
@@ -1 +0,0 @@
-Requests==2.31.0
--- a/examples/typescript-mentors/README.md
+++ b/examples/typescript-mentors/README.md
@@ -4,62 +4,18 @@ This example demonstrates how one would create a set of 'mentors' you can have a

 ## Usage

-1. Add llama2 to have the mentors ask your questions:
+```bash
+ts-node ./character-generator.ts "Lorne Greene"
+```

-   ```bash
-   ollama pull llama2
-   ```
+This will create `lornegreene/Modelfile`. Now you can create a model with this command:

-2. Install prerequisites:
+```bash
+ollama create lornegreene -f lornegreene/Modelfile
+```

-   ```bash
-   npm install
-   ```
+If you want to add your own mentors, you will have to update the code to look at your namespace instead of **mattw**. Also set the list of mentors to include yours.

-3. Ask a question:
-
-   ```bash
-   npm start "what is a jackalope"
-   ```
-
-You can also add your own character to be chosen at random when you ask a question.
-
-1. Make sure you have the right model installed:
-
-   ```bash
-   ollama pull stablebeluga2:70b-q4_K_M
-   ```
-  
-2. Create a new character:
-  
-   ```bash
-   npm run charactergen "Lorne Greene"
-   ```
-
-   You can choose any well-known person you like. This example will create `lornegreene/Modelfile`.
-
-3. Now you can create a model with this command:
-
-   ```bash
-   ollama create <YourNamespace>/lornegreene -f lornegreene/Modelfile
-   ```
-
-   `YourNamespace` is whatever name you set up when you signed up at [https://ollama.ai/signup](https://ollama.ai/signup).
-
-4. To add this to your mentors, you will have to update the code as follows. On line 8 of `mentors.ts`, add an object to the array, replacing `<YourNamespace>` with the namespace you used above.
-
-   ```bash
-   {ns: "<YourNamespace>", char: "Lorne Greene"}
-   ```
-
-## Review the Code
-
-There are two scripts you can run in this example. The first is the main script to ask the mentors a question. The other one lets you generate a character to add to the mentors. Both scripts are mostly about adjusting the prompts at each inference stage.
-
-### mentors.ts
-
-In the **main** function, it starts by generating a list of mentors. This chooses 3 from a list of interesting characters. Then we ask for a question, and then things get interesting. We set the prompt for each of the 3 mentors a little differently. And the 2nd and 3rd mentors see what the previous folks said. The other functions in mentors sets the prompts for each mentor.
-
-### character-generator.ts
-
-**Character Generator** simply customizes the prompt to build a character profile for any famous person. And most of the script is just tweaking the prompt. This uses Stable Beluga 2 70b parameters. The 70b models tend to do better writing a bio about a character than smaller models, and Stable Beluga seemed to do better than Llama 2. Since this is used at development time for the characters, it doesn't affect the runtime of asking the mentors for their input.
+```bash
+ts-node ./mentors.ts "What is a Jackalope?"
+```
--- a/examples/typescript-mentors/mentors.ts
+++ b/examples/typescript-mentors/mentors.ts
@@ -2,11 +2,10 @@ import { Ollama } from 'ollama-node';

 const mentorCount = 3;
 const ollama = new Ollama();
-type Mentor = { ns: string, char: string };

-function getMentors(): Mentor[] {
-  const mentors = [{ ns: 'mattw', char: 'Gary Vaynerchuk' }, { ns: 'mattw', char: 'Kanye West'}, {ns: 'mattw', char: 'Martha Stewart'}, {ns: 'mattw', char: 'Neil deGrasse Tyson'}, {ns: 'mattw', char: 'Owen Wilson'}, {ns: 'mattw', char: 'Ronald Reagan'}, {ns: 'mattw', char: 'Donald Trump'}, {ns: 'mattw', char: 'Barack Obama'}, {ns: 'mattw', char: 'Jeff Bezos'}];
-  const chosenMentors: Mentor[] = [];
+function getMentors(): string[] {
+  const mentors = ['Gary Vaynerchuk', 'Kanye West', 'Martha Stewart', 'Neil deGrasse Tyson', 'Owen Wilson', 'Ronald Reagan', 'Donald Trump', 'Barack Obama', 'Jeff Bezos'];
+  const chosenMentors: string[] = [];
  for (let i = 0; i < mentorCount; i++) {
    const mentor = mentors[Math.floor(Math.random() * mentors.length)];
    chosenMentors.push(mentor);
@@ -15,12 +14,12 @@ function getMentors(): Mentor[] {
  return chosenMentors;
 }

-function getMentorFileName(mentor: Mentor): string {
-  const model = mentor.char.toLowerCase().replace(/\s/g, '');
-  return `${mentor.ns}/${model}`;
+function getMentorFileName(mentor: string): string {
+  const model = mentor.toLowerCase().replace(/\s/g, '');
+  return `mattw/${model}`;
 }

-async function getSystemPrompt(mentor: Mentor, isLast: boolean, question: string): Promise<string> {
+async function getSystemPrompt(mentor: string, isLast: boolean, question: string): Promise<string> {
  ollama.setModel(getMentorFileName(mentor));
  const info = await ollama.showModelInfo()
  let SystemPrompt = info.system || '';
@@ -44,8 +43,8 @@ async function main() {
    ollama.setModel(getMentorFileName(mentor));
    ollama.setSystemPrompt(SystemPrompt);
    let output = '';
-    process.stdout.write(`\n${mentor.char}: `);
-    for await (const chunk of ollama.streamingGenerate(theConversation + `Continue the conversation as if you were ${mentor.char} on the question "${question}".`)) {
+    process.stdout.write(`\n${mentor}: `);
+    for await (const chunk of ollama.streamingGenerate(theConversation + `Continue the conversation as if you were ${mentor} on the question "${question}".`)) {
      if (chunk.response) {
        output += chunk.response;
        process.stdout.write(chunk.response);
@@ -53,7 +52,7 @@ async function main() {
        process.stdout.write('\n');
      }
    }
-    theConversation += `${mentor.char}: ${output}\n\n`
+    theConversation += `${mentor}: ${output}\n\n`
  }
 }

--- a/examples/typescript-mentors/package.json
+++ b/examples/typescript-mentors/package.json
@@ -1,15 +1,7 @@
 {
-  "scripts": {
-    "charactergen": "tsx character-generator.ts", 
-    "start": "tsx mentors.ts"
-  },
  "dependencies": {
    "fs": "^0.0.1-security",
    "ollama-node": "^0.0.3",
    "path": "^0.12.7"
-  },
-  "devDependencies": {
-    "tsx": "^4.6.2",
-    "typescript": "^5.3.3"
  }
 }
--- a/examples/typescript-simplechat/client.ts
+++ b/examples/typescript-simplechat/client.ts
@@ -1,77 +0,0 @@
-import * as readline from "readline";
-
-const model = "llama2";
-type Message = {
-  role: "assistant" | "user" | "system";
-  content: string;
-}
-const messages: Message[] = [{
-  role: "system",
-  content: "You are a helpful AI agent."
-}]
-
-const rl = readline.createInterface({
-  input: process.stdin,
-  output: process.stdout
-})
-
-async function chat(messages: Message[]): Promise<Message> {
-  const body = {
-    model: model,
-    messages: messages
-  }
-
-  const response = await fetch("http://localhost:11434/api/chat", {
-    method: "POST",
-    body: JSON.stringify(body)
-  })
-
-  const reader = response.body?.getReader()
-  if (!reader) {
-    throw new Error("Failed to read response body")
-  }
-  let content = ""
-  while (true) {
-    const { done, value } = await reader.read()
-    if (done) {
-      break;
-    }
-    const rawjson = new TextDecoder().decode(value);
-    const json = JSON.parse(rawjson)
-
-    if (json.done === false) {
-      process.stdout.write(json.message.content);
-      content += json.message.content
-    }
-
-  }
-  return { role: "assistant", content: content };
-}
-
-async function askQuestion(): Promise<void> {
-  return new Promise<void>((resolve) => {
-    rl.question("\n\nAsk a question: (press enter alone to quit)\n\n", async (user_input) => {
-      if (user_input.trim() === "") {
-        rl.close();
-        console.log("Thankyou. Goodbye.\n")
-        console.log("=======\nHere is the message history that was used in this conversation.\n=======\n")
-        messages.forEach(message => {
-          console.log(message)
-        })
-        resolve();
-      } else {
-        console.log();
-        messages.push({ role: "user", content: user_input });
-        messages.push(await chat(messages));
-        await askQuestion(); // Ask the next question
-      }
-    });
-  });
-}
-
-async function main() {
-  await askQuestion();
-
-}
-
-main();
--- a/examples/typescript-simplechat/package.json
+++ b/examples/typescript-simplechat/package.json
@@ -1,12 +0,0 @@
-{ 
-  "scripts": {
-    "start": "tsx client.ts"
-  }, 
-  "dependencies": {
-     "@types/node": "^20.10.4", 
-     "prompt-sync": "^4.2.0", 
-     "readline": "^1.3.0", 
-     "tsx": "^4.6.2", 
-     "typescript": "^5.3.3" 
-     } 
-    }
--- a/examples/typescript-simplechat/readme.md
+++ b/examples/typescript-simplechat/readme.md
@@ -1,35 +0,0 @@
-# Simple Chat Example
-
-The **chat** endpoint, available as of v0.1.14, is one of two ways to generate text from an LLM with Ollama. At a high level, you provide the endpoint an array of message objects with a role and content specified. Then with each output and prompt, you add more messages, which builds up the history.
-
-## Run the Example
-
-`npm start`
-
-## Review the Code
-
-You can see in the **chat** function that is actually calling the endpoint is simply done with:
-
-```typescript
-const body = {
-  model: model,
-  messages: messages
-}
-
-const response = await fetch("http://localhost:11434/api/chat", {
-  method: "POST",
-  body: JSON.stringify(body)
-})
-```
-
-With the **generate** endpoint, you need to provide a `prompt`. But with **chat**, you provide `messages`. And the resulting stream of responses includes a `message` object with a `content` field.
-
-The final JSON object doesn't provide the full content, so you will need to build the content yourself. In this example, **chat** takes the full array of messages and outputs the resulting message from this call of the chat endpoint.
-
-In the **askQuestion** function, we collect `user_input` and add it as a message to our messages, and that is passed to the chat function. When the LLM is done responding, the output is added as another message to the messages array.
-
-At the end, you will see a printout of all the messages.
-
-## Next Steps
-
-In this example, all generations are kept. You might want to experiment with summarizing everything older than 10 conversations to enable longer history with less context being used.
--- a/go.mod
+++ b/go.mod
@@ -5,18 +5,14 @@ go 1.20
 require (
 	github.com/emirpasic/gods v1.18.1
 	github.com/gin-gonic/gin v1.9.1
+	github.com/mattn/go-runewidth v0.0.14
+	github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db
 	github.com/olekukonko/tablewriter v0.0.5
 	github.com/spf13/cobra v1.7.0
-	github.com/stretchr/testify v1.8.4
 	golang.org/x/sync v0.3.0
 )

-require (
-	github.com/davecgh/go-spew v1.1.1 // indirect
-	github.com/mattn/go-runewidth v0.0.14 // indirect
-	github.com/pmezard/go-difflib v1.0.0 // indirect
-	github.com/rivo/uniseg v0.2.0 // indirect
-)
+require github.com/rivo/uniseg v0.2.0 // indirect

 require (
 	github.com/bytedance/sonic v1.9.1 // indirect
--- a/go.sum
+++ b/go.sum
@@ -63,6 +63,8 @@ github.com/mattn/go-isatty v0.0.19/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D
 github.com/mattn/go-runewidth v0.0.9/go.mod h1:H031xJmbD/WCDINGzjvQ9THkh0rPKHF+m2gUSrubnMI=
 github.com/mattn/go-runewidth v0.0.14 h1:+xnbZSEeDbOIg5/mE6JF0w6n9duR1l3/WmbinWVwUuU=
 github.com/mattn/go-runewidth v0.0.14/go.mod h1:Jdepj2loyihRzMpdS35Xk/zdY8IAYHsh153qUoGf23w=
+github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db h1:62I3jR2EmQ4l5rM/4FEfDWcRD+abF5XlKShorW5LRoQ=
+github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db/go.mod h1:l0dey0ia/Uv7NcFFVbCLtqEBQbrT4OCwCSKTEv6enCw=
 github.com/modern-go/concurrent v0.0.0-20180228061459-e0a39a4cb421/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
 github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd h1:TRLaZ9cD/w8PVh93nsPXa1VrQ6jlwL5oN8l14QlcNfg=
 github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
@@ -98,9 +100,8 @@ github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/
 github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
 github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
 github.com/stretchr/testify v1.8.2/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
+github.com/stretchr/testify v1.8.3 h1:RP3t2pwF7cMEbC1dqtB6poj3niw/9gnV4Cjg5oW5gtY=
 github.com/stretchr/testify v1.8.3/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
-github.com/stretchr/testify v1.8.4 h1:CcVxjf3Q8PM0mHUKJCdn+eZZtm5yQwehR5yeSVQQcUk=
-github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
 github.com/twitchyliquid64/golang-asm v0.15.1 h1:SU5vSMR7hnwNxj24w34ZyCi/FmDZTkS4MhqMhdFk5YI=
 github.com/twitchyliquid64/golang-asm v0.15.1/go.mod h1:a1lVb/DtPvCB8fslRZhAngC2+aY1QWCk3Cedj/Gdt08=
 github.com/ugorji/go v1.2.7/go.mod h1:nF9osbDWLy6bDVv/Rtoh6QgnvNDpmCalQV5urGCCS6M=
--- a/gpu/gpu.go
+++ b/gpu/gpu.go
@@ -1,149 +0,0 @@
-//go:build linux || windows
-
-package gpu
-
-/*
-#cgo linux LDFLAGS: -lrt -lpthread -ldl -lstdc++ -lm
-#cgo windows LDFLAGS: -lpthread
-
-#include "gpu_info.h"
-
-*/
-import "C"
-import (
-	"fmt"
-	"log"
-	"runtime"
-	"sync"
-	"unsafe"
-
-	"github.com/jmorganca/ollama/api"
-)
-
-type handles struct {
-	cuda *C.cuda_handle_t
-	rocm *C.rocm_handle_t
-}
-
-var gpuMutex sync.Mutex
-var gpuHandles *handles = nil
-
-// Note: gpuMutex must already be held
-func initGPUHandles() {
-	// TODO - if the ollama build is CPU only, don't do these checks as they're irrelevant and confusing
-	log.Printf("Detecting GPU type")
-	gpuHandles = &handles{nil, nil}
-	var resp C.cuda_init_resp_t
-	C.cuda_init(&resp)
-	if resp.err != nil {
-		log.Printf("CUDA not detected: %s", C.GoString(resp.err))
-		C.free(unsafe.Pointer(resp.err))
-
-		var resp C.rocm_init_resp_t
-		C.rocm_init(&resp)
-		if resp.err != nil {
-			log.Printf("ROCm not detected: %s", C.GoString(resp.err))
-			C.free(unsafe.Pointer(resp.err))
-		} else {
-			log.Printf("Radeon GPU detected")
-			rocm := resp.rh
-			gpuHandles.rocm = &rocm
-		}
-	} else {
-		log.Printf("Nvidia GPU detected")
-		cuda := resp.ch
-		gpuHandles.cuda = &cuda
-	}
-}
-
-func GetGPUInfo() GpuInfo {
-	// TODO - consider exploring lspci (and equivalent on windows) to check for
-	// GPUs so we can report warnings if we see Nvidia/AMD but fail to load the libraries
-	gpuMutex.Lock()
-	defer gpuMutex.Unlock()
-	if gpuHandles == nil {
-		initGPUHandles()
-	}
-
-	var memInfo C.mem_info_t
-	resp := GpuInfo{}
-	if gpuHandles.cuda != nil {
-		C.cuda_check_vram(*gpuHandles.cuda, &memInfo)
-		if memInfo.err != nil {
-			log.Printf("error looking up CUDA GPU memory: %s", C.GoString(memInfo.err))
-			C.free(unsafe.Pointer(memInfo.err))
-		} else {
-			resp.Library = "cuda"
-		}
-	} else if gpuHandles.rocm != nil {
-		C.rocm_check_vram(*gpuHandles.rocm, &memInfo)
-		if memInfo.err != nil {
-			log.Printf("error looking up ROCm GPU memory: %s", C.GoString(memInfo.err))
-			C.free(unsafe.Pointer(memInfo.err))
-		} else {
-			resp.Library = "rocm"
-		}
-	}
-	if resp.Library == "" {
-		C.cpu_check_ram(&memInfo)
-		// In the future we may offer multiple CPU variants to tune CPU features
-		if runtime.GOOS == "windows" {
-			resp.Library = "cpu"
-		} else {
-			resp.Library = "default"
-		}
-	}
-	if memInfo.err != nil {
-		log.Printf("error looking up CPU memory: %s", C.GoString(memInfo.err))
-		C.free(unsafe.Pointer(memInfo.err))
-		return resp
-	}
-	resp.FreeMemory = uint64(memInfo.free)
-	resp.TotalMemory = uint64(memInfo.total)
-	return resp
-}
-
-func getCPUMem() (memInfo, error) {
-	var ret memInfo
-	var info C.mem_info_t
-	C.cpu_check_ram(&info)
-	if info.err != nil {
-		defer C.free(unsafe.Pointer(info.err))
-		return ret, fmt.Errorf(C.GoString(info.err))
-	}
-	ret.FreeMemory = uint64(info.free)
-	ret.TotalMemory = uint64(info.total)
-	return ret, nil
-}
-
-func CheckVRAM() (int64, error) {
-	gpuInfo := GetGPUInfo()
-	if gpuInfo.FreeMemory > 0 && (gpuInfo.Library == "cuda" || gpuInfo.Library == "rocm") {
-		return int64(gpuInfo.FreeMemory), nil
-	}
-	return 0, fmt.Errorf("no GPU detected") // TODO - better handling of CPU based memory determiniation
-}
-
-func NumGPU(numLayer, fileSizeBytes int64, opts api.Options) int {
-	if opts.NumGPU != -1 {
-		return opts.NumGPU
-	}
-	info := GetGPUInfo()
-	if info.Library == "cpu" || info.Library == "default" {
-		return 0
-	}
-
-	/*
-		Calculate bytes per layer, this will roughly be the size of the model file divided by the number of layers.
-		We can store the model weights and the kv cache in vram,
-		to enable kv chache vram storage add two additional layers to the number of layers retrieved from the model file.
-	*/
-	bytesPerLayer := uint64(fileSizeBytes / numLayer)
-
-	// 75% of the absolute max number of layers we can fit in available VRAM, off-loading too many layers to the GPU can cause OOM errors
-	layers := int(info.FreeMemory/bytesPerLayer) * 3 / 4
-
-	log.Printf("%d MB VRAM available, loading up to %d %s GPU layers out of %d", info.FreeMemory/(1024*1024), layers, info.Library, numLayer)
-
-	return layers
-}
--- a/gpu/gpu_darwin.go
+++ b/gpu/gpu_darwin.go
@@ -1,50 +0,0 @@
-//go:build darwin
-
-package gpu
-
-import "C"
-import (
-	"runtime"
-
-	"github.com/jmorganca/ollama/api"
-)
-
-// CheckVRAM returns the free VRAM in bytes on Linux machines with NVIDIA GPUs
-func CheckVRAM() (int64, error) {
-	// TODO - assume metal, and return free memory?
-	return 0, nil
-
-}
-
-func GetGPUInfo() GpuInfo {
-	// TODO - Metal vs. x86 macs...
-	mem, _ := getCPUMem()
-	return GpuInfo{
-		Library: "default",
-		memInfo: mem,
-	}
-}
-
-func getCPUMem() (memInfo, error) {
-	return memInfo{
-		TotalMemory: 0,
-		FreeMemory:  0,
-	}, nil
-}
-
-func NumGPU(numLayer, fileSizeBytes int64, opts api.Options) int {
-	if opts.NumGPU != -1 {
-		return opts.NumGPU
-	}
-
-	// metal only supported on arm64
-	if runtime.GOARCH == "arm64" {
-		return 1
-	}
-
-	return 0
-}
-
-func nativeInit() error {
-	return nil
-}
--- a/gpu/gpu_info.h
+++ b/gpu/gpu_info.h
@@ -1,50 +0,0 @@
-#ifndef __APPLE__
-#ifndef __GPU_INFO_H__
-#define __GPU_INFO_H__
-#include <stdint.h>
-#include <stdio.h>
-#include <stdlib.h>
-
-#ifndef _WIN32
-#include <dlfcn.h>
-#define LOAD_LIBRARY(lib, flags) dlopen(lib, flags)
-#define LOAD_SYMBOL(handle, sym) dlsym(handle, sym)
-#define LOAD_ERR() strdup(dlerror())
-#define UNLOAD_LIBRARY(handle) dlclose(handle)
-#else
-#include <windows.h>
-#define LOAD_LIBRARY(lib, flags) LoadLibrary(lib)
-#define LOAD_SYMBOL(handle, sym) GetProcAddress(handle, sym)
-#define UNLOAD_LIBRARY(handle) FreeLibrary(handle)
-#define LOAD_ERR() ({\
-  LPSTR messageBuffer = NULL; \
-  size_t size = FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, \
-                                 NULL, GetLastError(), MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), (LPSTR)&messageBuffer, 0, NULL); \
-  char *resp = strdup(messageBuffer); \
-  LocalFree(messageBuffer); \
-  resp; \
-})
-
-#endif
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-typedef struct mem_info {
-  uint64_t total;
-  uint64_t free;
-  char *err;  // If non-nill, caller responsible for freeing
-} mem_info_t;
-
-void cpu_check_ram(mem_info_t *resp);
-
-#ifdef __cplusplus
-}
-#endif
-
-#include "gpu_info_cuda.h"
-#include "gpu_info_rocm.h"
-
-#endif  // __GPU_INFO_H__
-#endif  // __APPLE__
--- a/gpu/gpu_info_cpu.c
+++ b/gpu/gpu_info_cpu.c
@@ -1,43 +0,0 @@
-#include "gpu_info.h"
-// Fallbacks for CPU mode
-
-#ifdef _WIN32
-#include <sysinfoapi.h>
-void cpu_check_ram(mem_info_t *resp) {
-  resp->err = NULL;
-  MEMORYSTATUSEX info;
-  info.dwLength = sizeof(info);
-  if (GlobalMemoryStatusEx(&info) != 0) {
-    resp->total = info.ullTotalPhys;
-    resp->free = info.ullAvailPhys;
-  } else {
-    resp->err = LOAD_ERR();
-  }
-  return;
-}
-
-#elif __linux__
-#include <errno.h>
-#include <string.h>
-#include <sys/sysinfo.h>
-void cpu_check_ram(mem_info_t *resp) {
-  struct sysinfo info;
-  resp->err = NULL;
-  if (sysinfo(&info) != 0) {
-    resp->err = strdup(strerror(errno));
-  } else {
-    resp->total = info.totalram * info.mem_unit;
-    resp->free = info.freeram * info.mem_unit;
-  }
-  return;
-}
-
-#elif __APPLE__
-// TODO consider an Apple implementation that does something useful
-// mem_info_t cpu_check_ram() {
-//   mem_info_t resp = {0, 0, NULL};
-//   return resp;
-// }
-#else
-#error "Unsupported platform"
-#endif
--- a/gpu/gpu_info_cuda.c
+++ b/gpu/gpu_info_cuda.c
@@ -1,125 +0,0 @@
-#ifndef __APPLE__  // TODO - maybe consider nvidia support on intel macs?
-
-#include "gpu_info_cuda.h"
-
-#include <string.h>
-
-#ifndef _WIN32
-const char *cuda_lib_paths[] = {
-    "libnvidia-ml.so",
-    "/usr/local/cuda/lib64/libnvidia-ml.so",
-    "/usr/lib/x86_64-linux-gnu/nvidia/current/libnvidia-ml.so",
-    "/usr/lib/wsl/lib/libnvidia-ml.so.1",  // TODO Maybe glob?
-    NULL,
-};
-#else
-const char *cuda_lib_paths[] = {
-    "nvml.dll",
-    "",
-    NULL,
-};
-#endif
-
-#define CUDA_LOOKUP_SIZE 5
-
-void cuda_init(cuda_init_resp_t *resp) {
-  nvmlReturn_t ret;
-  resp->err = NULL;
-  const int buflen = 256;
-  char buf[buflen + 1];
-  int i;
-
-  struct lookup {
-    char *s;
-    void **p;
-  } l[CUDA_LOOKUP_SIZE] = {
-      {"nvmlInit_v2", (void *)&resp->ch.initFn},
-      {"nvmlShutdown", (void *)&resp->ch.shutdownFn},
-      {"nvmlDeviceGetHandleByIndex", (void *)&resp->ch.getHandle},
-      {"nvmlDeviceGetMemoryInfo", (void *)&resp->ch.getMemInfo},
-      {"nvmlDeviceGetCount_v2", (void *)&resp->ch.getCount},
-  };
-
-  for (i = 0; cuda_lib_paths[i] != NULL && resp->ch.handle == NULL; i++) {
-    resp->ch.handle = LOAD_LIBRARY(cuda_lib_paths[i], RTLD_LAZY);
-  }
-  if (!resp->ch.handle) {
-    // TODO improve error message, as the LOAD_ERR will have typically have the
-    // final path that was checked which might be confusing.
-    char *msg = LOAD_ERR();
-    snprintf(buf, buflen,
-             "Unable to load %s library to query for Nvidia GPUs: %s",
-             cuda_lib_paths[0], msg);
-    free(msg);
-    resp->err = strdup(buf);
-    return;
-  }
-
-  for (i = 0; i < CUDA_LOOKUP_SIZE; i++) {  // TODO - fix this to use a null terminated list
-    *l[i].p = LOAD_SYMBOL(resp->ch.handle, l[i].s);
-    if (!l[i].p) {
-      UNLOAD_LIBRARY(resp->ch.handle);
-      resp->ch.handle = NULL;
-      char *msg = LOAD_ERR();
-      snprintf(buf, buflen, "symbol lookup for %s failed: %s", l[i].s,
-               msg);
-      free(msg);
-      resp->err = strdup(buf);
-      return;
-    }
-  }
-
-  ret = (*resp->ch.initFn)();
-  if (ret != NVML_SUCCESS) {
-    snprintf(buf, buflen, "nvml vram init failure: %d", ret);
-    resp->err = strdup(buf);
-  }
-
-  return;
-}
-
-void cuda_check_vram(cuda_handle_t h, mem_info_t *resp) {
-  resp->err = NULL;
-  nvmlDevice_t device;
-  nvmlMemory_t memInfo = {0};
-  nvmlReturn_t ret;
-  const int buflen = 256;
-  char buf[buflen + 1];
-  int i;
-
-  if (h.handle == NULL) {
-    resp->err = strdup("nvml handle sn't initialized");
-    return;
-  }
-
-  unsigned int devices;
-  ret = (*h.getCount)(&devices);
-  if (ret != NVML_SUCCESS) {
-    snprintf(buf, buflen, "unable to get device count: %d", ret);
-    resp->err = strdup(buf);
-    return;
-  }
-
-  resp->total = 0;
-  resp->free = 0;
-
-  for (i = 0; i < devices; i++) {
-    ret = (*h.getHandle)(i, &device);
-    if (ret != NVML_SUCCESS) {
-      snprintf(buf, buflen, "unable to get device handle %d: %d", i, ret);
-      resp->err = strdup(buf);
-      return;
-    }
-
-    ret = (*h.getMemInfo)(device, &memInfo);
-    if (ret != NVML_SUCCESS) {
-      snprintf(buf, buflen, "device memory info lookup failure %d: %d", i, ret);
-      resp->err = strdup(buf);
-      return;
-    }
-
-    resp->total += memInfo.total;
-    resp->free += memInfo.free;
-  }
-}
-#endif  // __APPLE__
--- a/gpu/gpu_info_cuda.h
+++ b/gpu/gpu_info_cuda.h
@@ -1,36 +0,0 @@
-#ifndef __APPLE__
-#ifndef __GPU_INFO_CUDA_H__
-#define __GPU_INFO_CUDA_H__
-#include "gpu_info.h"
-
-// Just enough typedef's to dlopen/dlsym for memory information
-typedef enum nvmlReturn_enum {
-  NVML_SUCCESS = 0,
-  // Other values omitted for now...
-} nvmlReturn_t;
-typedef void *nvmlDevice_t;  // Opaque is sufficient
-typedef struct nvmlMemory_st {
-  unsigned long long total;
-  unsigned long long free;
-  unsigned long long used;
-} nvmlMemory_t;
-
-typedef struct cuda_handle {
-  void *handle;
-  nvmlReturn_t (*initFn)(void);
-  nvmlReturn_t (*shutdownFn)(void);
-  nvmlReturn_t (*getHandle)(unsigned int, nvmlDevice_t *);
-  nvmlReturn_t (*getMemInfo)(nvmlDevice_t, nvmlMemory_t *);
-  nvmlReturn_t (*getCount)(unsigned int *);
-} cuda_handle_t;
-
-typedef struct cuda_init_resp {
-  char *err;  // If err is non-null handle is invalid
-  cuda_handle_t ch;
-} cuda_init_resp_t;
-
-void cuda_init(cuda_init_resp_t *resp);
-void cuda_check_vram(cuda_handle_t ch, mem_info_t *resp);
-
-#endif  // __GPU_INFO_CUDA_H__
-#endif  // __APPLE__
--- a/gpu/gpu_info_rocm.c
+++ b/gpu/gpu_info_rocm.c
@@ -1,118 +0,0 @@
-#ifndef __APPLE__
-
-#include "gpu_info_rocm.h"
-
-#include <string.h>
-
-#ifndef _WIN32
-const char *rocm_lib_paths[] = {
-    "librocm_smi64.so",
-    "/opt/rocm/lib/librocm_smi64.so",
-    NULL,
-};
-#else
-// TODO untested
-const char *rocm_lib_paths[] = {
-    "rocm_smi64.dll",
-    "/opt/rocm/lib/rocm_smi64.dll",
-    NULL,
-};
-#endif
-
-void rocm_init(rocm_init_resp_t *resp) {
-  rsmi_status_t ret;
-  resp->err = NULL;
-  const int buflen = 256;
-  char buf[buflen + 1];
-  int i;
-  struct lookup {
-    char *s;
-    void **p;
-  } l[4] = {
-      {"rsmi_init", (void *)&resp->rh.initFn},
-      {"rsmi_shut_down", (void *)&resp->rh.shutdownFn},
-      {"rsmi_dev_memory_total_get", (void *)&resp->rh.totalMemFn},
-      {"rsmi_dev_memory_usage_get", (void *)&resp->rh.usageMemFn},
-      // { "rsmi_dev_id_get", (void*)&resp->rh.getHandle },
-  };
-
-  for (i = 0; rocm_lib_paths[i] != NULL && resp->rh.handle == NULL; i++) {
-    resp->rh.handle = LOAD_LIBRARY(rocm_lib_paths[i], RTLD_LAZY);
-  }
-  if (!resp->rh.handle) {
-    char *msg = LOAD_ERR();
-    snprintf(buf, buflen,
-             "Unable to load %s library to query for Radeon GPUs: %s\n",
-             rocm_lib_paths[0], msg);
-    free(msg);
-    resp->err = strdup(buf);
-    return;
-  }
-
-  for (i = 0; i < 4; i++) {
-    *l[i].p = LOAD_SYMBOL(resp->rh.handle, l[i].s);
-    if (!l[i].p) {
-      UNLOAD_LIBRARY(resp->rh.handle);
-      char *msg = LOAD_ERR();
-      snprintf(buf, buflen, "symbol lookup for %s failed: %s", l[i].s,
-               msg);
-      free(msg);
-      resp->err = strdup(buf);
-      return;
-    }
-  }
-
-  ret = (*resp->rh.initFn)(0);
-  if (ret != RSMI_STATUS_SUCCESS) {
-    snprintf(buf, buflen, "rocm vram init failure: %d", ret);
-    resp->err = strdup(buf);
-  }
-
-  return;
-}
-
-void rocm_check_vram(rocm_handle_t h, mem_info_t *resp) {
-  resp->err = NULL;
-  // uint32_t num_devices;
-  // uint16_t device;
-  uint64_t totalMem = 0;
-  uint64_t usedMem = 0;
-  rsmi_status_t ret;
-  const int buflen = 256;
-  char buf[buflen + 1];
-  int i;
-
-  if (h.handle == NULL) {
-    resp->err = strdup("nvml handle sn't initialized");
-    return;
-  }
-
-  // TODO - iterate through devices...  ret =
-  // rsmi_num_monitor_devices(&num_devices);
-
-  // ret = (*h.getHandle)(0, &device);
-  // if (ret != RSMI_STATUS_SUCCESS) {
-  //     printf("rocm vram device lookup failure: %d\n", ret);
-  //     return -1;
-  // }
-
-  // Get total memory - used memory for available memory
-  ret = (*h.totalMemFn)(0, RSMI_MEM_TYPE_VRAM, &totalMem);
-  if (ret != RSMI_STATUS_SUCCESS) {
-    snprintf(buf, buflen, "rocm total mem lookup failure: %d", ret);
-    resp->err = strdup(buf);
-    return;
-  }
-  ret = (*h.usageMemFn)(0, RSMI_MEM_TYPE_VRAM, &usedMem);
-  if (ret != RSMI_STATUS_SUCCESS) {
-    snprintf(buf, buflen, "rocm usage mem lookup failure: %d", ret);
-    resp->err = strdup(buf);
-    return;
-  }
-
-  resp->total = totalMem;
-  resp->free = totalMem - usedMem;
-  return;
-}
-
-#endif  // __APPLE__
--- a/gpu/gpu_info_rocm.h
+++ b/gpu/gpu_info_rocm.h
@@ -1,36 +0,0 @@
-#ifndef __APPLE__
-#ifndef __GPU_INFO_ROCM_H__
-#define __GPU_INFO_ROCM_H__
-#include "gpu_info.h"
-
-// Just enough typedef's to dlopen/dlsym for memory information
-typedef enum rsmi_status_return {
-  RSMI_STATUS_SUCCESS = 0,
-  // Other values omitted for now...
-} rsmi_status_t;
-
-typedef enum rsmi_memory_type {
-  RSMI_MEM_TYPE_VRAM = 0,
-  RSMI_MEM_TYPE_VIS_VRAM,
-  RSMI_MEM_TYPE_GTT,
-} rsmi_memory_type_t;
-
-typedef struct rocm_handle {
-  void *handle;
-  rsmi_status_t (*initFn)(uint64_t);
-  rsmi_status_t (*shutdownFn)(void);
-  rsmi_status_t (*totalMemFn)(uint32_t, rsmi_memory_type_t, uint64_t *);
-  rsmi_status_t (*usageMemFn)(uint32_t, rsmi_memory_type_t, uint64_t *);
-  // rsmi_status_t (*getHandle)(uint32_t, uint16_t *);
-} rocm_handle_t;
-
-typedef struct rocm_init_resp {
-  char *err;  // If err is non-null handle is invalid
-  rocm_handle_t rh;
-} rocm_init_resp_t;
-
-void rocm_init(rocm_init_resp_t *resp);
-void rocm_check_vram(rocm_handle_t rh, mem_info_t *resp);
-
-#endif  // __GPU_INFO_ROCM_H__
-#endif  // __APPLE__
--- a/gpu/gpu_test.go
+++ b/gpu/gpu_test.go
@@ -1,41 +0,0 @@
-package gpu
-
-import (
-	"runtime"
-	"testing"
-
-	"github.com/stretchr/testify/assert"
-)
-
-func TestBasicGetGPUInfo(t *testing.T) {
-	info := GetGPUInfo()
-	assert.Contains(t, "cuda rocm cpu default", info.Library)
-
-	switch runtime.GOOS {
-	case "darwin":
-		// TODO - remove this once MacOS returns some size for CPU
-		return
-	case "linux", "windows":
-		assert.Greater(t, info.TotalMemory, uint64(0))
-		assert.Greater(t, info.FreeMemory, uint64(0))
-	default:
-		return
-	}
-}
-
-func TestCPUMemInfo(t *testing.T) {
-	info, err := getCPUMem()
-	assert.NoError(t, err)
-	switch runtime.GOOS {
-	case "darwin":
-		t.Skip("CPU memory not populated on darwin")
-	case "linux", "windows":
-		assert.Greater(t, info.TotalMemory, uint64(0))
-		assert.Greater(t, info.FreeMemory, uint64(0))
-	default:
-		return
-	}
-
-}
-
-// TODO - add some logic to figure out card type through other means and actually verify we got back what we expected
--- a/gpu/types.go
+++ b/gpu/types.go
@@ -1,14 +0,0 @@
-package gpu
-
-type memInfo struct {
-	TotalMemory uint64 `json:"total_memory,omitempty"`
-	FreeMemory  uint64 `json:"free_memory,omitempty"`
-}
-
-// Beginning of an `ollama info` command
-type GpuInfo struct {
-	memInfo
-	Library string `json:"library,omitempty"`
-
-	// TODO add other useful attributes about the card here for discovery information
-}
--- a/llm/dynamic_shim.c
+++ b/llm/dynamic_shim.c
@@ -1,145 +0,0 @@
-#include "dynamic_shim.h"
-
-#include <stdio.h>
-#include <string.h>
-
-#ifdef __linux__
-#include <dlfcn.h>
-#define LOAD_LIBRARY(lib, flags) dlopen(lib, flags | RTLD_DEEPBIND)
-#define LOAD_SYMBOL(handle, sym) dlsym(handle, sym)
-#define LOAD_ERR() strdup(dlerror())
-#define UNLOAD_LIBRARY(handle) dlclose(handle)
-#elif _WIN32
-#include <windows.h>
-#define LOAD_LIBRARY(lib, flags) LoadLibrary(lib)
-#define LOAD_SYMBOL(handle, sym) GetProcAddress(handle, sym)
-#define UNLOAD_LIBRARY(handle) FreeLibrary(handle)
-inline char *LOAD_ERR() {
-  LPSTR messageBuffer = NULL;
-  size_t size = FormatMessageA(
-      FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM |
-          FORMAT_MESSAGE_IGNORE_INSERTS,
-      NULL, GetLastError(), MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
-      (LPSTR)&messageBuffer, 0, NULL);
-  char *resp = strdup(messageBuffer);
-  LocalFree(messageBuffer);
-  return resp;
-}
-#else
-#include <dlfcn.h>
-#define LOAD_LIBRARY(lib, flags) dlopen(lib, flags)
-#define LOAD_SYMBOL(handle, sym) dlsym(handle, sym)
-#define LOAD_ERR() strdup(dlerror())
-#define UNLOAD_LIBRARY(handle) dlclose(handle)
-#endif
-
-void dynamic_shim_init(const char *libPath, struct dynamic_llama_server *s,
-                       ext_server_resp_t *err) {
-  int i = 0;
-  struct lookup {
-    char *s;
-    void **p;
-  } l[] = {
-      {"llama_server_init", (void *)&s->llama_server_init},
-      {"llama_server_start", (void *)&s->llama_server_start},
-      {"llama_server_stop", (void *)&s->llama_server_stop},
-      {"llama_server_completion", (void *)&s->llama_server_completion},
-      {"llama_server_completion_next_result",
-       (void *)&s->llama_server_completion_next_result},
-      {"llama_server_completion_cancel",
-       (void *)&s->llama_server_completion_cancel},
-      {"llama_server_release_task_result",
-       (void *)&s->llama_server_release_task_result},
-      {"llama_server_tokenize", (void *)&s->llama_server_tokenize},
-      {"llama_server_detokenize", (void *)&s->llama_server_detokenize},
-      {"llama_server_embedding", (void *)&s->llama_server_embedding},
-      {"llama_server_release_json_resp",
-       (void *)&s->llama_server_release_json_resp},
-      {"", NULL},
-  };
-
-  printf("Lazy loading %s library\n", libPath);
-  s->handle = LOAD_LIBRARY(libPath, RTLD_NOW);
-  if (!s->handle) {
-    err->id = -1;
-    char *msg = LOAD_ERR();
-    snprintf(err->msg, err->msg_len,
-             "Unable to load dynamic server library: %s", msg);
-    free(msg);
-    return;
-  }
-
-  for (i = 0; l[i].p != NULL; i++) {
-    *l[i].p = LOAD_SYMBOL(s->handle, l[i].s);
-    if (!l[i].p) {
-      UNLOAD_LIBRARY(s->handle);
-      err->id = -1;
-      char *msg = LOAD_ERR();
-      snprintf(err->msg, err->msg_len, "symbol lookup for %s failed: %s",
-               l[i].s, msg);
-      free(msg);
-      return;
-    }
-  }
-}
-
-inline void dynamic_shim_llama_server_init(struct dynamic_llama_server s,
-                                           ext_server_params_t *sparams,
-                                           ext_server_resp_t *err) {
-  s.llama_server_init(sparams, err);
-}
-
-inline void dynamic_shim_llama_server_start(struct dynamic_llama_server s) {
-  s.llama_server_start();
-}
-
-inline void dynamic_shim_llama_server_stop(struct dynamic_llama_server s) {
-  s.llama_server_stop();
-}
-
-inline void dynamic_shim_llama_server_completion(struct dynamic_llama_server s,
-                                                 const char *json_req,
-                                                 ext_server_resp_t *resp) {
-  s.llama_server_completion(json_req, resp);
-}
-
-inline void dynamic_shim_llama_server_completion_next_result(
-    struct dynamic_llama_server s, const int task_id,
-    ext_server_task_result_t *result) {
-  s.llama_server_completion_next_result(task_id, result);
-}
-
-inline void dynamic_shim_llama_server_completion_cancel(
-    struct dynamic_llama_server s, const int task_id, ext_server_resp_t *err) {
-  s.llama_server_completion_cancel(task_id, err);
-}
-inline void dynamic_shim_llama_server_release_task_result(
-    struct dynamic_llama_server s, ext_server_task_result_t *result) {
-  s.llama_server_release_task_result(result);
-}
-
-inline void dynamic_shim_llama_server_tokenize(struct dynamic_llama_server s,
-                                               const char *json_req,
-                                               char **json_resp,
-                                               ext_server_resp_t *err) {
-  s.llama_server_tokenize(json_req, json_resp, err);
-}
-
-inline void dynamic_shim_llama_server_detokenize(struct dynamic_llama_server s,
-                                                 const char *json_req,
-                                                 char **json_resp,
-                                                 ext_server_resp_t *err) {
-  s.llama_server_detokenize(json_req, json_resp, err);
-}
-
-inline void dynamic_shim_llama_server_embedding(struct dynamic_llama_server s,
-                                                const char *json_req,
-                                                char **json_resp,
-                                                ext_server_resp_t *err) {
-  s.llama_server_embedding(json_req, json_resp, err);
-}
-
-inline void dynamic_shim_llama_server_release_json_resp(
-    struct dynamic_llama_server s, char **json_resp) {
-  s.llama_server_release_json_resp(json_resp);
-}
--- a/llm/dynamic_shim.h
+++ b/llm/dynamic_shim.h
@@ -1,74 +0,0 @@
-#include <stdlib.h>
-
-#include "ext_server.h"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-struct dynamic_llama_server {
-  void *handle;
-  void (*llama_server_init)(ext_server_params_t *sparams,
-                            ext_server_resp_t *err);
-  void (*llama_server_start)();
-  void (*llama_server_stop)();
-  void (*llama_server_completion)(const char *json_req,
-                                  ext_server_resp_t *resp);
-  void (*llama_server_completion_next_result)(const int task_id,
-                                              ext_server_task_result_t *result);
-  void (*llama_server_completion_cancel)(const int task_id,
-                                         ext_server_resp_t *err);
-  void (*llama_server_release_task_result)(ext_server_task_result_t *result);
-  void (*llama_server_tokenize)(const char *json_req, char **json_resp,
-                                ext_server_resp_t *err);
-  void (*llama_server_detokenize)(const char *json_req, char **json_resp,
-                                  ext_server_resp_t *err);
-  void (*llama_server_embedding)(const char *json_req, char **json_resp,
-                                 ext_server_resp_t *err);
-  void (*llama_server_release_json_resp)(char **json_resp);
-};
-
-void dynamic_shim_init(const char *libPath, struct dynamic_llama_server *s,
-                       ext_server_resp_t *err);
-
-// No good way to call C function pointers from Go so inline the indirection
-void dynamic_shim_llama_server_init(struct dynamic_llama_server s,
-                                    ext_server_params_t *sparams,
-                                    ext_server_resp_t *err);
-
-void dynamic_shim_llama_server_start(struct dynamic_llama_server s);
-
-void dynamic_shim_llama_server_stop(struct dynamic_llama_server s);
-
-void dynamic_shim_llama_server_completion(struct dynamic_llama_server s,
-                                          const char *json_req,
-                                          ext_server_resp_t *resp);
-
-void dynamic_shim_llama_server_completion_next_result(
-    struct dynamic_llama_server s, const int task_id,
-    ext_server_task_result_t *result);
-
-void dynamic_shim_llama_server_completion_cancel(struct dynamic_llama_server s,
-                                                 const int task_id,
-                                                 ext_server_resp_t *err);
-
-void dynamic_shim_llama_server_release_task_result(
-    struct dynamic_llama_server s, ext_server_task_result_t *result);
-
-void dynamic_shim_llama_server_tokenize(struct dynamic_llama_server s,
-                                        const char *json_req, char **json_resp,
-                                        ext_server_resp_t *err);
-
-void dynamic_shim_llama_server_detokenize(struct dynamic_llama_server s,
-                                          const char *json_req,
-                                          char **json_resp,
-                                          ext_server_resp_t *err);
-
-void dynamic_shim_llama_server_embedding(struct dynamic_llama_server s,
-                                         const char *json_req, char **json_resp,
-                                         ext_server_resp_t *err);
-void dynamic_shim_llama_server_release_json_resp(struct dynamic_llama_server s,
-                                                 char **json_resp);
-
-#ifdef __cplusplus
-}
-#endif
--- a/llm/ext_server/CMakeLists.txt
+++ b/llm/ext_server/CMakeLists.txt
@@ -1,29 +0,0 @@
-# Ollama specific CMakefile to include in llama.cpp/examples/server
-
-set(TARGET ext_server)
-option(LLAMA_SERVER_VERBOSE "Build verbose logging option for Server" ON)
-add_library(${TARGET} STATIC ../../../ext_server/ext_server.cpp)
-target_include_directories(${TARGET} PRIVATE ../../common)
-target_include_directories(${TARGET} PRIVATE ../..)
-target_include_directories(${TARGET} PRIVATE ../../..)
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
-target_compile_definitions(${TARGET} PUBLIC LLAMA_SERVER_LIBRARY=1)
-target_link_libraries(${TARGET} PRIVATE common llama llava ${CMAKE_THREAD_LIBS_INIT})
-target_compile_definitions(${TARGET} PRIVATE
-    SERVER_VERBOSE=$<BOOL:${LLAMA_SERVER_VERBOSE}>
-)
-
-if (BUILD_SHARED_LIBS)
-    set_target_properties(ext_server PROPERTIES POSITION_INDEPENDENT_CODE ON)
-    target_compile_definitions(ext_server PRIVATE LLAMA_SHARED LLAMA_BUILD)
-    add_library(ext_server_shared SHARED $<TARGET_OBJECTS:ext_server>)
-    target_link_libraries(ext_server_shared PRIVATE ggml llama llava common ${CMAKE_THREAD_LIBS_INIT})
-    install(TARGETS ext_server_shared LIBRARY)
-endif()
-
-if (CUDAToolkit_FOUND)
-    target_include_directories(${TARGET} PRIVATE ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
-    if (WIN32)
-        target_link_libraries(ext_server_shared PRIVATE nvml)
-    endif()
-endif()
--- a/llm/ext_server/README.md
+++ b/llm/ext_server/README.md
@@ -1,4 +0,0 @@
-# Extern C Server
-
-This directory contains a thin facade we layer on top of the Llama.cpp server
-to expose `extern C` interfaces to access the functionality through direct API calls in-process
--- a/llm/ext_server/ext_server.cpp
+++ b/llm/ext_server/ext_server.cpp
@@ -1,281 +0,0 @@
-#include "ext_server.h"
-
-// Necessary evil since the server types are not defined in a header
-#include "server.cpp"
-
-// Expose the llama server as a callable extern "C" API
-llama_server_context *llama = NULL;
-std::atomic<bool> ext_server_running(false);
-std::thread ext_server_thread;
-
-void llama_server_init(ext_server_params *sparams, ext_server_resp_t *err) {
-#if SERVER_VERBOSE != 1
-  log_disable();
-#endif
-  assert(err != NULL && sparams != NULL);
-  err->id = 0;
-  err->msg[0] = '\0';
-  try {
-    llama = new llama_server_context;
-    log_set_target(stdout);
-    gpt_params params;
-    params.n_ctx = sparams->n_ctx;
-    params.n_batch = sparams->n_batch;
-    if (sparams->n_threads > 0) {
-      params.n_threads = sparams->n_threads;
-    }
-    params.n_parallel = sparams->n_parallel;
-    params.rope_freq_base = sparams->rope_freq_base;
-    params.rope_freq_scale = sparams->rope_freq_scale;
-
-    if (sparams->memory_f16) {
-      params.cache_type_k = "f16";
-      params.cache_type_v = "f16";
-    } else {
-      params.cache_type_k = "f32";
-      params.cache_type_v = "f32";
-    }
-
-    params.n_gpu_layers = sparams->n_gpu_layers;
-    params.main_gpu = sparams->main_gpu;
-    params.use_mlock = sparams->use_mlock;
-    params.use_mmap = sparams->use_mmap;
-    params.numa = sparams->numa;
-    params.embedding = sparams->embedding;
-    if (sparams->model != NULL) {
-      params.model = sparams->model;
-    }
-
-    for (ext_server_lora_adapter *la = sparams->lora_adapters; la != NULL;
-         la = la->next) {
-      params.lora_adapter.push_back(std::make_tuple(la->adapter, la->scale));
-    }
-
-    if (sparams->mmproj != NULL) {
-      params.mmproj = std::string(sparams->mmproj);
-    }
-
-    llama_backend_init(params.numa);
-
-    // load the model
-    if (!llama->load_model(params)) {
-      // TODO - consider modifying the logging logic or patching load_model so
-      // we can capture more detailed error messages and pass them back to the
-      // caller for better UX
-      err->id = -1;
-      snprintf(err->msg, err->msg_len, "error loading model %s",
-               params.model.c_str());
-      return;
-    }
-
-    llama->initialize();
-  } catch (std::exception &e) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "exception %s", e.what());
-  } catch (...) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len,
-             "Unknown exception initializing llama server");
-  }
-}
-
-void llama_server_start() {
-  assert(llama != NULL);
-  // TODO mutex to protect thread creation
-  ext_server_thread = std::thread([&]() {
-    ext_server_running = true;
-    try {
-      LOG_TEE("llama server main loop starting\n");
-      ggml_time_init();
-      while (ext_server_running.load()) {
-        if (!llama->update_slots()) {
-          LOG_TEE(
-              "unexpected error in llama server update_slots - exiting main "
-              "loop\n");
-          break;
-        }
-      }
-    } catch (std::exception &e) {
-      LOG_TEE("caught exception in llama server main loop: %s\n", e.what());
-    } catch (...) {
-      LOG_TEE("caught unknown exception in llama server main loop\n");
-    }
-    LOG_TEE("\nllama server shutting down\n");
-    llama_backend_free();
-  });
-}
-
-void llama_server_stop() {
-  assert(llama != NULL);
-  // TODO - too verbose, remove once things are solid
-  LOG_TEE("requesting llama server shutdown\n");
-  ext_server_running = false;
-  ext_server_thread.join();
-  delete llama;
-  llama = NULL;
-  LOG_TEE("llama server shutdown complete\n");
-}
-
-void llama_server_completion(const char *json_req, ext_server_resp_t *resp) {
-  assert(llama != NULL && json_req != NULL && resp != NULL);
-  resp->id = -1;
-  resp->msg[0] = '\0';
-  try {
-    json data = json::parse(json_req);
-    resp->id = llama->request_completion(data, false, false, -1);
-  } catch (std::exception &e) {
-    snprintf(resp->msg, resp->msg_len, "exception %s", e.what());
-  } catch (...) {
-    snprintf(resp->msg, resp->msg_len, "Unknown exception during completion");
-  }
-}
-
-void llama_server_completion_next_result(const int task_id,
-                                         ext_server_task_result_t *resp) {
-  assert(llama != NULL && resp != NULL);
-  std::string msg;
-  resp->id = -1;
-  resp->stop = false;
-  resp->error = false;
-  resp->json_resp = NULL;
-  std::string result_json;
-  try {
-    task_result result = llama->next_result(task_id);
-    result_json =
-        result.result_json.dump(-1, ' ', false, json::error_handler_t::replace);
-    resp->id = result.id;
-    resp->stop = result.stop;
-    resp->error = result.error;
-    if (result.error) {
-      llama->request_cancel(task_id);
-    } else if (result.stop) {
-      llama->request_cancel(task_id);
-    }
-  } catch (std::exception &e) {
-    resp->error = true;
-    resp->id = -1;
-    result_json = "{\"error\":\"exception " + std::string(e.what()) + "\"}";
-    LOG_TEE("llama server completion exception %s\n", e.what());
-  } catch (...) {
-    resp->error = true;
-    resp->id = -1;
-    result_json = "{\"error\":\"Unknown exception during completion\"}";
-    LOG_TEE("llama server completion unknown exception\n");
-  }
-  const std::string::size_type size = result_json.size() + 1;
-  resp->json_resp = new char[size];
-  snprintf(resp->json_resp, size, "%s", result_json.c_str());
-}
-
-void llama_server_release_task_result(ext_server_task_result_t *result) {
-  if (result == NULL || result->json_resp == NULL) {
-    return;
-  }
-  delete[] result->json_resp;
-}
-
-void llama_server_completion_cancel(const int task_id, ext_server_resp_t *err) {
-  assert(llama != NULL && err != NULL);
-  err->id = 0;
-  err->msg[0] = '\0';
-  try {
-    llama->request_cancel(task_id);
-  } catch (std::exception &e) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "exception %s", e.what());
-  } catch (...) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len,
-             "Unknown exception completion cancel in llama server");
-  }
-}
-
-void llama_server_tokenize(const char *json_req, char **json_resp,
-                           ext_server_resp_t *err) {
-  assert(llama != NULL && json_req != NULL && json_resp != NULL && err != NULL);
-  *json_resp = NULL;
-  err->id = 0;
-  err->msg[0] = '\0';
-  try {
-    const json body = json::parse(json_req);
-    std::vector<llama_token> tokens;
-    if (body.count("content") != 0) {
-      tokens = llama->tokenize(body["content"], false);
-    }
-    const json data = format_tokenizer_response(tokens);
-    std::string result_json = data.dump();
-    const std::string::size_type size = result_json.size() + 1;
-    *json_resp = new char[size];
-    snprintf(*json_resp, size, "%s", result_json.c_str());
-  } catch (std::exception &e) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "exception %s", e.what());
-  } catch (...) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "Unknown exception during tokenize");
-  }
-}
-
-void llama_server_release_json_resp(char **json_resp) {
-  if (json_resp == NULL || *json_resp == NULL) {
-    return;
-  }
-  delete[] *json_resp;
-}
-
-void llama_server_detokenize(const char *json_req, char **json_resp,
-                             ext_server_resp_t *err) {
-  assert(llama != NULL && json_req != NULL && json_resp != NULL && err != NULL);
-  *json_resp = NULL;
-  err->id = 0;
-  err->msg[0] = '\0';
-  try {
-    const json body = json::parse(json_req);
-    std::string content;
-    if (body.count("tokens") != 0) {
-      const std::vector<llama_token> tokens = body["tokens"];
-      content = tokens_to_str(llama->ctx, tokens.cbegin(), tokens.cend());
-    }
-    const json data = format_detokenized_response(content);
-    std::string result_json = data.dump();
-    const std::string::size_type size = result_json.size() + 1;
-    *json_resp = new char[size];
-    snprintf(*json_resp, size, "%s", result_json.c_str());
-  } catch (std::exception &e) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "exception %s", e.what());
-  } catch (...) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "Unknown exception during detokenize");
-  }
-}
-
-void llama_server_embedding(const char *json_req, char **json_resp,
-                            ext_server_resp_t *err) {
-  assert(llama != NULL && json_req != NULL && json_resp != NULL && err != NULL);
-  *json_resp = NULL;
-  err->id = 0;
-  err->msg[0] = '\0';
-  try {
-    const json body = json::parse(json_req);
-    json prompt;
-    if (body.count("content") != 0) {
-      prompt = body["content"];
-    } else {
-      prompt = "";
-    }
-    const int task_id = llama->request_completion(
-        {{"prompt", prompt}, {"n_predict", 0}}, false, true, -1);
-    task_result result = llama->next_result(task_id);
-    std::string result_json = result.result_json.dump();
-    const std::string::size_type size = result_json.size() + 1;
-    *json_resp = new char[size];
-    snprintf(*json_resp, size, "%s", result_json.c_str());
-  } catch (std::exception &e) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "exception %s", e.what());
-  } catch (...) {
-    err->id = -1;
-    snprintf(err->msg, err->msg_len, "Unknown exception during embedding");
-  }
-}
--- a/llm/ext_server/ext_server.h
+++ b/llm/ext_server/ext_server.h
@@ -1,94 +0,0 @@
-#if defined(LLAMA_SERVER_LIBRARY)
-#ifndef LLAMA_SERVER_H
-#define LLAMA_SERVER_H
-#include <stdbool.h>
-#include <stddef.h>
-#include <stdint.h>
-#include <stdio.h>
-
-int __main(int argc, char **argv);
-
-// This exposes extern C entrypoints into the llama_server
-// To enable the server compile with LLAMA_SERVER_LIBRARY
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-typedef struct ext_server_resp {
-  int id;          // < 0 on error
-  size_t msg_len;  // caller must allocate msg and set msg_len
-  char *msg;
-} ext_server_resp_t;
-
-// Allocated and freed by caller
-typedef struct ext_server_lora_adapter {
-  char *adapter;
-  float scale;
-  struct ext_server_lora_adapter *next;
-} ext_server_lora_adapter_t;
-
-// Allocated and freed by caller
-typedef struct ext_server_params {
-  char *model;
-  uint32_t n_ctx;         // token context window, 0 = from model
-  uint32_t n_batch;       // prompt processing maximum batch size
-  uint32_t n_threads;     // number of threads to use for generation
-  int32_t n_parallel;     // number of parallel sequences to decodewra
-  float rope_freq_base;   // RoPE base frequency, 0 = from model
-  float rope_freq_scale;  // RoPE frequency scaling factor, 0 = from model
-  bool memory_f16;        // use f16 instead of f32 for memory kv
-  int32_t n_gpu_layers;  // number of layers to store in VRAM (-1 - use default)
-  int32_t main_gpu;      // the GPU that is used for scratch and small tensors
-  bool use_mlock;        // force system to keep model in RAM
-  bool use_mmap;         // use mmap if possible
-  bool numa;             // attempt optimizations that help on some NUMA systems
-  bool embedding;        // get only sentence embedding
-  ext_server_lora_adapter_t *lora_adapters;
-  char *mmproj;
-} ext_server_params_t;
-
-typedef struct ext_server_task_result {
-  int id;
-  bool stop;
-  bool error;
-  char *json_resp;  // null terminated, memory managed by ext_server
-} ext_server_task_result_t;
-
-// Initialize the server once per process
-// err->id = 0 for success and err->msg[0] = NULL
-// err->id != 0 for failure, and err->msg contains error message
-void llama_server_init(ext_server_params_t *sparams, ext_server_resp_t *err);
-
-// Run the main loop, called once per init
-void llama_server_start();
-// Stop the main loop and free up resources allocated in init and start.  Init
-// must be called again to reuse
-void llama_server_stop();
-
-// json_req null terminated string, memory managed by caller
-// resp->id >= 0 on success (task ID)
-// resp->id < 0 on error, and resp->msg contains error message
-void llama_server_completion(const char *json_req, ext_server_resp_t *resp);
-
-// Caller must call llama_server_release_task_result to free resp->json_resp
-void llama_server_completion_next_result(const int task_id,
-                                         ext_server_task_result_t *result);
-void llama_server_completion_cancel(const int task_id, ext_server_resp_t *err);
-void llama_server_release_task_result(ext_server_task_result_t *result);
-
-// Caller must call llama_server_releaes_json_resp to free json_resp if err.id <
-// 0
-void llama_server_tokenize(const char *json_req, char **json_resp,
-                           ext_server_resp_t *err);
-void llama_server_detokenize(const char *json_req, char **json_resp,
-                             ext_server_resp_t *err);
-void llama_server_embedding(const char *json_req, char **json_resp,
-                            ext_server_resp_t *err);
-void llama_server_release_json_resp(char **json_resp);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif
-#endif  // LLAMA_SERVER_LIBRARY
--- a/llm/ext_server_common.go
+++ b/llm/ext_server_common.go
@@ -1,360 +0,0 @@
-package llm
-
-/*
-#cgo CFLAGS: -I${SRCDIR}/ext_server -I${SRCDIR}/llama.cpp -I${SRCDIR}/llama.cpp/common -I${SRCDIR}/llama.cpp/examples/server
-#cgo CFLAGS: -DNDEBUG -DLLAMA_SERVER_LIBRARY=1 -D_XOPEN_SOURCE=600 -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64
-#cgo CFLAGS: -Wmissing-noreturn -Wall -Wextra -Wcast-qual -Wno-unused-function -Wno-array-bounds
-#cgo CPPFLAGS: -Ofast -Wall -Wextra -Wno-unused-function -Wno-unused-variable -Wno-deprecated-declarations -Wno-unused-but-set-variable
-#cgo darwin CFLAGS: -D_DARWIN_C_SOURCE
-#cgo darwin CPPFLAGS:  -DGGML_USE_ACCELERATE
-#cgo darwin CPPFLAGS: -DGGML_USE_METAL -DGGML_METAL_NDEBUG
-#cgo darwin LDFLAGS: -lc++ -framework Accelerate
-#cgo darwin LDFLAGS: -framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders
-#cgo darwin LDFLAGS: ${SRCDIR}/llama.cpp/build/darwin/metal/lib/libcommon.a
-#cgo darwin LDFLAGS: ${SRCDIR}/llama.cpp/build/darwin/metal/lib/libext_server.a
-#cgo darwin LDFLAGS: ${SRCDIR}/llama.cpp/build/darwin/metal/lib/libllama.a
-#cgo darwin LDFLAGS: ${SRCDIR}/llama.cpp/build/darwin/metal/lib/libggml_static.a
-#cgo linux CFLAGS: -D_GNU_SOURCE
-#cgo linux windows CFLAGS: -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_MMV_Y=1 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_USE_CUBLAS
-#cgo linux LDFLAGS: -L/usr/local/cuda/targets/x86_64-linux/lib -L/usr/local/cuda/lib64 -L/usr/local/cuda/targets/x86_64-linux/lib/stubs
-#cgo linux LDFLAGS: ${SRCDIR}/llama.cpp/build/linux/cpu/lib/libext_server.a
-#cgo linux LDFLAGS: ${SRCDIR}/llama.cpp/build/linux/cpu/lib/libcommon.a
-#cgo linux LDFLAGS: ${SRCDIR}/llama.cpp/build/linux/cpu/lib/libllama.a
-#cgo linux LDFLAGS: ${SRCDIR}/llama.cpp/build/linux/cpu/lib/libggml_static.a
-#cgo linux LDFLAGS: -lrt -ldl -lstdc++ -lm
-#cgo linux windows LDFLAGS: -lpthread
-
-#include <stdlib.h>
-#include "ext_server.h"
-
-*/
-import "C"
-import (
-	"bytes"
-	"context"
-	"encoding/json"
-	"fmt"
-	"log"
-	"os"
-	"strings"
-	"sync"
-	"time"
-	"unsafe"
-
-	"github.com/jmorganca/ollama/api"
-	"github.com/jmorganca/ollama/gpu"
-)
-
-type extServer interface {
-	LLM
-	llama_server_init(sparams *C.ext_server_params_t, err *C.ext_server_resp_t)
-	llama_server_start()
-	llama_server_stop()
-	llama_server_completion(json_req *C.char, resp *C.ext_server_resp_t)
-	llama_server_completion_next_result(task_id C.int, resp *C.ext_server_task_result_t)
-	llama_server_completion_cancel(task_id C.int, err *C.ext_server_resp_t)
-	llama_server_release_task_result(result *C.ext_server_task_result_t)
-	llama_server_tokenize(json_req *C.char, json_resp **C.char, err *C.ext_server_resp_t)
-	llama_server_detokenize(json_req *C.char, json_resp **C.char, err *C.ext_server_resp_t)
-	llama_server_embedding(json_req *C.char, json_resp **C.char, err *C.ext_server_resp_t)
-	llama_server_release_json_resp(json_resp **C.char)
-}
-
-// Note: current implementation does not support concurrent instantiations
-var mutex sync.Mutex
-
-func newExtServerResp(len C.size_t) C.ext_server_resp_t {
-	var resp C.ext_server_resp_t
-	resp.msg_len = len
-	bytes := make([]byte, len)
-	resp.msg = (*C.char)(C.CBytes(bytes))
-	return resp
-}
-
-func freeExtServerResp(resp C.ext_server_resp_t) {
-	if resp.msg_len == 0 {
-		return
-	}
-	C.free(unsafe.Pointer(resp.msg))
-}
-
-func extServerResponseToErr(resp C.ext_server_resp_t) error {
-	return fmt.Errorf(C.GoString(resp.msg))
-}
-
-func newExtServer(server extServer, model string, adapters, projectors []string, numLayers int64, opts api.Options) (extServer, error) {
-	if !mutex.TryLock() {
-		log.Printf("concurrent llm servers not yet supported, waiting for prior server to complete")
-		mutex.Lock()
-	}
-	fileInfo, err := os.Stat(model)
-	if err != nil {
-		return nil, err
-	}
-	var sparams C.ext_server_params_t
-	sparams.model = C.CString(model)
-	defer C.free(unsafe.Pointer(sparams.model))
-
-	numGPU := gpu.NumGPU(numLayers, fileInfo.Size(), opts)
-
-	sparams.embedding = true
-	sparams.n_ctx = C.uint(opts.NumCtx)
-	sparams.n_batch = C.uint(opts.NumBatch)
-	sparams.n_gpu_layers = C.int(numGPU)
-	sparams.main_gpu = C.int(opts.MainGPU)
-	sparams.n_parallel = 1 // TODO - wire up concurrency
-
-	// Always use the value encoded in the model
-	sparams.rope_freq_base = 0.0
-	sparams.rope_freq_scale = 0.0
-	sparams.memory_f16 = C.bool(opts.F16KV)
-	sparams.use_mlock = C.bool(opts.UseMLock)
-	sparams.use_mmap = C.bool(opts.UseMMap)
-	sparams.numa = C.bool(opts.UseNUMA)
-
-	sparams.lora_adapters = nil
-	for i := 0; i < len(adapters); i++ {
-		la := (*C.ext_server_lora_adapter_t)(C.malloc(C.sizeof_ext_server_lora_adapter_t))
-		defer C.free(unsafe.Pointer(la))
-		la.adapter = C.CString(adapters[i])
-		defer C.free(unsafe.Pointer(la.adapter))
-		la.scale = C.float(1.0) // TODO expose scale/weights up through ollama UX
-		la.next = nil
-		if i == 0 {
-			sparams.lora_adapters = la
-		} else {
-			tmp := sparams.lora_adapters
-			for ; tmp.next != nil; tmp = tmp.next {
-			}
-			tmp.next = la
-		}
-	}
-
-	if len(projectors) > 0 {
-		// TODO: applying multiple projectors is not supported by the llama.cpp server yet
-		sparams.mmproj = C.CString(projectors[0])
-		defer C.free(unsafe.Pointer(sparams.mmproj))
-	} else {
-		sparams.mmproj = nil
-	}
-
-	sparams.n_threads = C.uint(opts.NumThread)
-
-	log.Printf("Initializing internal llama server")
-	resp := newExtServerResp(128)
-	defer freeExtServerResp(resp)
-	server.llama_server_init(&sparams, &resp)
-	if resp.id < 0 {
-		return nil, extServerResponseToErr(resp)
-	}
-
-	log.Printf("Starting internal llama main loop")
-	server.llama_server_start()
-	return server, nil
-}
-
-func predict(ctx context.Context, llm extServer, predict PredictOpts, fn func(PredictResult)) error {
-	resp := newExtServerResp(128)
-	defer freeExtServerResp(resp)
-	var imageData []ImageData
-	if len(predict.Images) > 0 {
-		for cnt, i := range predict.Images {
-			imageData = append(imageData, ImageData{Data: i, ID: cnt})
-		}
-	}
-	log.Printf("loaded %d images", len(imageData))
-
-	request := map[string]any{
-		"prompt":            predict.Prompt,
-		"stream":            true,
-		"n_predict":         predict.Options.NumPredict,
-		"n_keep":            predict.Options.NumKeep,
-		"temperature":       predict.Options.Temperature,
-		"top_k":             predict.Options.TopK,
-		"top_p":             predict.Options.TopP,
-		"tfs_z":             predict.Options.TFSZ,
-		"typical_p":         predict.Options.TypicalP,
-		"repeat_last_n":     predict.Options.RepeatLastN,
-		"repeat_penalty":    predict.Options.RepeatPenalty,
-		"presence_penalty":  predict.Options.PresencePenalty,
-		"frequency_penalty": predict.Options.FrequencyPenalty,
-		"mirostat":          predict.Options.Mirostat,
-		"mirostat_tau":      predict.Options.MirostatTau,
-		"mirostat_eta":      predict.Options.MirostatEta,
-		"penalize_nl":       predict.Options.PenalizeNewline,
-		"seed":              predict.Options.Seed,
-		"stop":              predict.Options.Stop,
-		"image_data":        imageData,
-		"cache_prompt":      true,
-	}
-
-	if predict.Format == "json" {
-		request["grammar"] = jsonGrammar
-	}
-
-	retryDelay := 100 * time.Microsecond
-	for retries := 0; retries < maxRetries; retries++ {
-		if retries > 0 {
-			time.Sleep(retryDelay) // wait before retrying
-			retryDelay *= 2        // exponential backoff
-		}
-
-		// Handling JSON marshaling with special characters unescaped.
-		buffer := &bytes.Buffer{}
-		enc := json.NewEncoder(buffer)
-		enc.SetEscapeHTML(false)
-
-		if err := enc.Encode(request); err != nil {
-			return fmt.Errorf("failed to marshal data: %w", err)
-		}
-
-		req := C.CString(buffer.String())
-		defer C.free(unsafe.Pointer(req))
-
-		llm.llama_server_completion(req, &resp)
-		if resp.id < 0 {
-			return extServerResponseToErr(resp)
-		}
-
-		retryNeeded := false
-	out:
-		for {
-			select {
-			case <-ctx.Done():
-				// This handles the request cancellation
-				llm.llama_server_completion_cancel(resp.id, &resp)
-				if resp.id < 0 {
-					return extServerResponseToErr(resp)
-				} else {
-					return nil
-				}
-			default:
-				var result C.ext_server_task_result_t
-				llm.llama_server_completion_next_result(resp.id, &result)
-				json_resp := C.GoString(result.json_resp)
-				llm.llama_server_release_task_result(&result)
-
-				var p prediction
-				if err := json.Unmarshal([]byte(json_resp), &p); err != nil {
-					llm.llama_server_completion_cancel(resp.id, &resp)
-					if resp.id < 0 {
-						return fmt.Errorf("error unmarshaling llm prediction response: %w and cancel %s", err, C.GoString(resp.msg))
-					} else {
-						return fmt.Errorf("error unmarshaling llm prediction response: %w", err)
-					}
-				}
-
-				if bool(result.error) && strings.Contains(json_resp, "slot unavailable") {
-					retryNeeded = true
-					// task will already be canceled
-					break out
-				}
-
-				if p.Content != "" {
-					fn(PredictResult{
-						Content: p.Content,
-					})
-				}
-
-				if p.Stop {
-					fn(PredictResult{
-						Done:               true,
-						PromptEvalCount:    p.Timings.PromptN,
-						PromptEvalDuration: parseDurationMs(p.Timings.PromptMS),
-						EvalCount:          p.Timings.PredictedN,
-						EvalDuration:       parseDurationMs(p.Timings.PredictedMS),
-					})
-					return nil
-				}
-			}
-		}
-		if !retryNeeded {
-			return nil // success
-		}
-	}
-
-	// should never reach here ideally
-	return fmt.Errorf("max retries exceeded")
-}
-
-func encode(llm extServer, ctx context.Context, prompt string) ([]int, error) {
-	data, err := json.Marshal(TokenizeRequest{Content: prompt})
-	if err != nil {
-		return nil, fmt.Errorf("marshaling encode data: %w", err)
-	}
-	req := C.CString(string(data))
-	defer C.free(unsafe.Pointer(req))
-	var json_resp *C.char
-	resp := newExtServerResp(128)
-	defer freeExtServerResp(resp)
-	llm.llama_server_tokenize(req, &json_resp, &resp)
-	if resp.id < 0 {
-		return nil, extServerResponseToErr(resp)
-	}
-	defer llm.llama_server_release_json_resp(&json_resp)
-
-	var encoded TokenizeResponse
-	if err2 := json.Unmarshal([]byte(C.GoString(json_resp)), &encoded); err2 != nil {
-		return nil, fmt.Errorf("unmarshal encode response: %w", err2)
-	}
-
-	return encoded.Tokens, err
-}
-
-func decode(llm extServer, ctx context.Context, tokens []int) (string, error) {
-	if len(tokens) == 0 {
-		return "", nil
-	}
-	data, err := json.Marshal(DetokenizeRequest{Tokens: tokens})
-	if err != nil {
-		return "", fmt.Errorf("marshaling decode data: %w", err)
-	}
-
-	req := C.CString(string(data))
-	defer C.free(unsafe.Pointer(req))
-	var json_resp *C.char
-	resp := newExtServerResp(128)
-	defer freeExtServerResp(resp)
-	llm.llama_server_detokenize(req, &json_resp, &resp)
-	if resp.id < 0 {
-		return "", extServerResponseToErr(resp)
-	}
-	defer llm.llama_server_release_json_resp(&json_resp)
-
-	var decoded DetokenizeResponse
-	if err2 := json.Unmarshal([]byte(C.GoString(json_resp)), &decoded); err2 != nil {
-		return "", fmt.Errorf("unmarshal encode response: %w", err2)
-	}
-
-	return decoded.Content, err
-}
-
-func embedding(llm extServer, ctx context.Context, input string) ([]float64, error) {
-	data, err := json.Marshal(TokenizeRequest{Content: input})
-	if err != nil {
-		return nil, fmt.Errorf("error marshaling embed data: %w", err)
-	}
-
-	req := C.CString(string(data))
-	defer C.free(unsafe.Pointer(req))
-	var json_resp *C.char
-	resp := newExtServerResp(128)
-	defer freeExtServerResp(resp)
-	llm.llama_server_embedding(req, &json_resp, &resp)
-	if resp.id < 0 {
-		return nil, extServerResponseToErr(resp)
-	}
-	defer llm.llama_server_release_json_resp(&json_resp)
-
-	var embedding EmbeddingResponse
-	if err := json.Unmarshal([]byte(C.GoString(json_resp)), &embedding); err != nil {
-		return nil, fmt.Errorf("unmarshal tokenize response: %w", err)
-	}
-
-	return embedding.Embedding, nil
-}
-
-func close(llm extServer) {
-	llm.llama_server_stop()
-	mutex.Unlock()
-}
--- a/llm/ext_server_default.go
+++ b/llm/ext_server_default.go
@@ -1,80 +0,0 @@
-//go:build !windows
-
-package llm
-
-/*
-#include <stdlib.h>
-#include "ext_server.h"
-
-*/
-import "C"
-import (
-	"context"
-
-	"github.com/jmorganca/ollama/api"
-)
-
-type llamaExtServer struct {
-	api.Options
-}
-
-func (llm *llamaExtServer) llama_server_init(sparams *C.ext_server_params_t, err *C.ext_server_resp_t) {
-	C.llama_server_init(sparams, err)
-}
-func (llm *llamaExtServer) llama_server_start() {
-	C.llama_server_start()
-}
-func (llm *llamaExtServer) llama_server_stop() {
-	C.llama_server_stop()
-}
-
-func (llm *llamaExtServer) llama_server_completion(json_req *C.char, resp *C.ext_server_resp_t) {
-	C.llama_server_completion(json_req, resp)
-}
-func (llm *llamaExtServer) llama_server_completion_next_result(task_id C.int, resp *C.ext_server_task_result_t) {
-	C.llama_server_completion_next_result(task_id, resp)
-}
-func (llm *llamaExtServer) llama_server_completion_cancel(task_id C.int, err *C.ext_server_resp_t) {
-	C.llama_server_completion_cancel(task_id, err)
-}
-func (llm *llamaExtServer) llama_server_release_task_result(result *C.ext_server_task_result_t) {
-	C.llama_server_release_task_result(result)
-}
-
-func (llm *llamaExtServer) llama_server_tokenize(json_req *C.char, json_resp **C.char, err *C.ext_server_resp_t) {
-	C.llama_server_tokenize(json_req, json_resp, err)
-}
-func (llm *llamaExtServer) llama_server_detokenize(json_req *C.char, json_resp **C.char, err *C.ext_server_resp_t) {
-	C.llama_server_detokenize(json_req, json_resp, err)
-}
-func (llm *llamaExtServer) llama_server_embedding(json_req *C.char, json_resp **C.char, err *C.ext_server_resp_t) {
-	C.llama_server_embedding(json_req, json_resp, err)
-}
-func (llm *llamaExtServer) llama_server_release_json_resp(json_resp **C.char) {
-	C.llama_server_release_json_resp(json_resp)
-}
-
-func newDefaultExtServer(model string, adapters, projectors []string, numLayers int64, opts api.Options) (extServer, error) {
-	server := &llamaExtServer{opts}
-	return newExtServer(server, model, adapters, projectors, numLayers, opts)
-}
-
-func (llm *llamaExtServer) Predict(ctx context.Context, pred PredictOpts, fn func(PredictResult)) error {
-	return predict(ctx, llm, pred, fn)
-}
-
-func (llm *llamaExtServer) Encode(ctx context.Context, prompt string) ([]int, error) {
-	return encode(llm, ctx, prompt)
-}
-
-func (llm *llamaExtServer) Decode(ctx context.Context, tokens []int) (string, error) {
-	return decode(llm, ctx, tokens)
-}
-
-func (llm *llamaExtServer) Embedding(ctx context.Context, input string) ([]float64, error) {
-	return embedding(llm, ctx, input)
-}
-
-func (llm *llamaExtServer) Close() {
-	close(llm)
-}
--- a/llm/ext_server_windows.go
+++ b/llm/ext_server_windows.go
@@ -1,12 +0,0 @@
-package llm
-
-import (
-	"github.com/jmorganca/ollama/api"
-)
-
-func newDefaultExtServer(model string, adapters, projectors []string, numLayers int64, opts api.Options) (extServer, error) {
-	// On windows we always load the llama.cpp libraries dynamically to avoid startup DLL dependencies
-	// This ensures we can update the PATH at runtime to get everything loaded
-
-	return newDynamicShimExtServer(AvailableShims["cpu"], model, adapters, projectors, numLayers, opts)
-}
--- a/llm/falcon.go
+++ b/llm/falcon.go
@@ -0,0 +1,20 @@
+package llm
+
+const (
+	falconModelType7B   = 32
+	falconModelType40B  = 60
+	falconModelType180B = 80
+)
+
+func falconModelType(numLayer uint32) string {
+	switch numLayer {
+	case 32:
+		return "7B"
+	case 60:
+		return "40B"
+	case 80:
+		return "180B"
+	default:
+		return "unknown"
+	}
+}
--- a/llm/generate/gen_common.sh
+++ b/llm/generate/gen_common.sh
@@ -1,57 +0,0 @@
-# common logic accross linux and darwin
-
-init_vars() {
-    LLAMACPP_DIR=../llama.cpp
-    CMAKE_DEFS=""
-    CMAKE_TARGETS="--target ggml --target ggml_static --target llama --target build_info --target common --target ext_server --target llava_static"
-    if echo "${CGO_CFLAGS}" | grep -- '-g' >/dev/null; then
-        CMAKE_DEFS="-DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_VERBOSE_MAKEFILE=on -DLLAMA_GPROF=on -DLLAMA_SERVER_VERBOSE=on"
-    else
-        # TODO - add additional optimization flags...
-        CMAKE_DEFS="-DCMAKE_BUILD_TYPE=Release -DLLAMA_SERVER_VERBOSE=off"
-    fi
-}
-
-git_module_setup() {
-    if [ -n "${OLLAMA_SKIP_PATCHING}" ]; then
-        echo "Skipping submodule initialization"
-        return
-    fi
-    # Make sure the tree is clean after the directory moves
-    if [ -d "${LLAMACPP_DIR}/gguf" ]; then
-        echo "Cleaning up old submodule"
-        rm -rf ${LLAMACPP_DIR}
-    fi
-    git submodule init
-    git submodule update --force ${LLAMACPP_DIR}
-
-}
-
-apply_patches() {
-    # Wire up our CMakefile
-    if ! grep ollama ${LLAMACPP_DIR}/examples/server/CMakeLists.txt; then
-        echo 'include (../../../ext_server/CMakeLists.txt) # ollama' >>${LLAMACPP_DIR}/examples/server/CMakeLists.txt
-    fi
-    # Avoid duplicate main symbols when we link into the cgo binary
-    sed -e 's/int main(/int __main(/g' <${LLAMACPP_DIR}/examples/server/server.cpp >${LLAMACPP_DIR}/examples/server/server.cpp.tmp &&
-        mv ${LLAMACPP_DIR}/examples/server/server.cpp.tmp ${LLAMACPP_DIR}/examples/server/server.cpp
-}
-
-build() {
-    cmake -S ${LLAMACPP_DIR} -B ${BUILD_DIR} ${CMAKE_DEFS}
-    cmake --build ${BUILD_DIR} ${CMAKE_TARGETS} -j8
-}
-
-install() {
-    rm -rf ${BUILD_DIR}/lib
-    mkdir -p ${BUILD_DIR}/lib
-    cp ${BUILD_DIR}/examples/server/libext_server.a ${BUILD_DIR}/lib
-    cp ${BUILD_DIR}/common/libcommon.a ${BUILD_DIR}/lib
-    cp ${BUILD_DIR}/libllama.a ${BUILD_DIR}/lib
-    cp ${BUILD_DIR}/libggml_static.a ${BUILD_DIR}/lib
-}
-
-# Keep the local tree clean after we're done with the build
-cleanup() {
-    (cd ${LLAMACPP_DIR}/examples/server/ && git checkout CMakeLists.txt server.cpp)
-}
--- a/llm/generate/gen_darwin.sh
+++ b/llm/generate/gen_darwin.sh
@@ -1,32 +0,0 @@
-#!/bin/bash
-# This script is intended to run inside the go generate
-# working directory must be ./llm/generate/
-
-# TODO - add hardening to detect missing tools (cmake, etc.)
-
-set -ex
-set -o pipefail
-echo "Starting darwin generate script"
-source $(dirname $0)/gen_common.sh
-init_vars
-CMAKE_DEFS="-DCMAKE_OSX_DEPLOYMENT_TARGET=11.0 -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on ${CMAKE_DEFS}"
-BUILD_DIR="${LLAMACPP_DIR}/build/darwin/metal"
-case "${GOARCH}" in
-"amd64")
-    CMAKE_DEFS="-DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
-    ;;
-"arm64")
-    CMAKE_DEFS="-DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64 ${CMAKE_DEFS}"
-    ;;
-*)
-    echo "GOARCH must be set"
-    echo "this script is meant to be run from within go generate"
-    exit 1
-    ;;
-esac
-
-git_module_setup
-apply_patches
-build
-install
-cleanup
--- a/llm/generate/gen_linux.sh
+++ b/llm/generate/gen_linux.sh
@@ -1,116 +0,0 @@
-#!/bin/bash
-# This script is intended to run inside the go generate
-# working directory must be llm/generate/
-
-# First we build our default built-in library which will be linked into the CGO
-# binary as a normal dependency. This default build is CPU based.
-#
-# Then we build a CUDA dynamic library (although statically linked with the CUDA
-# library dependencies for maximum portability)
-#
-# Then if we detect ROCm, we build a dynamically loaded ROCm lib.  ROCm is particularly
-# important to be a dynamic lib even if it's the only GPU library detected because
-# we can't redistribute the objectfiles but must rely on dynamic libraries at
-# runtime, which could lead the server not to start if not present.
-
-set -ex
-set -o pipefail
-
-# See https://llvm.org/docs/AMDGPUUsage.html#processors for reference
-amdGPUs() {
-    GPU_LIST=(
-        "gfx803"
-        "gfx900"
-        "gfx906:xnack-"
-        "gfx908:xnack-"
-        "gfx90a:xnack+"
-        "gfx90a:xnack-"
-        "gfx1010"
-        "gfx1012"
-        "gfx1030"
-        "gfx1100"
-        "gfx1101"
-        "gfx1102"
-    )
-    (
-        IFS=$';'
-        echo "'${GPU_LIST[*]}'"
-    )
-}
-
-echo "Starting linux generate script"
-if [ -z "${CUDACXX}" -a -x /usr/local/cuda/bin/nvcc ]; then
-    export CUDACXX=/usr/local/cuda/bin/nvcc
-fi
-COMMON_CMAKE_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off"
-source $(dirname $0)/gen_common.sh
-init_vars
-git_module_setup
-apply_patches
-
-#
-# CPU first for the default library
-#
-CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS}"
-BUILD_DIR="${LLAMACPP_DIR}/build/linux/cpu"
-
-build
-install
-
-# Placeholder to keep go embed happy until we start building dynamic CPU lib variants
-touch ${BUILD_DIR}/lib/dummy.so
-
-if [ -d /usr/local/cuda/lib64/ ]; then
-    echo "CUDA libraries detected - building dynamic CUDA library"
-    init_vars
-    CMAKE_DEFS="-DLLAMA_CUBLAS=on ${COMMON_CMAKE_DEFS} ${CMAKE_DEFS}"
-    BUILD_DIR="${LLAMACPP_DIR}/build/linux/cuda"
-    CUDA_LIB_DIR=/usr/local/cuda/lib64
-    build
-    install
-    gcc -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.so \
-        -Wl,--whole-archive \
-        ${BUILD_DIR}/lib/libext_server.a \
-        ${BUILD_DIR}/lib/libcommon.a \
-        ${BUILD_DIR}/lib/libllama.a \
-        -Wl,--no-whole-archive \
-        ${CUDA_LIB_DIR}/libcudart_static.a \
-        ${CUDA_LIB_DIR}/libcublas_static.a \
-        ${CUDA_LIB_DIR}/libcublasLt_static.a \
-        ${CUDA_LIB_DIR}/libcudadevrt.a \
-        ${CUDA_LIB_DIR}/libculibos.a \
-        -lrt -lpthread -ldl -lstdc++ -lm
-fi
-
-if [ -z "${ROCM_PATH}" ]; then
-    # Try the default location in case it exists
-    ROCM_PATH=/opt/rocm
-fi
-
-if [ -z "${CLBlast_DIR}" ]; then
-    # Try the default location in case it exists
-    if [ -d /usr/lib/cmake/CLBlast ]; then
-        export CLBlast_DIR=/usr/lib/cmake/CLBlast
-    fi
-fi
-
-if [ -d "${ROCM_PATH}" ]; then
-    echo "ROCm libraries detected - building dynamic ROCm library"
-    init_vars
-    CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DLLAMA_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
-    BUILD_DIR="${LLAMACPP_DIR}/build/linux/rocm"
-    build
-    install
-    gcc -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.so \
-        -Wl,--whole-archive \
-        ${BUILD_DIR}/lib/libext_server.a \
-        ${BUILD_DIR}/lib/libcommon.a \
-        ${BUILD_DIR}/lib/libllama.a \
-        -Wl,--no-whole-archive \
-        -lrt -lpthread -ldl -lstdc++ -lm \
-        -L/opt/rocm/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ \
-        -Wl,-rpath,/opt/rocm/lib,-rpath,/opt/amdgpu/lib/x86_64-linux-gnu/ \
-        -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu
-fi
-
-cleanup
--- a/llm/generate/gen_windows.ps1
+++ b/llm/generate/gen_windows.ps1
@@ -1,87 +0,0 @@
-#!powershell
-
-$ErrorActionPreference = "Stop"
-
-function init_vars {
-    $script:llamacppDir = "../llama.cpp"
-    $script:cmakeDefs = @("-DBUILD_SHARED_LIBS=on", "-DLLAMA_NATIVE=off", "-DLLAMA_F16C=off", "-DLLAMA_FMA=off", "-DLLAMA_AVX512=off", "-DLLAMA_AVX2=off", "-DLLAMA_AVX=on", "-A","x64")
-    $script:cmakeTargets = @("ggml", "ggml_static", "llama", "build_info", "common", "ext_server_shared", "llava_static")
-    if ($env:CGO_CFLAGS -contains "-g") {
-        $script:cmakeDefs += @("-DCMAKE_VERBOSE_MAKEFILE=on", "-DLLAMA_SERVER_VERBOSE=on")
-        $script:config = "RelWithDebInfo"
-    } else {
-        $script:cmakeDefs += @("-DLLAMA_SERVER_VERBOSE=off")
-        $script:config = "Release"
-    }
-}
-
-function git_module_setup {
-    # TODO add flags to skip the init/patch logic to make it easier to mod llama.cpp code in-repo
-    & git submodule init
-    if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
-    & git submodule update --force "${script:llamacppDir}"
-    if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
-}
-
-function apply_patches {
-    # Wire up our CMakefile
-    if (!(Select-String -Path "${script:llamacppDir}/examples/server/CMakeLists.txt" -Pattern 'ollama')) {
-        Add-Content -Path "${script:llamacppDir}/examples/server/CMakeLists.txt" -Value 'include (../../../ext_server/CMakeLists.txt) # ollama'
-    }
-    # Avoid duplicate main symbols when we link into the cgo binary
-    $content = Get-Content -Path "${script:llamacppDir}/examples/server/server.cpp"
-    $content = $content -replace 'int main\(', 'int __main('
-    Set-Content -Path "${script:llamacppDir}/examples/server/server.cpp" -Value $content
-}
-
-function build {
-    write-host "generating config with: cmake -S ${script:llamacppDir} -B $script:buildDir $script:cmakeDefs"
-    & cmake --version
-    & cmake -S "${script:llamacppDir}" -B $script:buildDir $script:cmakeDefs
-    if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
-    write-host "building with: cmake --build $script:buildDir --config $script:config ($script:cmakeTargets | ForEach-Object { "--target", $_ })"
-    & cmake --build $script:buildDir --config $script:config ($script:cmakeTargets | ForEach-Object { "--target", $_ })
-    if ($LASTEXITCODE -ne 0) { exit($LASTEXITCODE)}
-}
-
-function install {
-    rm -ea 0 -recurse -force -path "${script:buildDir}/lib"
-    md "${script:buildDir}/lib" -ea 0 > $null
-    cp "${script:buildDir}/bin/${script:config}/ext_server_shared.dll" "${script:buildDir}/lib"
-    cp "${script:buildDir}/bin/${script:config}/llama.dll" "${script:buildDir}/lib"
-
-    # Display the dll dependencies in the build log
-    dumpbin /dependents "${script:buildDir}/bin/${script:config}/ext_server_shared.dll" | select-string ".dll"
-}
-
-function cleanup {
-    Set-Location "${script:llamacppDir}/examples/server"
-    git checkout CMakeLists.txt server.cpp
-}
-
-init_vars
-git_module_setup
-apply_patches
-
-# first build CPU based
-$script:buildDir="${script:llamacppDir}/build/windows/cpu"
-
-build
-install
-
-# Then build cuda as a dynamically loaded library
-init_vars
-$script:buildDir="${script:llamacppDir}/build/windows/cuda"
-$script:cmakeDefs += @("-DLLAMA_CUBLAS=ON")
-build
-install
-
-# TODO - actually implement ROCm support on windows
-$script:buildDir="${script:llamacppDir}/build/windows/rocm"
-
-rm -ea 0 -recurse -force -path "${script:buildDir}/lib"
-md "${script:buildDir}/lib" -ea 0 > $null
-echo $null >> "${script:buildDir}/lib/.generated"
-
-cleanup
-write-host "`ngo generate completed"
--- a/llm/generate/generate_darwin.go
+++ b/llm/generate/generate_darwin.go
@@ -1,3 +0,0 @@
-package generate
-
-//go:generate sh ./gen_darwin.sh
--- a/llm/generate/generate_linux.go
+++ b/llm/generate/generate_linux.go
@@ -1,3 +0,0 @@
-package generate
-
-//go:generate bash ./gen_linux.sh
--- a/llm/generate/generate_windows.go
+++ b/llm/generate/generate_windows.go
@@ -1,3 +0,0 @@
-package generate
-
-//go:generate powershell -ExecutionPolicy Bypass -File ./gen_windows.ps1
--- a/llm/ggml.go
+++ b/llm/ggml.go
@@ -7,10 +7,9 @@ import (
 )

 type GGML struct {
+	magic uint32
 	container
 	model
-
-	Size int64
 }

 const (
@@ -83,7 +82,65 @@ type model interface {

 type container interface {
 	Name() string
-	Decode(*readSeekOffset) (model, error)
+	Decode(io.Reader) (model, error)
+}
+
+type containerGGML struct{}
+
+func (c *containerGGML) Name() string {
+	return "ggml"
+}
+
+func (c *containerGGML) Decode(r io.Reader) (model, error) {
+	return nil, nil
+}
+
+type containerGGMF struct {
+	version uint32
+}
+
+func (c *containerGGMF) Name() string {
+	return "ggmf"
+}
+
+func (c *containerGGMF) Decode(r io.Reader) (model, error) {
+	var version uint32
+	binary.Read(r, binary.LittleEndian, &version)
+
+	switch version {
+	case 1:
+	default:
+		return nil, errors.New("invalid version")
+	}
+
+	c.version = version
+	return nil, nil
+}
+
+type containerGGJT struct {
+	version uint32
+}
+
+func (c *containerGGJT) Name() string {
+	return "ggjt"
+}
+
+func (c *containerGGJT) Decode(r io.Reader) (model, error) {
+	var version uint32
+	binary.Read(r, binary.LittleEndian, &version)
+
+	switch version {
+	case 1, 2, 3:
+	default:
+		return nil, errors.New("invalid version")
+	}
+
+	c.version = version
+
+	// different model types may have different layouts for hyperparameters
+	var llama llamaModel
+	binary.Read(r, binary.LittleEndian, &llama.hyperparameters)
+	return &llama, nil
 }

 type containerLORA struct {
@@ -94,9 +151,9 @@ func (c *containerLORA) Name() string {
 	return "ggla"
 }

-func (c *containerLORA) Decode(ro *readSeekOffset) (model, error) {
+func (c *containerLORA) Decode(r io.Reader) (model, error) {
 	var version uint32
-	binary.Read(ro, binary.LittleEndian, &version)
+	binary.Read(r, binary.LittleEndian, &version)

 	switch version {
 	case 1:
@@ -105,10 +162,6 @@ func (c *containerLORA) Decode(ro *readSeekOffset) (model, error) {
 	}

 	c.version = version
-
-	// remaining file contents aren't decoded
-	ro.Seek(0, io.SeekEnd)
-
 	return nil, nil
 }

@@ -126,60 +179,34 @@ const (
 	FILE_MAGIC_GGUF_BE = 0x47475546
 )

-var ErrUnsupportedFormat = errors.New("unsupported model format")
-
 func DecodeGGML(r io.ReadSeeker) (*GGML, error) {
-	ro := readSeekOffset{ReadSeeker: r}
+	var ggml GGML
+	binary.Read(r, binary.LittleEndian, &ggml.magic)

-	var magic uint32
-	if err := binary.Read(&ro, binary.LittleEndian, &magic); err != nil {
-		return nil, err
-	}
-
-	var c container
-	switch magic {
-	case FILE_MAGIC_GGML, FILE_MAGIC_GGMF, FILE_MAGIC_GGJT:
-		return nil, ErrUnsupportedFormat
+	switch ggml.magic {
+	case FILE_MAGIC_GGML:
+		ggml.container = &containerGGML{}
+	case FILE_MAGIC_GGMF:
+		ggml.container = &containerGGMF{}
+	case FILE_MAGIC_GGJT:
+		ggml.container = &containerGGJT{}
 	case FILE_MAGIC_GGLA:
-		c = &containerLORA{}
+		ggml.container = &containerLORA{}
 	case FILE_MAGIC_GGUF_LE:
-		c = &containerGGUF{bo: binary.LittleEndian}
+		ggml.container = &containerGGUF{bo: binary.LittleEndian}
 	case FILE_MAGIC_GGUF_BE:
-		c = &containerGGUF{bo: binary.BigEndian}
+		ggml.container = &containerGGUF{bo: binary.BigEndian}
 	default:
 		return nil, errors.New("invalid file magic")
 	}

-	model, err := c.Decode(&ro)
+	model, err := ggml.Decode(r)
 	if err != nil {
 		return nil, err
 	}

+	ggml.model = model
+
 	// final model type
-	return &GGML{
-		container: c,
-		model:     model,
-		Size:      ro.offset,
-	}, nil
-}
-
-type readSeekOffset struct {
-	io.ReadSeeker
-	offset int64
-}
-
-func (rso *readSeekOffset) Seek(offset int64, whence int) (int64, error) {
-	offset, err := rso.ReadSeeker.Seek(offset, whence)
-	if err != nil {
-		return 0, err
-	}
-
-	rso.offset = offset
-	return offset, nil
-}
-
-func (rso *readSeekOffset) Read(p []byte) (int, error) {
-	n, err := rso.ReadSeeker.Read(p)
-	rso.offset += int64(n)
-	return n, err
+	return &ggml, nil
 }
--- a/llm/gguf.go
+++ b/llm/gguf.go
@@ -23,24 +23,26 @@ type containerGGUF struct {
 		NumTensor uint64
 		NumKV     uint64
 	}
+
+	parameters uint64
 }

 func (c *containerGGUF) Name() string {
 	return "gguf"
 }

-func (c *containerGGUF) Decode(rso *readSeekOffset) (model, error) {
-	binary.Read(rso, c.bo, &c.Version)
+func (c *containerGGUF) Decode(r io.Reader) (model, error) {
+	binary.Read(r, c.bo, &c.Version)

 	switch c.Version {
 	case 1:
-		binary.Read(rso, c.bo, &c.V1)
+		binary.Read(r, c.bo, &c.V1)
 	default:
-		binary.Read(rso, c.bo, &c.V2)
+		binary.Read(r, c.bo, &c.V2)
 	}

 	model := newGGUFModel(c)
-	if err := model.Decode(rso); err != nil {
+	if err := model.Decode(r); err != nil {
 		return nil, err
 	}

@@ -65,23 +67,9 @@ const (

 type kv map[string]any

-type tensor struct {
-	name   string
-	kind   uint32
-	offset uint64
-	size   uint64
-
-	// shape is the number of elements in each dimension
-	shape [4]uint64
-}
-
 type ggufModel struct {
 	*containerGGUF
-
 	kv
-	tensors []tensor
-
-	parameters uint64
 }

 func newGGUFModel(container *containerGGUF) *ggufModel {
@@ -108,7 +96,8 @@ func (llm *ggufModel) NumKV() uint64 {
 }

 func (llm *ggufModel) ModelFamily() string {
-	if t, ok := llm.kv["general.architecture"].(string); ok {
+	t, ok := llm.kv["general.architecture"].(string)
+	if ok {
 		return t
 	}

@@ -120,60 +109,82 @@ func (llm *ggufModel) ModelType() string {
 		return format.HumanNumber(llm.parameters)
 	}

+	switch llm.ModelFamily() {
+	case "llama":
+		if blocks, ok := llm.kv["llama.block_count"].(uint32); ok {
+			heads, headsOK := llm.kv["llama.head_count"].(uint32)
+			headKVs, headsKVsOK := llm.kv["llama.head_count_kv"].(uint32)
+			if headsOK && headsKVsOK && heads/headKVs == 8 {
+				return "70B"
+			}
+
+			return llamaModelType(blocks)
+		}
+	case "falcon":
+		if blocks, ok := llm.kv["falcon.block_count"].(uint32); ok {
+			return falconModelType(blocks)
+		}
+	case "starcoder":
+		if blocks, ok := llm.kv["starcoder.block_count"].(uint32); ok {
+			return starCoderModelType(blocks)
+		}
+	}
+
 	return "unknown"
 }

 func (llm *ggufModel) FileType() string {
-	if t, ok := llm.kv["general.file_type"].(uint32); ok {
+	t, ok := llm.kv["general.file_type"].(uint32)
+	if ok {
 		return fileType(t)
 	}

 	return "unknown"
 }

-func (llm *ggufModel) Decode(rso *readSeekOffset) error {
+func (llm *ggufModel) Decode(r io.Reader) error {
 	// decode key-values
 	for i := 0; uint64(i) < llm.NumKV(); i++ {
-		k, err := llm.readString(rso)
+		k, err := llm.readString(r)
 		if err != nil {
 			return err
 		}

-		vtype := llm.readU32(rso)
+		vtype := llm.readU32(r)

 		var v any
 		switch vtype {
 		case ggufTypeUint8:
-			v = llm.readU8(rso)
+			v = llm.readU8(r)
 		case ggufTypeInt8:
-			v = llm.readI8(rso)
+			v = llm.readI8(r)
 		case ggufTypeUint16:
-			v = llm.readU16(rso)
+			v = llm.readU16(r)
 		case ggufTypeInt16:
-			v = llm.readI16(rso)
+			v = llm.readI16(r)
 		case ggufTypeUint32:
-			v = llm.readU32(rso)
+			v = llm.readU32(r)
 		case ggufTypeInt32:
-			v = llm.readI32(rso)
+			v = llm.readI32(r)
 		case ggufTypeUint64:
-			v = llm.readU64(rso)
+			v = llm.readU64(r)
 		case ggufTypeInt64:
-			v = llm.readI64(rso)
+			v = llm.readI64(r)
 		case ggufTypeFloat32:
-			v = llm.readF32(rso)
+			v = llm.readF32(r)
 		case ggufTypeFloat64:
-			v = llm.readF64(rso)
+			v = llm.readF64(r)
 		case ggufTypeBool:
-			v = llm.readBool(rso)
+			v = llm.readBool(r)
 		case ggufTypeString:
-			s, err := llm.readString(rso)
+			s, err := llm.readString(r)
 			if err != nil {
 				return err
 			}

 			v = s
 		case ggufTypeArray:
-			a, err := llm.readArray(rso)
+			a, err := llm.readArray(r)
 			if err != nil {
 				return err
 			}
@@ -188,85 +199,21 @@ func (llm *ggufModel) Decode(rso *readSeekOffset) error {

 	// decode tensors
 	for i := 0; uint64(i) < llm.NumTensor(); i++ {
-		name, err := llm.readString(rso)
-		if err != nil {
+		if _, err := llm.readString(r); err != nil {
 			return err
 		}

-		// dims is the number of dimensions in the tensor
-		dims := llm.readU32(rso)
+		dimensions := llm.readU32(r)

-		shape := [4]uint64{1, 1, 1, 1}
-		for i := 0; uint32(i) < dims; i++ {
-			shape[i] = llm.readU64(rso)
+		var elements uint64 = 1
+		for i := 0; uint32(i) < dimensions; i++ {
+			elements *= llm.readU64(r)
 		}

-		kind := llm.readU32(rso)
-		offset := llm.readU64(rso)
+		llm.readU32(r) // type
+		llm.readU64(r) // offset

-		var blockSize uint64
-		switch {
-		case kind < 2:
-			blockSize = 1
-		case kind < 10:
-			blockSize = 32
-		default:
-			blockSize = 256
-		}
-
-		var typeSize uint64
-		switch kind {
-		case 0: // FP32
-			typeSize = 4
-		case 1: // FP16
-			typeSize = 2
-		case 2: // Q4_0
-			typeSize = 2 + blockSize/2
-		case 3: // Q4_1
-			typeSize = 2 + 2 + blockSize/2
-		case 6: // Q5_0
-			typeSize = 2 + 4 + blockSize/2
-		case 7: // Q5_1
-			typeSize = 2 + 2 + 4 + blockSize/2
-		case 8: // Q8_0
-			typeSize = 2 + blockSize
-		case 9: // Q8_1
-			typeSize = 4 + 4 + blockSize
-		case 10: // Q2_K
-			typeSize = blockSize/16 + blockSize/4 + 2 + 2
-		case 11: // Q3_K
-			typeSize = blockSize/8 + blockSize/4 + 12 + 2
-		case 12: // Q4_K
-			typeSize = 2 + 2 + 12 + blockSize/2
-		case 13: // Q5_K
-			typeSize = 2 + 2 + 12 + blockSize/8 + blockSize/2
-		case 14: // Q6_K
-			typeSize = blockSize/2 + blockSize/4 + blockSize/16 + 2
-		}
-
-		parameters := shape[0] * shape[1] * shape[2] * shape[3]
-		size := parameters * typeSize / blockSize
-
-		llm.tensors = append(llm.tensors, tensor{
-			name:   name,
-			kind:   kind,
-			offset: offset,
-			size:   size,
-			shape:  shape,
-		})
-
-		llm.parameters += parameters
-	}
-
-	alignment, ok := llm.kv["general.alignment"].(uint32)
-	if !ok {
-		alignment = 32
-	}
-
-	rso.Seek(int64(alignment)-rso.offset%int64(alignment), io.SeekCurrent)
-	for _, tensor := range llm.tensors {
-		padded := (int64(tensor.size) + int64(alignment) - 1) & ^(int64(alignment) - 1)
-		rso.Seek(padded, io.SeekCurrent)
+		llm.parameters += elements
 	}

 	return nil
--- a/llm/llama.cpp
+++ b/llm/llama.cpp
--- a/llm/llama.cpp/generate_darwin_amd64.go
+++ b/llm/llama.cpp/generate_darwin_amd64.go
@@ -0,0 +1,18 @@
+package llm
+
+//go:generate git submodule init
+
+//go:generate git submodule update --force ggml
+//go:generate git -C ggml apply ../patches/0001-add-detokenize-endpoint.patch
+//go:generate git -C ggml apply ../patches/0002-34B-model-support.patch
+//go:generate git -C ggml apply ../patches/0003-metal-fix-synchronization-in-new-matrix-multiplicati.patch
+//go:generate git -C ggml apply ../patches/0004-metal-add-missing-barriers-for-mul-mat-2699.patch
+//go:generate cmake -S ggml -B ggml/build/cpu -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_NAME=Darwin -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
+//go:generate cmake --build ggml/build/cpu --target server --config Release
+//go:generate mv ggml/build/cpu/bin/server ggml/build/cpu/bin/ollama-runner
+
+//go:generate git submodule update --force gguf
+//go:generate git -C gguf apply ../patches/0001-update-default-log-target.patch
+//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_METAL=off -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_NAME=Darwin -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0 -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=on
+//go:generate cmake --build gguf/build/cpu --target server --config Release
+//go:generate mv gguf/build/cpu/bin/server gguf/build/cpu/bin/ollama-runner
--- a/llm/llama.cpp/generate_darwin_arm64.go
+++ b/llm/llama.cpp/generate_darwin_arm64.go
@@ -0,0 +1,18 @@
+package llm
+
+//go:generate git submodule init
+
+//go:generate git submodule update --force ggml
+//go:generate git -C ggml apply ../patches/0001-add-detokenize-endpoint.patch
+//go:generate git -C ggml apply ../patches/0002-34B-model-support.patch
+//go:generate git -C ggml apply ../patches/0003-metal-fix-synchronization-in-new-matrix-multiplicati.patch
+//go:generate git -C ggml apply ../patches/0004-metal-add-missing-barriers-for-mul-mat-2699.patch
+//go:generate cmake -S ggml -B ggml/build/metal -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
+//go:generate cmake --build ggml/build/metal --target server --config Release
+//go:generate mv ggml/build/metal/bin/server ggml/build/metal/bin/ollama-runner
+
+//go:generate git submodule update --force gguf
+//go:generate git -C gguf apply ../patches/0001-update-default-log-target.patch
+//go:generate cmake -S gguf -B gguf/build/metal -DLLAMA_METAL=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=arm64 -DCMAKE_OSX_ARCHITECTURES=arm64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
+//go:generate cmake --build gguf/build/metal --target server --config Release
+//go:generate mv gguf/build/metal/bin/server gguf/build/metal/bin/ollama-runner
--- a/llm/llama.cpp/generate_linux.go
+++ b/llm/llama.cpp/generate_linux.go
@@ -0,0 +1,26 @@
+package llm
+
+//go:generate git submodule init
+
+//go:generate git submodule update --force ggml
+//go:generate git -C ggml apply ../patches/0001-add-detokenize-endpoint.patch
+//go:generate git -C ggml apply ../patches/0002-34B-model-support.patch
+//go:generate git -C ggml apply ../patches/0005-ggml-support-CUDA-s-half-type-for-aarch64-1455-2670.patch
+//go:generate git -C ggml apply ../patches/0001-copy-cuda-runtime-libraries.patch
+//go:generate cmake -S ggml -B ggml/build/cpu -DLLAMA_K_QUANTS=on
+//go:generate cmake --build ggml/build/cpu --target server --config Release
+//go:generate mv ggml/build/cpu/bin/server ggml/build/cpu/bin/ollama-runner
+
+//go:generate git submodule update --force gguf
+//go:generate git -C gguf apply ../patches/0001-copy-cuda-runtime-libraries.patch
+//go:generate git -C gguf apply ../patches/0001-update-default-log-target.patch
+//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off
+//go:generate cmake --build gguf/build/cpu --target server --config Release
+//go:generate mv gguf/build/cpu/bin/server gguf/build/cpu/bin/ollama-runner
+
+//go:generate cmake -S ggml -B ggml/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
+//go:generate cmake --build ggml/build/cuda --target server --config Release
+//go:generate mv ggml/build/cuda/bin/server ggml/build/cuda/bin/ollama-runner
+//go:generate cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off -DLLAMA_CUDA_PEER_MAX_BATCH_SIZE=0
+//go:generate cmake --build gguf/build/cuda --target server --config Release
+//go:generate mv gguf/build/cuda/bin/server gguf/build/cuda/bin/ollama-runner
--- a/llm/llama.cpp/generate_windows.go
+++ b/llm/llama.cpp/generate_windows.go
@@ -0,0 +1,24 @@
+package llm
+
+//go:generate git submodule init
+
+//go:generate git submodule update --force ggml
+//go:generate git -C ggml apply ../patches/0001-add-detokenize-endpoint.patch
+//go:generate git -C ggml apply ../patches/0002-34B-model-support.patch
+//go:generate cmake -S ggml -B ggml/build/cpu -DLLAMA_K_QUANTS=on
+//go:generate cmake --build ggml/build/cpu --target server --config Release
+//go:generate cmd /c move ggml\build\cpu\bin\Release\server.exe ggml\build\cpu\bin\Release\ollama-runner.exe
+
+//go:generate git submodule update --force gguf
+//go:generate git -C gguf apply ../patches/0001-update-default-log-target.patch
+//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off
+//go:generate cmake --build gguf/build/cpu --target server --config Release
+//go:generate cmd /c move gguf\build\cpu\bin\Release\server.exe gguf\build\cpu\bin\Release\ollama-runner.exe
+
+//go:generate cmake -S ggml -B ggml/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
+//go:generate cmake --build ggml/build/cuda --target server --config Release
+//go:generate cmd /c move ggml\build\cuda\bin\Release\server.exe ggml\build\cuda\bin\Release\ollama-runner.exe
+
+//go:generate cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off
+//go:generate cmake --build gguf/build/cuda --target server --config Release
+//go:generate cmd /c move gguf\build\cuda\bin\Release\server.exe gguf\build\cuda\bin\Release\ollama-runner.exe
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Matt Williams	05162c56aa	Update readme.md	2023-11-29 10:45:07 -08:00
Matt Williams	edd1a2b6e8	function calling for python. already had ts. Signed-off-by: Matt Williams <m@technovangelist.com>	2023-11-29 10:06:11 -08:00