JSON mode: add `"format" as an api parameter (#1051)

* add `"format": "json"` as an API parameter
---------
Co-authored-by: Bruce MacDonald <brucewmacdonald@gmail.com>

Dockerfile

@@ -1,16 +1,16 @@
 FROM nvidia/cuda:11.8.0-devel-ubuntu22.04
 
 ARG TARGETARCH
-ARG VERSION=0.0.0
 ARG GOFLAGS="'-ldflags=-w -s'"
 
 WORKDIR /go/src/github.com/jmorganca/ollama
 RUN apt-get update && apt-get install -y git build-essential cmake
-ADD https://dl.google.com/go/go1.21.1.linux-$TARGETARCH.tar.gz /tmp/go1.21.1.tar.gz
-RUN mkdir -p /usr/local && tar xz -C /usr/local </tmp/go1.21.1.tar.gz
+ADD https://dl.google.com/go/go1.21.3.linux-$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
 
 COPY . .
 ENV GOARCH=$TARGETARCH
+ENV GOFLAGS=$GOFLAGS
 RUN /usr/local/go/bin/go generate ./... \
     && /usr/local/go/bin/go build .
 

Dockerfile.build

@@ -1,6 +1,5 @@
-
 # centos7 amd64 dependencies
-FROM --platform=linux/amd64 nvidia/cuda:11.8.0-devel-centos7 AS base-amd64
+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
@@ -8,25 +7,25 @@ RUN wget "https://github.com/Kitware/CMake/releases/download/v3.27.6/cmake-3.27.
 ENV PATH /opt/rh/devtoolset-10/root/usr/bin:$PATH
 
 # centos8 arm64 dependencies
-FROM --platform=linux/arm64 nvidia/cuda:11.4.3-devel-centos8 AS base-arm64
+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/go1.21.1.linux-$TARGETARCH.tar.gz /tmp/go1.21.1.tar.gz
-RUN mkdir -p /usr/local && tar xz -C /usr/local </tmp/go1.21.1.tar.gz
+ADD https://dl.google.com/go/go1.21.3.linux-$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 . .
+
 ENV GOOS=linux
 ENV GOARCH=$TARGETARCH
-
-ARG VERSION=0.0.0
-ARG GOFLAGS="'-ldflags -w -s'"
+ENV GOFLAGS=$GOFLAGS
 
 RUN /usr/local/go/bin/go generate ./... && \
     /usr/local/go/bin/go build .

README.md

@@ -13,7 +13,11 @@ Get up and running with large language models locally.
 
 ### macOS
 
-[Download](https://ollama.ai/download/Ollama-darwin.zip) 
+[Download](https://ollama.ai/download/Ollama-darwin.zip)
+
+### Windows
+
+Coming soon!
 
 ### Linux & WSL2
 
@@ -23,9 +27,9 @@ curl https://ollama.ai/install.sh | sh
 
 [Manual install instructions](https://github.com/jmorganca/ollama/blob/main/docs/linux.md)
 
-### Windows 
+### Docker
 
-coming soon
+The official [Ollama Docker image](https://hub.docker.com/r/ollama/ollama) `ollama/ollama` is available on Docker Hub.
 
 ## Quickstart
 
@@ -37,7 +41,7 @@ ollama run llama2
 
 ## Model library
 
-Ollama supports a list of open-source models available on [ollama.ai/library](https://ollama.ai/library "ollama model library")
+Ollama supports a list of open-source models available on [ollama.ai/library](https://ollama.ai/library 'ollama model library')
 
 Here are some example open-source models that can be downloaded:
 
@@ -56,31 +60,35 @@ Here are some example open-source models that can be downloaded:
 
 ## Customize your own model
 
-### Import from GGUF or GGML
+### Import from GGUF
 
-Ollama supports importing GGUF and GGML file formats in the Modelfile. This means if you have a model that is not in the Ollama library, you can create it, iterate on it, and upload it to the Ollama library to share with others when you are ready.
+Ollama supports importing GGUF models in the Modelfile:
 
-1. Create a file named Modelfile, and add a `FROM` instruction with the local filepath to the model you want to import.
+1. Create a file named `Modelfile`, with a `FROM` instruction with the local filepath to the model you want to import.
 
    ```
    FROM ./vicuna-33b.Q4_0.gguf
    ```
 
-3. Create the model in Ollama
+2. Create the model in Ollama
 
    ```
-   ollama create name -f path_to_modelfile
+   ollama create example -f Modelfile
    ```
 
-5. Run the model
+3. Run the model
 
    ```
-   ollama run name
+   ollama run example
    ```
 
+### Import from PyTorch or Safetensors
+
+See the [guide](docs/import.md) on importing models for more information.
+
 ### Customize a prompt
 
-Models from the Ollama library can be customized with a prompt. The example
+Models from the Ollama library can be customized with a prompt. For example, to customize the `llama2` model:
 
 ```
 ollama pull llama2
@@ -109,7 +117,7 @@ ollama run mario
 Hello! It's your friend Mario.
 ```
 
-For more examples, see the [examples](./examples) directory. For more information on working with a Modelfile, see the [Modelfile](./docs/modelfile.md) documentation.
+For more examples, see the [examples](examples) directory. For more information on working with a Modelfile, see the [Modelfile](docs/modelfile.md) documentation.
 
 ## CLI Reference
 
@@ -151,7 +159,7 @@ I'm a basic program that prints the famous "Hello, world!" message to the consol
 ### Pass in prompt as arguments
 
 ```
-$ ollama run llama2 "summarize this file:" "$(cat README.md)"
+$ ollama run llama2 "Summarize this file: $(cat README.md)"
  Ollama is a lightweight, extensible framework for building and running language models on the local machine. It provides a simple API for creating, running, and managing models, as well as a library of pre-built models that can be easily used in a variety of applications.
 ```
 
@@ -170,8 +178,7 @@ ollama list
 Install `cmake` and `go`:
 
 ```
-brew install cmake
-brew install go
+brew install cmake go
 ```
 
 Then generate dependencies and build:
@@ -195,9 +202,8 @@ Finally, in a separate shell, run a model:
 
 ## REST API
 
-> See the [API documentation](./docs/api.md) for all endpoints.
-
-Ollama has an API for running and managing models. For example to generate text from a model:
+Ollama has a REST API for running and managing models.
+For example, to generate text from a model:
 
 ```
 curl -X POST http://localhost:11434/api/generate -d '{
@@ -206,18 +212,44 @@ curl -X POST http://localhost:11434/api/generate -d '{
 }'
 ```
 
+See the [API documentation](./docs/api.md) for all endpoints.
+
 ## Community Integrations
 
+### Web & Desktop
+
+- [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)
+- [Minimalistic React UI for Ollama Models](https://github.com/richawo/minimal-llm-ui)
+- [Web UI](https://github.com/ollama-webui/ollama-webui)
+- [Ollamac](https://github.com/kevinhermawan/Ollamac)
+
+### Terminal
+
+- [oterm](https://github.com/ggozad/oterm)
+- [Ellama Emacs client](https://github.com/s-kostyaev/ellama)
+- [Emacs client](https://github.com/zweifisch/ollama)
+- [gen.nvim](https://github.com/David-Kunz/gen.nvim)
+- [gptel Emacs client](https://github.com/karthink/gptel)
+
+### Libraries
+
 - [LangChain](https://python.langchain.com/docs/integrations/llms/ollama) and [LangChain.js](https://js.langchain.com/docs/modules/model_io/models/llms/integrations/ollama) with [example](https://js.langchain.com/docs/use_cases/question_answering/local_retrieval_qa)
 - [LlamaIndex](https://gpt-index.readthedocs.io/en/stable/examples/llm/ollama.html)
+- [LiteLLM](https://github.com/BerriAI/litellm)
+- [OllamaSharp for .NET](https://github.com/awaescher/OllamaSharp)
+- [Ollama-rs for Rust](https://github.com/pepperoni21/ollama-rs)
+- [Ollama4j for Java](https://github.com/amithkoujalgi/ollama4j)
+- [ModelFusion Typescript Library](https://modelfusion.dev/integration/model-provider/ollama)
+
+### Extensions & Plugins
+
 - [Raycast extension](https://github.com/MassimilianoPasquini97/raycast_ollama)
 - [Discollama](https://github.com/mxyng/discollama) (Discord bot inside the Ollama discord channel)
 - [Continue](https://github.com/continuedev/continue)
 - [Obsidian Ollama plugin](https://github.com/hinterdupfinger/obsidian-ollama)
+- [Logseq Ollama plugin](https://github.com/omagdy7/ollama-logseq)
 - [Dagger Chatbot](https://github.com/samalba/dagger-chatbot)
-- [LiteLLM](https://github.com/BerriAI/litellm)
 - [Discord AI Bot](https://github.com/mekb-turtle/discord-ai-bot)
-- [HTML UI](https://github.com/rtcfirefly/ollama-ui)
-- [Typescript UI](https://github.com/ollama-interface/Ollama-Gui?tab=readme-ov-file)
-- [Dumbar](https://github.com/JerrySievert/Dumbar)
-- [Emacs client](https://github.com/zweifisch/ollama)
+- [Hass Ollama Conversation](https://github.com/ej52/hass-ollama-conversation)

api/client.go

@@ -7,25 +7,20 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
+	"net"
 	"net/http"
 	"net/url"
 	"os"
 	"runtime"
 	"strings"
 
+	"github.com/jmorganca/ollama/format"
 	"github.com/jmorganca/ollama/version"
 )
 
-const DefaultHost = "127.0.0.1:11434"
-
-var (
-	envHost = os.Getenv("OLLAMA_HOST")
-)
-
 type Client struct {
-	Base    url.URL
-	HTTP    http.Client
-	Headers http.Header
+	base *url.URL
+	http http.Client
 }
 
 func checkError(resp *http.Response, body []byte) error {
@@ -44,34 +39,56 @@ func checkError(resp *http.Response, body []byte) error {
 	return apiError
 }
 
-// Host returns the default host to use for the client. It is determined in the following order:
-// 1. The OLLAMA_HOST environment variable
-// 2. The default host (localhost:11434)
-func Host() string {
-	if envHost != "" {
-		return envHost
-	}
-	return DefaultHost
-}
+func ClientFromEnvironment() (*Client, error) {
+	defaultPort := "11434"
 
-// FromEnv creates a new client using Host() as the host. An error is returns
-// if the host is invalid.
-func FromEnv() (*Client, error) {
-	h := Host()
-	if !strings.HasPrefix(h, "http://") && !strings.HasPrefix(h, "https://") {
-		h = "http://" + h
+	scheme, hostport, ok := strings.Cut(os.Getenv("OLLAMA_HOST"), "://")
+	switch {
+	case !ok:
+		scheme, hostport = "http", os.Getenv("OLLAMA_HOST")
+	case scheme == "http":
+		defaultPort = "80"
+	case scheme == "https":
+		defaultPort = "443"
 	}
 
-	u, err := url.Parse(h)
+	// trim trailing slashes
+	hostport = strings.TrimRight(hostport, "/")
+
+	host, port, err := net.SplitHostPort(hostport)
 	if err != nil {
-		return nil, fmt.Errorf("could not parse host: %w", err)
+		host, port = "127.0.0.1", defaultPort
+		if ip := net.ParseIP(strings.Trim(hostport, "[]")); ip != nil {
+			host = ip.String()
+		} else if hostport != "" {
+			host = hostport
+		}
 	}
 
-	if u.Port() == "" {
-		u.Host += ":11434"
+	client := Client{
+		base: &url.URL{
+			Scheme: scheme,
+			Host:   net.JoinHostPort(host, port),
+		},
 	}
 
-	return &Client{Base: *u, HTTP: http.Client{}}, nil
+	mockRequest, err := http.NewRequest(http.MethodHead, client.base.String(), nil)
+	if err != nil {
+		return nil, err
+	}
+
+	proxyURL, err := http.ProxyFromEnvironment(mockRequest)
+	if err != nil {
+		return nil, err
+	}
+
+	client.http = http.Client{
+		Transport: &http.Transport{
+			Proxy: http.ProxyURL(proxyURL),
+		},
+	}
+
+	return &client, nil
 }
 
 func (c *Client) do(ctx context.Context, method, path string, reqData, respData any) error {
@@ -86,7 +103,7 @@ func (c *Client) do(ctx context.Context, method, path string, reqData, respData
 		reqBody = bytes.NewReader(data)
 	}
 
-	requestURL := c.Base.JoinPath(path)
+	requestURL := c.base.JoinPath(path)
 	request, err := http.NewRequestWithContext(ctx, method, requestURL.String(), reqBody)
 	if err != nil {
 		return err
@@ -96,11 +113,7 @@ func (c *Client) do(ctx context.Context, method, path string, reqData, respData
 	request.Header.Set("Accept", "application/json")
 	request.Header.Set("User-Agent", fmt.Sprintf("ollama/%s (%s %s) Go/%s", version.Version, runtime.GOARCH, runtime.GOOS, runtime.Version()))
 
-	for k, v := range c.Headers {
-		request.Header[k] = v
-	}
-
-	respObj, err := c.HTTP.Do(request)
+	respObj, err := c.http.Do(request)
 	if err != nil {
 		return err
 	}
@@ -123,6 +136,8 @@ func (c *Client) do(ctx context.Context, method, path string, reqData, respData
 	return nil
 }
 
+const maxBufferSize = 512 * format.KiloByte
+
 func (c *Client) stream(ctx context.Context, method, path string, data any, fn func([]byte) error) error {
 	var buf *bytes.Buffer
 	if data != nil {
@@ -134,23 +149,26 @@ func (c *Client) stream(ctx context.Context, method, path string, data any, fn f
 		buf = bytes.NewBuffer(bts)
 	}
 
-	requestURL := c.Base.JoinPath(path)
+	requestURL := c.base.JoinPath(path)
 	request, err := http.NewRequestWithContext(ctx, method, requestURL.String(), buf)
 	if err != nil {
 		return err
 	}
 
 	request.Header.Set("Content-Type", "application/json")
-	request.Header.Set("Accept", "application/json")
+	request.Header.Set("Accept", "application/x-ndjson")
 	request.Header.Set("User-Agent", fmt.Sprintf("ollama/%s (%s %s) Go/%s", version.Version, runtime.GOARCH, runtime.GOOS, runtime.Version()))
 
-	response, err := http.DefaultClient.Do(request)
+	response, err := c.http.Do(request)
 	if err != nil {
 		return err
 	}
 	defer response.Body.Close()
 
 	scanner := bufio.NewScanner(response.Body)
+	// increase the buffer size to avoid running out of space
+	scanBuf := make([]byte, 0, maxBufferSize)
+	scanner.Buffer(scanBuf, maxBufferSize)
 	for scanner.Scan() {
 		var errorResponse struct {
 			Error string `json:"error,omitempty"`

api/client_test.go

@@ -0,0 +1,43 @@
+package api
+
+import "testing"
+
+func TestClientFromEnvironment(t *testing.T) {
+	type testCase struct {
+		value  string
+		expect string
+		err    error
+	}
+
+	testCases := map[string]*testCase{
+		"empty":                      {value: "", expect: "http://127.0.0.1:11434"},
+		"only address":               {value: "1.2.3.4", expect: "http://1.2.3.4:11434"},
+		"only port":                  {value: ":1234", expect: "http://:1234"},
+		"address and port":           {value: "1.2.3.4:1234", expect: "http://1.2.3.4:1234"},
+		"scheme http and address":    {value: "http://1.2.3.4", expect: "http://1.2.3.4:80"},
+		"scheme https and address":   {value: "https://1.2.3.4", expect: "https://1.2.3.4:443"},
+		"scheme, address, and port":  {value: "https://1.2.3.4:1234", expect: "https://1.2.3.4:1234"},
+		"hostname":                   {value: "example.com", expect: "http://example.com:11434"},
+		"hostname and port":          {value: "example.com:1234", expect: "http://example.com:1234"},
+		"scheme http and hostname":   {value: "http://example.com", expect: "http://example.com:80"},
+		"scheme https and hostname":  {value: "https://example.com", expect: "https://example.com:443"},
+		"scheme, hostname, and port": {value: "https://example.com:1234", expect: "https://example.com:1234"},
+		"trailing slash":             {value: "example.com/", expect: "http://example.com:11434"},
+		"trailing slash port":        {value: "example.com:1234/", expect: "http://example.com:1234"},
+	}
+
+	for k, v := range testCases {
+		t.Run(k, func(t *testing.T) {
+			t.Setenv("OLLAMA_HOST", v.value)
+
+			client, err := ClientFromEnvironment()
+			if err != v.err {
+				t.Fatalf("expected %s, got %s", v.err, err)
+			}
+
+			if client.base.String() != v.expect {
+				t.Fatalf("expected %s, got %s", v.expect, client.base.String())
+			}
+		})
+	}
+}

api/types.go

@@ -3,7 +3,6 @@ package api
 import (
 	"encoding/json"
 	"fmt"
-	"log"
 	"math"
 	"os"
 	"reflect"
@@ -37,10 +36,57 @@ type GenerateRequest struct {
 	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"`
 }
 
+// Options specfied in GenerateRequest, if you add a new option here add it to the API docs also
+type Options struct {
+	Runner
+
+	// Predict options used at runtime
+	NumKeep          int      `json:"num_keep,omitempty"`
+	Seed             int      `json:"seed,omitempty"`
+	NumPredict       int      `json:"num_predict,omitempty"`
+	TopK             int      `json:"top_k,omitempty"`
+	TopP             float32  `json:"top_p,omitempty"`
+	TFSZ             float32  `json:"tfs_z,omitempty"`
+	TypicalP         float32  `json:"typical_p,omitempty"`
+	RepeatLastN      int      `json:"repeat_last_n,omitempty"`
+	Temperature      float32  `json:"temperature,omitempty"`
+	RepeatPenalty    float32  `json:"repeat_penalty,omitempty"`
+	PresencePenalty  float32  `json:"presence_penalty,omitempty"`
+	FrequencyPenalty float32  `json:"frequency_penalty,omitempty"`
+	Mirostat         int      `json:"mirostat,omitempty"`
+	MirostatTau      float32  `json:"mirostat_tau,omitempty"`
+	MirostatEta      float32  `json:"mirostat_eta,omitempty"`
+	PenalizeNewline  bool     `json:"penalize_newline,omitempty"`
+	Stop             []string `json:"stop,omitempty"`
+}
+
+// Runner options which must be set when the model is loaded into memory
+type Runner struct {
+	UseNUMA            bool    `json:"numa,omitempty"`
+	NumCtx             int     `json:"num_ctx,omitempty"`
+	NumBatch           int     `json:"num_batch,omitempty"`
+	NumGQA             int     `json:"num_gqa,omitempty"`
+	NumGPU             int     `json:"num_gpu,omitempty"`
+	MainGPU            int     `json:"main_gpu,omitempty"`
+	LowVRAM            bool    `json:"low_vram,omitempty"`
+	F16KV              bool    `json:"f16_kv,omitempty"`
+	LogitsAll          bool    `json:"logits_all,omitempty"`
+	VocabOnly          bool    `json:"vocab_only,omitempty"`
+	UseMMap            bool    `json:"use_mmap,omitempty"`
+	UseMLock           bool    `json:"use_mlock,omitempty"`
+	EmbeddingOnly      bool    `json:"embedding_only,omitempty"`
+	RopeFrequencyBase  float32 `json:"rope_frequency_base,omitempty"`
+	RopeFrequencyScale float32 `json:"rope_frequency_scale,omitempty"`
+	NumThread          int     `json:"num_thread,omitempty"`
+}
+
 type EmbeddingRequest struct {
 	Model  string `json:"model"`
 	Prompt string `json:"prompt"`
@@ -53,8 +99,9 @@ type EmbeddingResponse struct {
 }
 
 type CreateRequest struct {
-	Name string `json:"name"`
-	Path string `json:"path"`
+	Name   string `json:"name"`
+	Path   string `json:"path"`
+	Stream *bool  `json:"stream,omitempty"`
 }
 
 type DeleteRequest struct {
@@ -81,6 +128,9 @@ type CopyRequest struct {
 type PullRequest struct {
 	Name     string `json:"name"`
 	Insecure bool   `json:"insecure,omitempty"`
+	Username string `json:"username"`
+	Password string `json:"password"`
+	Stream   *bool  `json:"stream,omitempty"`
 }
 
 type ProgressResponse struct {
@@ -93,6 +143,9 @@ type ProgressResponse struct {
 type PushRequest struct {
 	Name     string `json:"name"`
 	Insecure bool   `json:"insecure,omitempty"`
+	Username string `json:"username"`
+	Password string `json:"password"`
+	Stream   *bool  `json:"stream,omitempty"`
 }
 
 type ListResponse struct {
@@ -113,7 +166,7 @@ type TokenResponse struct {
 type GenerateResponse struct {
 	Model     string    `json:"model"`
 	CreatedAt time.Time `json:"created_at"`
-	Response  string    `json:"response,omitempty"`
+	Response  string    `json:"response"`
 
 	Done    bool  `json:"done"`
 	Context []int `json:"context,omitempty"`
@@ -154,48 +207,7 @@ func (r *GenerateResponse) Summary() {
 	}
 }
 
-type Options struct {
-	Seed int `json:"seed,omitempty"`
-
-	// Backend options
-	UseNUMA bool `json:"numa,omitempty"`
-
-	// Model options
-	NumCtx             int     `json:"num_ctx,omitempty"`
-	NumKeep            int     `json:"num_keep,omitempty"`
-	NumBatch           int     `json:"num_batch,omitempty"`
-	NumGQA             int     `json:"num_gqa,omitempty"`
-	NumGPU             int     `json:"num_gpu,omitempty"`
-	MainGPU            int     `json:"main_gpu,omitempty"`
-	LowVRAM            bool    `json:"low_vram,omitempty"`
-	F16KV              bool    `json:"f16_kv,omitempty"`
-	LogitsAll          bool    `json:"logits_all,omitempty"`
-	VocabOnly          bool    `json:"vocab_only,omitempty"`
-	UseMMap            bool    `json:"use_mmap,omitempty"`
-	UseMLock           bool    `json:"use_mlock,omitempty"`
-	EmbeddingOnly      bool    `json:"embedding_only,omitempty"`
-	RopeFrequencyBase  float32 `json:"rope_frequency_base,omitempty"`
-	RopeFrequencyScale float32 `json:"rope_frequency_scale,omitempty"`
-
-	// Predict options
-	NumPredict       int      `json:"num_predict,omitempty"`
-	TopK             int      `json:"top_k,omitempty"`
-	TopP             float32  `json:"top_p,omitempty"`
-	TFSZ             float32  `json:"tfs_z,omitempty"`
-	TypicalP         float32  `json:"typical_p,omitempty"`
-	RepeatLastN      int      `json:"repeat_last_n,omitempty"`
-	Temperature      float32  `json:"temperature,omitempty"`
-	RepeatPenalty    float32  `json:"repeat_penalty,omitempty"`
-	PresencePenalty  float32  `json:"presence_penalty,omitempty"`
-	FrequencyPenalty float32  `json:"frequency_penalty,omitempty"`
-	Mirostat         int      `json:"mirostat,omitempty"`
-	MirostatTau      float32  `json:"mirostat_tau,omitempty"`
-	MirostatEta      float32  `json:"mirostat_eta,omitempty"`
-	PenalizeNewline  bool     `json:"penalize_newline,omitempty"`
-	Stop             []string `json:"stop,omitempty"`
-
-	NumThread int `json:"num_thread,omitempty"`
-}
+var ErrInvalidOpts = fmt.Errorf("invalid options")
 
 func (opts *Options) FromMap(m map[string]interface{}) error {
 	valueOpts := reflect.ValueOf(opts).Elem() // names of the fields in the options struct
@@ -210,6 +222,7 @@ func (opts *Options) FromMap(m map[string]interface{}) error {
 		}
 	}
 
+	invalidOpts := []string{}
 	for key, val := range m {
 		if opt, ok := jsonOpts[key]; ok {
 			field := valueOpts.FieldByName(opt.Name)
@@ -227,44 +240,39 @@ func (opts *Options) FromMap(m map[string]interface{}) error {
 						// when JSON unmarshals numbers, it uses float64, not int
 						field.SetInt(int64(t))
 					default:
-						log.Printf("could not convert model parameter %v to int, skipped", key)
+						return fmt.Errorf("option %q must be of type integer", key)
 					}
 				case reflect.Bool:
 					val, ok := val.(bool)
 					if !ok {
-						log.Printf("could not convert model parameter %v to bool, skipped", key)
-						continue
+						return fmt.Errorf("option %q must be of type boolean", key)
 					}
 					field.SetBool(val)
 				case reflect.Float32:
 					// JSON unmarshals to float64
 					val, ok := val.(float64)
 					if !ok {
-						log.Printf("could not convert model parameter %v to float32, skipped", key)
-						continue
+						return fmt.Errorf("option %q must be of type float32", key)
 					}
 					field.SetFloat(val)
 				case reflect.String:
 					val, ok := val.(string)
 					if !ok {
-						log.Printf("could not convert model parameter %v to string, skipped", key)
-						continue
+						return fmt.Errorf("option %q must be of type string", key)
 					}
 					field.SetString(val)
 				case reflect.Slice:
 					// JSON unmarshals to []interface{}, not []string
 					val, ok := val.([]interface{})
 					if !ok {
-						log.Printf("could not convert model parameter %v to slice, skipped", key)
-						continue
+						return fmt.Errorf("option %q must be of type array", key)
 					}
 					// convert []interface{} to []string
 					slice := make([]string, len(val))
 					for i, item := range val {
 						str, ok := item.(string)
 						if !ok {
-							log.Printf("could not convert model parameter %v to slice of strings, skipped", key)
-							continue
+							return fmt.Errorf("option %q must be of an array of strings", key)
 						}
 						slice[i] = str
 					}
@@ -273,45 +281,53 @@ func (opts *Options) FromMap(m map[string]interface{}) error {
 					return fmt.Errorf("unknown type loading config params: %v", field.Kind())
 				}
 			}
+		} else {
+			invalidOpts = append(invalidOpts, key)
 		}
 	}
+
+	if len(invalidOpts) > 0 {
+		return fmt.Errorf("%w: %v", ErrInvalidOpts, strings.Join(invalidOpts, ", "))
+	}
 	return nil
 }
 
 func DefaultOptions() Options {
 	return Options{
-		Seed: -1,
-
-		UseNUMA: false,
-
-		NumCtx:             2048,
-		NumKeep:            -1,
-		NumBatch:           512,
-		NumGPU:             -1, // -1 here indicates that NumGPU should be set dynamically
-		NumGQA:             1,
-		LowVRAM:            false,
-		F16KV:              true,
-		UseMMap:            true,
-		UseMLock:           false,
-		RopeFrequencyBase:  10000.0,
-		RopeFrequencyScale: 1.0,
-		EmbeddingOnly:      true,
-
-		RepeatLastN:      64,
-		RepeatPenalty:    1.1,
-		FrequencyPenalty: 0.0,
-		PresencePenalty:  0.0,
+		// options set on request to runner
+		NumPredict:       -1,
+		NumKeep:          0,
 		Temperature:      0.8,
 		TopK:             40,
 		TopP:             0.9,
 		TFSZ:             1.0,
 		TypicalP:         1.0,
+		RepeatLastN:      64,
+		RepeatPenalty:    1.1,
+		PresencePenalty:  0.0,
+		FrequencyPenalty: 0.0,
 		Mirostat:         0,
 		MirostatTau:      5.0,
 		MirostatEta:      0.1,
 		PenalizeNewline:  true,
+		Seed:             -1,
 
-		NumThread: 0, // let the runtime decide
+		Runner: Runner{
+			// options set when the model is loaded
+			NumCtx:             2048,
+			RopeFrequencyBase:  10000.0,
+			RopeFrequencyScale: 1.0,
+			NumBatch:           512,
+			NumGPU:             -1, // -1 here indicates that NumGPU should be set dynamically
+			NumGQA:             1,
+			NumThread:          0, // let the runtime decide
+			LowVRAM:            false,
+			F16KV:              true,
+			UseMLock:           false,
+			UseMMap:            true,
+			UseNUMA:            false,
+			EmbeddingOnly:      true,
+		},
 	}
 }
 

app/forge.config.ts

@@ -47,16 +47,6 @@ const config: ForgeConfig = {
   },
   rebuildConfig: {},
   makers: [new MakerSquirrel({}), new MakerZIP({}, ['darwin'])],
-  publishers: [
-    new PublisherGithub({
-      repository: {
-        name: 'ollama',
-        owner: 'jmorganca',
-      },
-      draft: false,
-      prerelease: true,
-    }),
-  ],
   hooks: {
     readPackageJson: async (_, packageJson) => {
       return { ...packageJson, version: process.env.VERSION || packageJson.version }

app/package.json

@@ -46,7 +46,7 @@
     "chmodr": "^1.2.0",
     "copy-webpack-plugin": "^11.0.0",
     "css-loader": "^6.8.1",
-    "electron": "25.2.0",
+    "electron": "25.9.2",
     "eslint": "^8.43.0",
     "eslint-plugin-import": "^2.27.5",
     "fork-ts-checker-webpack-plugin": "^7.3.0",

app/src/index.ts

@@ -162,13 +162,56 @@ app.on('before-quit', () => {
   }
 })
 
+const updateURL = `https://ollama.ai/api/update?os=${process.platform}&arch=${
+  process.arch
+}&version=${app.getVersion()}&id=${id()}`
+
+let latest = ''
+async function isNewReleaseAvailable() {
+  try {
+    const response = await fetch(updateURL)
+
+    if (!response.ok) {
+      return false
+    }
+
+    if (response.status === 204) {
+      return false
+    }
+
+    const data = await response.json()
+
+    const url = data?.url
+    if (!url) {
+      return false
+    }
+
+    if (latest === url) {
+      return false
+    }
+
+    latest = url
+
+    return true
+  } catch (error) {
+    logger.error(`update check failed - ${error}`)
+    return false
+  }
+}
+
+async function checkUpdate() {
+  const available = await isNewReleaseAvailable()
+  if (available) {
+    logger.info('checking for update')
+    autoUpdater.checkForUpdates()
+  }
+}
+
 function init() {
   if (app.isPackaged) {
-    autoUpdater.checkForUpdates()
+    checkUpdate()
     setInterval(() => {
-      if (!updateAvailable) {
-        autoUpdater.checkForUpdates()
-      }
+      checkUpdate()
     }, 60 * 60 * 1000)
   }
 
@@ -246,11 +289,7 @@ function id(): string {
   return uuid
 }
 
-autoUpdater.setFeedURL({
-  url: `https://ollama.ai/api/update?os=${process.platform}&arch=${
-    process.arch
-  }&version=${app.getVersion()}&id=${id()}`,
-})
+autoUpdater.setFeedURL({ url: updateURL })
 
 autoUpdater.on('error', e => {
   logger.error(`update check failed - ${e.message}`)

cmd/cmd.go

@@ -11,6 +11,7 @@ import (
 	"io"
 	"log"
 	"net"
+	"net/http"
 	"os"
 	"os/exec"
 	"os/signal"
@@ -22,7 +23,6 @@ import (
 
 	"github.com/dustin/go-humanize"
 	"github.com/olekukonko/tablewriter"
-	"github.com/pdevine/readline"
 	"github.com/spf13/cobra"
 	"golang.org/x/crypto/ssh"
 	"golang.org/x/term"
@@ -30,30 +30,11 @@ import (
 	"github.com/jmorganca/ollama/api"
 	"github.com/jmorganca/ollama/format"
 	"github.com/jmorganca/ollama/progressbar"
+	"github.com/jmorganca/ollama/readline"
 	"github.com/jmorganca/ollama/server"
 	"github.com/jmorganca/ollama/version"
 )
 
-type Painter struct {
-	IsMultiLine bool
-}
-
-func (p Painter) Paint(line []rune, _ int) []rune {
-	termType := os.Getenv("TERM")
-	if termType == "xterm-256color" && len(line) == 0 {
-		var prompt string
-		if p.IsMultiLine {
-			prompt = "Use \"\"\" to end multi-line input"
-		} else {
-			prompt = "Send a message (/? for help)"
-		}
-		return []rune(fmt.Sprintf("\033[38;5;245m%s\033[%dD\033[0m", prompt, len(prompt)))
-	}
-	// add a space and a backspace to prevent the cursor from walking up the screen
-	line = append(line, []rune(" \b")...)
-	return line
-}
-
 func CreateHandler(cmd *cobra.Command, args []string) error {
 	filename, _ := cmd.Flags().GetString("file")
 	filename, err := filepath.Abs(filename)
@@ -61,7 +42,7 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -78,18 +59,12 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 				spinner.Stop()
 			}
 			currentDigest = resp.Digest
-			switch {
-			case strings.Contains(resp.Status, "embeddings"):
-				bar = progressbar.Default(resp.Total, resp.Status)
-				bar.Set64(resp.Completed)
-			default:
-				// pulling
-				bar = progressbar.DefaultBytes(
-					resp.Total,
-					resp.Status,
-				)
-				bar.Set64(resp.Completed)
-			}
+			// pulling
+			bar = progressbar.DefaultBytes(
+				resp.Total,
+				resp.Status,
+			)
+			bar.Set64(resp.Completed)
 		} else if resp.Digest == currentDigest && resp.Digest != "" {
 			bar.Set64(resp.Completed)
 		} else {
@@ -119,24 +94,21 @@ func CreateHandler(cmd *cobra.Command, args []string) error {
 }
 
 func RunHandler(cmd *cobra.Command, args []string) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
 
-	models, err := client.List(context.Background())
-	if err != nil {
-		return err
-	}
-
-	canonicalModelPath := server.ParseModelPath(args[0])
-	for _, model := range models.Models {
-		if model.Name == canonicalModelPath.GetShortTagname() {
-			return RunGenerate(cmd, args)
+	name := args[0]
+	// check if the model exists on the server
+	_, 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, args); err != nil {
+			return err
 		}
-	}
-
-	if err := PullHandler(cmd, args); err != nil {
+	case err != nil:
 		return err
 	}
 
@@ -144,7 +116,7 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 }
 
 func PushHandler(cmd *cobra.Command, args []string) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -188,7 +160,7 @@ func PushHandler(cmd *cobra.Command, args []string) error {
 }
 
 func ListHandler(cmd *cobra.Command, args []string) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -221,7 +193,7 @@ func ListHandler(cmd *cobra.Command, args []string) error {
 }
 
 func DeleteHandler(cmd *cobra.Command, args []string) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -237,7 +209,7 @@ func DeleteHandler(cmd *cobra.Command, args []string) error {
 }
 
 func ShowHandler(cmd *cobra.Command, args []string) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -315,7 +287,7 @@ func ShowHandler(cmd *cobra.Command, args []string) error {
 }
 
 func CopyHandler(cmd *cobra.Command, args []string) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -338,7 +310,7 @@ func PullHandler(cmd *cobra.Command, args []string) error {
 }
 
 func pull(model string, insecure bool) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -380,7 +352,20 @@ func pull(model string, insecure bool) error {
 func RunGenerate(cmd *cobra.Command, args []string) error {
 	if len(args) > 1 {
 		// join all args into a single prompt
-		return generate(cmd, args[0], strings.Join(args[1:], " "))
+		wordWrap := false
+		if term.IsTerminal(int(os.Stdout.Fd())) {
+			wordWrap = true
+		}
+
+		nowrap, err := cmd.Flags().GetBool("nowordwrap")
+		if err != nil {
+			return err
+		}
+		if nowrap {
+			wordWrap = false
+		}
+
+		return generate(cmd, args[0], strings.Join(args[1:], " "), wordWrap)
 	}
 
 	if readline.IsTerminal(int(os.Stdin.Fd())) {
@@ -392,8 +377,8 @@ func RunGenerate(cmd *cobra.Command, args []string) error {
 
 type generateContextKey string
 
-func generate(cmd *cobra.Command, model, prompt string) error {
-	client, err := api.FromEnv()
+func generate(cmd *cobra.Command, model, prompt string, wordWrap bool) error {
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -408,24 +393,9 @@ func generate(cmd *cobra.Command, model, prompt string) error {
 		generateContext = []int{}
 	}
 
-	var wrapTerm bool
-	termType := os.Getenv("TERM")
-	if termType == "xterm-256color" {
-		wrapTerm = true
-	}
-
 	termWidth, _, err := term.GetSize(int(0))
 	if err != nil {
-		wrapTerm = false
-	}
-
-	// override wrapping if the user turned it off
-	nowrap, err := cmd.Flags().GetBool("nowordwrap")
-	if err != nil {
-		return err
-	}
-	if nowrap {
-		wrapTerm = false
+		wordWrap = false
 	}
 
 	cancelCtx, cancel := context.WithCancel(context.Background())
@@ -452,7 +422,7 @@ func generate(cmd *cobra.Command, model, prompt string) error {
 
 		latest = response
 
-		if wrapTerm {
+		if wordWrap {
 			for _, ch := range response.Response {
 				if currentLineLength+1 > termWidth-5 {
 					// backtrack the length of the last word and clear to the end of the line
@@ -481,18 +451,7 @@ func generate(cmd *cobra.Command, model, prompt string) error {
 	}
 
 	if err := client.Generate(cancelCtx, &request, fn); err != nil {
-		if strings.Contains(err.Error(), "failed to load model") {
-			// tell the user to check the server log, if it exists locally
-			home, nestedErr := os.UserHomeDir()
-			if nestedErr != nil {
-				// return the original error
-				return err
-			}
-			logPath := filepath.Join(home, ".ollama", "logs", "server.log")
-			if _, nestedErr := os.Stat(logPath); nestedErr == nil {
-				err = fmt.Errorf("%w\nFor more details, check the error logs at %s", err, logPath)
-			}
-		} else if strings.Contains(err.Error(), "context canceled") && abort {
+		if strings.Contains(err.Error(), "context canceled") && abort {
 			spinner.Finish()
 			return nil
 		}
@@ -527,69 +486,84 @@ func generate(cmd *cobra.Command, model, prompt string) error {
 }
 
 func generateInteractive(cmd *cobra.Command, model string) error {
-	home, err := os.UserHomeDir()
-	if err != nil {
-		return err
-	}
-
 	// load the model
-	if err := generate(cmd, model, ""); err != nil {
+	if err := generate(cmd, model, "", false); err != nil {
 		return err
 	}
 
-	completer := readline.NewPrefixCompleter(
-		readline.PcItem("/help"),
-		readline.PcItem("/list"),
-		readline.PcItem("/set",
-			readline.PcItem("history"),
-			readline.PcItem("nohistory"),
-			readline.PcItem("wordwrap"),
-			readline.PcItem("nowordwrap"),
-			readline.PcItem("verbose"),
-			readline.PcItem("quiet"),
-		),
-		readline.PcItem("/show",
-			readline.PcItem("license"),
-			readline.PcItem("modelfile"),
-			readline.PcItem("parameters"),
-			readline.PcItem("system"),
-			readline.PcItem("template"),
-		),
-		readline.PcItem("/exit"),
-		readline.PcItem("/bye"),
-	)
-
 	usage := func() {
-		fmt.Fprintln(os.Stderr, "commands:")
-		fmt.Fprintln(os.Stderr, completer.Tree("  "))
+		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, "")
 	}
 
-	var painter Painter
-
-	config := readline.Config{
-		Painter:      &painter,
-		Prompt:       ">>> ",
-		HistoryFile:  filepath.Join(home, ".ollama", "history"),
-		AutoComplete: completer,
+	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 verbose      Show LLM stats")
+		fmt.Fprintln(os.Stderr, "  /set quiet        Disable LLM stats")
+		fmt.Fprintln(os.Stderr, "")
 	}
 
-	scanner, err := readline.NewEx(&config)
+	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, "")
+	}
+
+	prompt := readline.Prompt{
+		Prompt:         ">>> ",
+		AltPrompt:      "... ",
+		Placeholder:    "Send a message (/? for help)",
+		AltPlaceholder: `Use """ to end multi-line input`,
+	}
+
+	scanner, err := readline.New(prompt)
 	if err != nil {
 		return err
 	}
-	defer scanner.Close()
+
+	var wordWrap bool
+	termType := os.Getenv("TERM")
+	if termType == "xterm-256color" {
+		wordWrap = true
+	}
+
+	// override wrapping if the user turned it off
+	nowrap, err := cmd.Flags().GetBool("nowordwrap")
+	if err != nil {
+		return err
+	}
+	if nowrap {
+		wordWrap = false
+	}
+
+	fmt.Print(readline.StartBracketedPaste)
+	defer fmt.Printf(readline.EndBracketedPaste)
 
 	var multiLineBuffer string
-	var isMultiLine bool
 
 	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("Use Ctrl-D or /bye to exit.")
+				fmt.Println("\nUse Ctrl-D or /bye to exit.")
 			}
 
 			continue
@@ -600,23 +574,19 @@ func generateInteractive(cmd *cobra.Command, model string) error {
 		line = strings.TrimSpace(line)
 
 		switch {
-		case isMultiLine:
+		case scanner.Prompt.UseAlt:
 			if strings.HasSuffix(line, `"""`) {
-				isMultiLine = false
-				painter.IsMultiLine = isMultiLine
+				scanner.Prompt.UseAlt = false
 				multiLineBuffer += strings.TrimSuffix(line, `"""`)
 				line = multiLineBuffer
 				multiLineBuffer = ""
-				scanner.SetPrompt(">>> ")
 			} else {
 				multiLineBuffer += line + " "
 				continue
 			}
 		case strings.HasPrefix(line, `"""`):
-			isMultiLine = true
-			painter.IsMultiLine = isMultiLine
+			scanner.Prompt.UseAlt = true
 			multiLineBuffer = strings.TrimPrefix(line, `"""`) + " "
-			scanner.SetPrompt("... ")
 			continue
 		case strings.HasPrefix(line, "/list"):
 			args := strings.Fields(line)
@@ -632,10 +602,10 @@ func generateInteractive(cmd *cobra.Command, model string) error {
 				case "nohistory":
 					scanner.HistoryDisable()
 				case "wordwrap":
-					cmd.Flags().Set("nowordwrap", "false")
+					wordWrap = true
 					fmt.Println("Set 'wordwrap' mode.")
 				case "nowordwrap":
-					cmd.Flags().Set("nowordwrap", "true")
+					wordWrap = false
 					fmt.Println("Set 'nowordwrap' mode.")
 				case "verbose":
 					cmd.Flags().Set("verbose", "true")
@@ -643,29 +613,21 @@ func generateInteractive(cmd *cobra.Command, model string) error {
 				case "quiet":
 					cmd.Flags().Set("verbose", "false")
 					fmt.Println("Set 'quiet' mode.")
-				case "mode":
-					if len(args) > 2 {
-						switch args[2] {
-						case "vim":
-							scanner.SetVimMode(true)
-						case "emacs", "default":
-							scanner.SetVimMode(false)
-						default:
-							usage()
-						}
-					} else {
-						usage()
-					}
 				default:
 					fmt.Printf("Unknown command '/set %s'. Type /? for help\n", args[1])
 				}
 			} else {
-				usage()
+				usageSet()
 			}
 		case strings.HasPrefix(line, "/show"):
 			args := strings.Fields(line)
 			if len(args) > 1 {
-				resp, err := server.GetModelInfo(model)
+				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
@@ -673,23 +635,49 @@ func generateInteractive(cmd *cobra.Command, model string) error {
 
 				switch args[1] {
 				case "license":
-					fmt.Println(resp.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":
-					fmt.Println(resp.Parameters)
+					if resp.Parameters == "" {
+						fmt.Print("No parameters were specified for this model.\n\n")
+					} else {
+						fmt.Println(resp.Parameters)
+					}
 				case "system":
-					fmt.Println(resp.System)
+					if resp.System == "" {
+						fmt.Print("No system prompt was specified for this model.\n\n")
+					} else {
+						fmt.Println(resp.System)
+					}
 				case "template":
-					fmt.Println(resp.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 == "/help", line == "/?":
-			usage()
 		case line == "/exit", line == "/bye":
 			return nil
 		case strings.HasPrefix(line, "/"):
@@ -698,7 +686,7 @@ func generateInteractive(cmd *cobra.Command, model string) error {
 		}
 
 		if len(line) > 0 && line[0] != '/' {
-			if err := generate(cmd, model, line); err != nil {
+			if err := generate(cmd, model, line, wordWrap); err != nil {
 				return err
 			}
 		}
@@ -710,7 +698,7 @@ func generateBatch(cmd *cobra.Command, model string) error {
 	for scanner.Scan() {
 		prompt := scanner.Text()
 		fmt.Printf(">>> %s\n", prompt)
-		if err := generate(cmd, model, prompt); err != nil {
+		if err := generate(cmd, model, prompt, false); err != nil {
 			return err
 		}
 	}
@@ -741,21 +729,6 @@ func RunServer(cmd *cobra.Command, _ []string) error {
 		origins = strings.Split(o, ",")
 	}
 
-	if noprune := os.Getenv("OLLAMA_NOPRUNE"); noprune == "" {
-		if err := server.PruneLayers(); err != nil {
-			return err
-		}
-
-		manifestsPath, err := server.GetManifestPath()
-		if err != nil {
-			return err
-		}
-
-		if err := server.PruneDirectory(manifestsPath); err != nil {
-			return err
-		}
-	}
-
 	return server.Serve(ln, origins)
 }
 
@@ -840,7 +813,7 @@ func startMacApp(client *api.Client) error {
 }
 
 func checkServerHeartbeat(_ *cobra.Command, _ []string) error {
-	client, err := api.FromEnv()
+	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return err
 	}
@@ -878,7 +851,7 @@ func NewCLI() *cobra.Command {
 	createCmd := &cobra.Command{
 		Use:     "create MODEL",
 		Short:   "Create a model from a Modelfile",
-		Args:    cobra.MinimumNArgs(1),
+		Args:    cobra.ExactArgs(1),
 		PreRunE: checkServerHeartbeat,
 		RunE:    CreateHandler,
 	}
@@ -888,7 +861,7 @@ func NewCLI() *cobra.Command {
 	showCmd := &cobra.Command{
 		Use:     "show MODEL",
 		Short:   "Show information for a model",
-		Args:    cobra.MinimumNArgs(1),
+		Args:    cobra.ExactArgs(1),
 		PreRunE: checkServerHeartbeat,
 		RunE:    ShowHandler,
 	}
@@ -915,13 +888,14 @@ func NewCLI() *cobra.Command {
 		Use:     "serve",
 		Aliases: []string{"start"},
 		Short:   "Start ollama",
+		Args:    cobra.ExactArgs(0),
 		RunE:    RunServer,
 	}
 
 	pullCmd := &cobra.Command{
 		Use:     "pull MODEL",
 		Short:   "Pull a model from a registry",
-		Args:    cobra.MinimumNArgs(1),
+		Args:    cobra.ExactArgs(1),
 		PreRunE: checkServerHeartbeat,
 		RunE:    PullHandler,
 	}
@@ -931,7 +905,7 @@ func NewCLI() *cobra.Command {
 	pushCmd := &cobra.Command{
 		Use:     "push MODEL",
 		Short:   "Push a model to a registry",
-		Args:    cobra.MinimumNArgs(1),
+		Args:    cobra.ExactArgs(1),
 		PreRunE: checkServerHeartbeat,
 		RunE:    PushHandler,
 	}
@@ -947,15 +921,15 @@ func NewCLI() *cobra.Command {
 	}
 
 	copyCmd := &cobra.Command{
-		Use:     "cp",
+		Use:     "cp SOURCE TARGET",
 		Short:   "Copy a model",
-		Args:    cobra.MinimumNArgs(2),
+		Args:    cobra.ExactArgs(2),
 		PreRunE: checkServerHeartbeat,
 		RunE:    CopyHandler,
 	}
 
 	deleteCmd := &cobra.Command{
-		Use:     "rm",
+		Use:     "rm MODEL [MODEL...]",
 		Short:   "Remove a model",
 		Args:    cobra.MinimumNArgs(1),
 		PreRunE: checkServerHeartbeat,

docs/api.md

@@ -12,7 +12,6 @@
 - [Push a Model](#push-a-model)
 - [Generate Embeddings](#generate-embeddings)
 
-
 ## Conventions
 
 ### Model names
@@ -39,30 +38,39 @@ Generate a response for a given prompt with a provided model. This is a streamin
 
 - `model`: (required) the [model name](#model-names)
 - `prompt`: the prompt to generate a response for
+- `format`: the format to return a response in. Currently the only accepted value is `json`
 
-Advanced parameters:
+Advanced parameters (optional):
 
 - `options`: additional model parameters listed in the documentation for the [Modelfile](./modelfile.md#valid-parameters-and-values) such as `temperature`
 - `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 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.
 
-### Request
+### JSON mode
+
+Enable JSON mode by setting the `format` parameter to `json` and specifying the model should use JSON in the `prompt`. This will structure the response as valid JSON. See the JSON mode [example](#request-json-mode) below.
+
+### Examples
+
+#### Request
 
 ```shell
 curl -X POST http://localhost:11434/api/generate -d '{
-  "model": "llama2:7b",
+  "model": "llama2",
   "prompt": "Why is the sky blue?"
 }'
 ```
 
-### Response
+#### Response
 
-A stream of JSON objects:
+A stream of JSON objects is returned:
 
 ```json
 {
-  "model": "llama2:7b",
+  "model": "llama2",
   "created_at": "2023-08-04T08:52:19.385406455-07:00",
   "response": "The",
   "done": false
@@ -80,13 +88,15 @@ The final response in the stream also includes additional data about the generat
 - `eval_count`: number of tokens the response
 - `eval_duration`: time in nanoseconds spent generating the response
 - `context`: an encoding of the conversation used in this response, this can be sent in the next request to keep a conversational memory
+- `response`: empty if the response was streamed, if not streamed, this will contain the full response
 
 To calculate how fast the response is generated in tokens per second (token/s), divide `eval_count` / `eval_duration`.
 
 ```json
 {
-  "model": "llama2:7b",
+  "model": "llama2",
   "created_at": "2023-08-04T19:22:45.499127Z",
+  "response": "",
   "context": [1, 2, 3],
   "done": true,
   "total_duration": 5589157167,
@@ -100,6 +110,182 @@ To calculate how fast the response is generated in tokens per second (token/s),
 }
 ```
 
+#### Request (No streaming)
+
+```shell
+curl -X POST http://localhost:11434/api/generate -d '{
+  "model": "llama2:7b",
+  "prompt": "Why is the sky blue?",
+  "stream": false
+}'
+```
+
+#### Response
+
+If `stream` is set to `false`, the response will be a single JSON object:
+
+```json
+{
+  "model": "llama2:7b",
+  "created_at": "2023-08-04T19:22:45.499127Z",
+  "response": "The sky is blue because it is the color of the sky.",
+  "context": [1, 2, 3],
+  "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 -X POST 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)
+
+```shell
+curl -X POST http://localhost:11434/api/generate -d '{
+  "model": "llama2",
+  "prompt": "What color is the sky at different times of the day? Respond using JSON",
+  "format": "json",
+  "stream": false
+}'
+```
+
+#### Response
+
+```json
+{
+  "model": "llama2",
+  "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,
+  "total_duration": 4661289125,
+  "load_duration": 1714434500,
+  "prompt_eval_count": 36,
+  "prompt_eval_duration": 264132000,
+  "eval_count": 75,
+  "eval_duration": 2112149000
+}
+```
+
+The value of `response` will be a string containing JSON similar to:
+
+```json
+{
+  "morning": {
+    "color": "blue"
+  },
+  "noon": {
+    "color": "blue-gray"
+  },
+  "afternoon": {
+    "color": "warm gray"
+  },
+  "evening": {
+    "color": "orange"
+  }
+}
+```
+
+#### 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.
+
+```shell
+curl -X POST http://localhost:11434/api/generate -d '{
+  "model": "llama2:7b",
+  "prompt": "Why is the sky blue?",
+  "stream": false,
+  "options": {
+    "num_keep": 5,
+    "seed": 42,
+    "num_predict": 100,
+    "top_k": 20,
+    "top_p": 0.9,
+    "tfs_z": 0.5,
+    "typical_p": 0.7,
+    "repeat_last_n": 33,
+    "temperature": 0.8,
+    "repeat_penalty": 1.2,
+    "presence_penalty": 1.5,
+    "frequency_penalty": 1.0,
+    "mirostat": 1,
+    "mirostat_tau": 0.8,
+    "mirostat_eta": 0.6,
+    "penalize_newline": true,
+    "stop": ["\n", "user:"],
+    "numa": false,
+    "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,
+    "embedding_only": false,
+    "rope_frequency_base": 1.1,
+    "rope_frequency_scale": 0.8,
+    "num_thread": 8
+    }
+}'
+```
+
+#### Response
+
+```json
+{
+  "model": "llama2:7b",
+  "created_at": "2023-08-04T19:22:45.499127Z",
+  "response": "The sky is blue because it is the color of the sky.",
+  "context": [1, 2, 3],
+  "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
+}
+```
+
 ## Create a Model
 
 ```shell
@@ -112,8 +298,11 @@ Create a model from a [`Modelfile`](./modelfile.md)
 
 - `name`: name of the model to create
 - `path`: path to the Modelfile
+- `stream`: (optional) if `false` the response will be returned as a single response object, rather than a stream of objects
 
-### Request
+### Examples
+
+#### Request
 
 ```shell
 curl -X POST http://localhost:11434/api/create -d '{
@@ -122,7 +311,7 @@ curl -X POST http://localhost:11434/api/create -d '{
 }'
 ```
 
-### Response
+#### Response
 
 A stream of JSON objects. When finished, `status` is `success`.
 
@@ -140,13 +329,17 @@ GET /api/tags
 
 List models that are available locally.
 
-### Request
+### Examples
+
+#### Request
 
 ```shell
 curl http://localhost:11434/api/tags
 ```
 
-### Response
+#### Response
+
+A single JSON object will be returned.
 
 ```json
 {
@@ -177,22 +370,24 @@ Show details about a model including modelfile, template, parameters, license, a
 
 - `name`: name of the model to show
 
-### Request
+### Examples
 
-```shell  
+#### Request
+
+```shell
 curl http://localhost:11434/api/show -d '{
   "name": "llama2:7b"
 }'
 ```
 
-### Response
+#### Response
 
 ```json
 {
-    "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] "
+  "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] "
 }
 ```
 
@@ -204,7 +399,9 @@ POST /api/copy
 
 Copy a model. Creates a model with another name from an existing model.
 
-### Request
+### Examples
+
+#### Request
 
 ```shell
 curl http://localhost:11434/api/copy -d '{
@@ -213,6 +410,10 @@ curl http://localhost:11434/api/copy -d '{
 }'
 ```
 
+#### Response
+
+The only response is a 200 OK if successful.
+
 ## Delete a Model
 
 ```shell
@@ -223,9 +424,11 @@ Delete a model and its data.
 
 ### Parameters
 
-- `model`: model name to delete
+- `name`: model name to delete
 
-### Request
+### Examples
+
+#### Request
 
 ```shell
 curl -X DELETE http://localhost:11434/api/delete -d '{
@@ -233,6 +436,10 @@ curl -X DELETE http://localhost:11434/api/delete -d '{
 }'
 ```
 
+#### Response
+
+If successful, the only response is a 200 OK.
+
 ## Pull a Model
 
 ```shell
@@ -245,8 +452,11 @@ Download a model from the ollama library. Cancelled pulls are resumed from where
 
 - `name`: name of the model to pull
 - `insecure`: (optional) allow insecure connections to the library. Only use this if you are pulling from your own library during development.
+- `stream`: (optional) if `false` the response will be returned as a single response object, rather than a stream of objects
 
-### Request
+### Examples
+
+#### Request
 
 ```shell
 curl -X POST http://localhost:11434/api/pull -d '{
@@ -254,13 +464,51 @@ curl -X POST http://localhost:11434/api/pull -d '{
 }'
 ```
 
-### Response
+#### Response
+
+If `stream` is not specified, or set to `true`, a stream of JSON objects is returned:
+
+The first object is the manifest:
+
+```json
+{
+  "status": "pulling manifest"
+}
+```
+
+Then there is a series of downloading responses. Until any of the download is completed, the `completed` key may not be included. The number of files to be downloaded depends on the number of layers specified in the manifest.
 
 ```json
 {
   "status": "downloading digestname",
   "digest": "digestname",
-  "total": 2142590208
+  "total": 2142590208,
+  "completed": 241970
+}
+```
+
+After all the files are downloaded, the final responses are:
+
+```json
+{
+    "status": "verifying sha256 digest"
+}
+{
+    "status": "writing manifest"
+}
+{
+    "status": "removing any unused layers"
+}
+{
+    "status": "success"
+}
+```
+
+if `stream` is set to false, then the response is a single JSON object:
+
+```json
+{
+  "status": "success"
 }
 ```
 
@@ -275,9 +523,12 @@ Upload a model to a model library. Requires registering for ollama.ai and adding
 ### Parameters
 
 - `name`: name of the model to push in the form of `<namespace>/<model>:<tag>`
-- `insecure`: (optional) allow insecure connections to the library. Only use this if you are pushing to your library during development.  
+- `insecure`: (optional) allow insecure connections to the library. Only use this if you are pushing to your library during development.
+- `stream`: (optional) if `false` the response will be returned as a single response object, rather than a stream of objects
 
-### Request
+### Examples
+
+#### Request
 
 ```shell
 curl -X POST http://localhost:11434/api/push -d '{
@@ -285,20 +536,21 @@ curl -X POST http://localhost:11434/api/push -d '{
 }'
 ```
 
-### Response
+#### Response
 
-Streaming response that starts with:
+If `stream` is not specified, or set to `true`, a stream of JSON objects is returned:
 
 ```json
-{"status":"retrieving manifest"}
+{ "status": "retrieving manifest" }
 ```
 
 and then:
 
 ```json
 {
-"status":"starting upload","digest":"sha256:bc07c81de745696fdf5afca05e065818a8149fb0c77266fb584d9b2cba3711ab",
-"total":1928429856
+  "status": "starting upload",
+  "digest": "sha256:bc07c81de745696fdf5afca05e065818a8149fb0c77266fb584d9b2cba3711ab",
+  "total": 1928429856
 }
 ```
 
@@ -306,9 +558,10 @@ Then there is a series of uploading responses:
 
 ```json
 {
-"status":"starting upload",
-"digest":"sha256:bc07c81de745696fdf5afca05e065818a8149fb0c77266fb584d9b2cba3711ab",
-"total":1928429856}
+  "status": "starting upload",
+  "digest": "sha256:bc07c81de745696fdf5afca05e065818a8149fb0c77266fb584d9b2cba3711ab",
+  "total": 1928429856
+}
 ```
 
 Finally, when the upload is complete:
@@ -318,6 +571,12 @@ Finally, when the upload is complete:
 {"status":"success"}
 ```
 
+If `stream` is set to `false`, then the response is a single JSON object:
+
+```json
+{ "status": "success" }
+```
+
 ## Generate Embeddings
 
 ```shell
@@ -335,7 +594,9 @@ Advanced parameters:
 
 - `options`: additional model parameters listed in the documentation for the [Modelfile](./modelfile.md#valid-parameters-and-values) such as `temperature`
 
-### Request
+### Examples
+
+#### Request
 
 ```shell
 curl -X POST http://localhost:11434/api/embeddings -d '{
@@ -344,12 +605,13 @@ curl -X POST http://localhost:11434/api/embeddings -d '{
 }'
 ```
 
-### Response
+#### Response
 
 ```json
 {
-  "embeddings": [
+  "embedding": [
     0.5670403838157654, 0.009260174818336964, 0.23178744316101074, -0.2916173040866852, -0.8924556970596313,
     0.8785552978515625, -0.34576427936553955, 0.5742510557174683, -0.04222835972905159, -0.137906014919281
   ]
-}```
+}
+```

docs/development.md

@@ -10,25 +10,25 @@ Install required tools:
 - go version 1.20 or higher
 - gcc version 11.4.0 or higher
 
-```
+```bash
 brew install go cmake gcc
 ```
 
 Get the required libraries:
 
-```
+```bash
 go generate ./...
 ```
 
 Then build ollama:
 
-```
+```bash
 go build .
 ```
 
 Now you can run `ollama`:
 
-```
+```bash
 ./ollama
 ```
 

docs/faq.md

@@ -1,19 +1,98 @@
 # FAQ
 
-## How can I expose the Ollama server?
+## How can I view the logs?
+
+On macOS:
 
 ```
+cat ~/.ollama/logs/server.log
+```
+
+On Linux:
+
+```
+journalctl -u 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.
+
+On macOS:
+
+```bash
 OLLAMA_HOST=0.0.0.0:11435 ollama serve
 ```
 
-By default, Ollama allows cross origin requests from `127.0.0.1` and `0.0.0.0`. To support more origins, you can use the `OLLAMA_ORIGINS` environment variable:
+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:
+
+On macOS:
+
+```bash
 OLLAMA_ORIGINS=http://192.168.1.1:*,https://example.com ollama serve
 ```
 
+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: Raw model data is stored under `~/.ollama/models`.
-* Linux: Raw model data is stored under `/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`
 
+
+
+Below the models directory you will find a structure similar to the following:
+
+```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.

docs/import.md

@@ -0,0 +1,198 @@
+# Import a model
+
+This guide walks through importing a GGUF, PyTorch or Safetensors model.
+
+## Importing (GGUF)
+
+### Step 1: Write a `Modelfile`
+
+Start by creating a `Modelfile`. This file is the blueprint for your model, specifying weights, parameters, prompt templates and more.
+
+```
+FROM ./mistral-7b-v0.1.Q4_0.gguf
+```
+
+(Optional) many chat models require a prompt template in order to answer correctly. A default prompt template can be specified with the `TEMPLATE` instruction in the `Modelfile`:
+
+```
+FROM ./q4_0.bin
+TEMPLATE "[INST] {{ .Prompt }} [/INST]"
+```
+
+### Step 2: Create the Ollama model
+
+Finally, create a model from your `Modelfile`:
+
+```
+ollama create example -f Modelfile
+```
+
+### Step 3: Run your model
+
+Next, test the model with `ollama run`:
+
+```
+ollama run example "What is your favourite condiment?"
+```
+
+## Importing (PyTorch & Safetensors)
+
+### Supported models
+
+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`).
+
+### Step 1: Clone the HuggingFace repository (optional)
+
+If the model is currently hosted in a HuggingFace repository, first clone that repository to download the raw model.
+
+```
+git lfs install
+git clone https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1
+cd Mistral-7B-Instruct-v0.1
+```
+
+### Step 2: Convert and quantize to a `.bin` file (optional, for PyTorch and Safetensors)
+
+If the model is in PyTorch or Safetensors format, a [Docker image](https://hub.docker.com/r/ollama/quantize) with the tooling required to convert and quantize models is available.
+
+First, Install [Docker](https://www.docker.com/get-started/).
+
+Next, to convert and quantize your model, run:
+
+```
+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 (we will use this file to create the Ollama model)
+
+### Step 3: Write a `Modelfile`
+
+Next, create a `Modelfile` for your model:
+
+```
+FROM ./q4_0.bin
+```
+
+(Optional) many chat models require a prompt template in order to answer correctly. A default prompt template can be specified with the `TEMPLATE` instruction in the `Modelfile`:
+
+```
+FROM ./q4_0.bin
+TEMPLATE "[INST] {{ .Prompt }} [/INST]"
+```
+
+### Step 4: Create the Ollama model
+
+Finally, create a model from your `Modelfile`:
+
+```
+ollama create example -f Modelfile
+```
+
+### Step 5: Run your model
+
+Next, test the model with `ollama run`:
+
+```
+ollama run example "What is your favourite condiment?"
+```
+
+## Publishing your model (optional – early alpha)
+
+Publishing models is in early alpha. If you'd like to publish your model to share with others, follow these steps:
+
+1. Create [an account](https://ollama.ai/signup)
+2. Run `cat ~/.ollama/id_ed25519.pub` to view your Ollama public key. Copy this to the clipboard.
+3. Add your public key to your [Ollama account](https://ollama.ai/settings/keys)
+
+Next, copy your model to your username's namespace:
+
+```
+ollama cp example <your username>/example
+```
+
+Then push the model:
+
+```
+ollama push <your username>/example
+```
+
+After publishing, your model will be available at `https://ollama.ai/<your username>/example`.
+
+## Quantization reference
+
+The quantization options are as follow (from highest highest to lowest levels of quantization). Note: some architectures such as Falcon do not support K quants.
+
+- `q2_K`
+- `q3_K`
+- `q3_K_S`
+- `q3_K_M`
+- `q3_K_L`
+- `q4_0` (recommended)
+- `q4_1`
+- `q4_K`
+- `q4_K_S`
+- `q4_K_M`
+- `q5_0`
+- `q5_1`
+- `q5_K`
+- `q5_K_S`
+- `q5_K_M`
+- `q6_K`
+- `q8_0`
+
+## Manually converting & quantizing models
+
+### Prerequisites
+
+Start by cloning the `llama.cpp` repo to your machine in another directory:
+
+```
+git clone https://github.com/ggerganov/llama.cpp.git
+cd llama.cpp
+```
+
+Next, install the Python dependencies:
+
+```
+pip install -r requirements.txt
+```
+
+Finally, build the `quantize` tool:
+
+```
+make quantize
+```
+
+### Convert the model
+
+Run the correct conversion script for your model architecture:
+
+```shell
+# LlamaForCausalLM or MistralForCausalLM
+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>
+```
+
+### Quantize the model
+
+```
+quantize <path to model dir>/ggml-model-f32.bin <path to model dir>/q4_0.bin q4_0
+```

docs/linux.md

@@ -1,49 +1,29 @@
-# Installing Ollama on Linux
+# Ollama on Linux
 
-> Note: A one line installer for Ollama is available by running:
+## Install
+
+Install Ollama running this one-liner:
 >
-> ```
-> curl https://ollama.ai/install.sh | sh
-> ```
+```bash
+curl https://ollama.ai/install.sh | sh
+```
 
-## Download the `ollama` binary
+## Manual install
+
+### Download the `ollama` binary
 
 Ollama is distributed as a self-contained binary. Download it to a directory in your PATH:
 
-```
+```bash
 sudo curl -L https://ollama.ai/download/ollama-linux-amd64 -o /usr/bin/ollama
 sudo chmod +x /usr/bin/ollama
 ```
 
-## Start Ollama
-
-Start Ollama by running `ollama serve`:
-
-```
-ollama serve
-```
-
-Once Ollama is running, run a model in another terminal session:
-
-```
-ollama run llama2
-```
-
-## Install CUDA drivers (optional – for Nvidia GPUs)
-
-[Download and install](https://developer.nvidia.com/cuda-downloads) CUDA.
-
-Verify that the drivers are installed by running the following command, which should print details about your GPU:
-
-```
-nvidia-smi
-```
-
-## Adding Ollama as a startup service (optional)
+### Adding Ollama as a startup service (recommended)
 
 Create a user for Ollama:
 
-```
+```bash
 sudo useradd -r -s /bin/false -m -d /usr/share/ollama ollama
 ```
 
@@ -60,7 +40,6 @@ User=ollama
 Group=ollama
 Restart=always
 RestartSec=3
-Environment="HOME=/usr/share/ollama"
 
 [Install]
 WantedBy=default.target
@@ -68,16 +47,70 @@ WantedBy=default.target
 
 Then start the service:
 
-```
+```bash
 sudo systemctl daemon-reload
 sudo systemctl enable ollama
 ```
 
-### Viewing logs
+### Install CUDA drivers (optional – for Nvidia GPUs)
+
+[Download and install](https://developer.nvidia.com/cuda-downloads) CUDA.
+
+Verify that the drivers are installed by running the following command, which should print details about your GPU:
+
+```bash
+nvidia-smi
+```
+
+### Start Ollama
+
+Start Ollama using `systemd`:
+
+```bash
+sudo systemctl start ollama
+```
+
+## Update
+
+Update ollama by running the install script again:
+
+```bash
+curl https://ollama.ai/install.sh | sh
+```
+
+Or by downloading the ollama binary:
+
+```bash
+sudo curl -L https://ollama.ai/download/ollama-linux-amd64 -o /usr/bin/ollama
+sudo chmod +x /usr/bin/ollama
+```
+
+## Viewing logs
 
 To view logs of Ollama running as a startup service, run:
 
-```
+```bash
 journalctl -u ollama
 ```
 
+## Uninstall
+
+Remove the ollama service:
+
+```bash
+sudo systemctl stop ollama
+sudo systemctl disable ollama
+sudo rm /etc/systemd/system/ollama.service
+```
+
+Remove the ollama binary from your bin directory (either `/usr/local/bin`, `/usr/bin`, or `/bin`):
+
+```bash
+sudo rm $(which ollama)
+```
+
+Remove the downloaded models and Ollama service user:
+```bash
+sudo rm -r /usr/share/ollama
+sudo userdel ollama
+```

docs/modelfile.md

@@ -1,6 +1,6 @@
 # Ollama Model File
 
-> Note: this model file syntax is in development
+> Note: this `Modelfile` syntax is in development
 
 A model file is the blueprint to create and share models with Ollama.
 
@@ -12,7 +12,6 @@ A model file is the blueprint to create and share models with Ollama.
   - [FROM (Required)](#from-required)
     - [Build from llama2](#build-from-llama2)
     - [Build from a bin file](#build-from-a-bin-file)
-  - [EMBED](#embed)
   - [PARAMETER](#parameter)
     - [Valid Parameters and Values](#valid-parameters-and-values)
   - [TEMPLATE](#template)
@@ -24,7 +23,7 @@ A model file is the blueprint to create and share models with Ollama.
 
 ## Format
 
-The format of the Modelfile:
+The format of the `Modelfile`:
 
 ```modelfile
 # comment
@@ -42,9 +41,9 @@ INSTRUCTION arguments
 
 ## Examples
 
-An example of a model file creating a mario blueprint:
+An example of a `Modelfile` creating a mario blueprint:
 
-```
+```modelfile
 FROM llama2
 # sets the temperature to 1 [higher is more creative, lower is more coherent]
 PARAMETER temperature 1
@@ -57,9 +56,9 @@ SYSTEM You are Mario from super mario bros, acting as an assistant.
 
 To use this:
 
-1. Save it as a file (eg. `Modelfile`)
-2. `ollama create NAME -f <location of the file eg. ./Modelfile>'`
-3. `ollama run NAME`
+1. Save it as a file (e.g. `Modelfile`)
+2. `ollama create choose-a-model-name -f <location of the file e.g. ./Modelfile>'`
+3. `ollama run choose-a-model-name`
 4. Start using the model!
 
 More examples are available in the [examples directory](../examples).
@@ -68,45 +67,34 @@ More examples are available in the [examples directory](../examples).
 
 ### FROM (Required)
 
-The FROM instruction defines the base model to use when creating a model.
+The `FROM` instruction defines the base model to use when creating a model.
 
-```
+```modelfile
 FROM <model name>:<tag>
 ```
 
 #### Build from llama2
 
-```
+```modelfile
 FROM llama2
 ```
 
 A list of available base models:
 <https://github.com/jmorganca/ollama#model-library>
 
-#### Build from a bin file
+#### Build from a `bin` file
 
-```
+```modelfile
 FROM ./ollama-model.bin
 ```
 
-This bin file location should be specified as an absolute path or relative to the Modelfile location.
-
-### EMBED
-
-The EMBED instruction is used to add embeddings of files to a model. This is useful for adding custom data that the model can reference when generating an answer. Note that currently only text files are supported, formatted with each line as one embedding.
-
-```
-FROM <model name>:<tag>
-EMBED <file path>.txt
-EMBED <different file path>.txt
-EMBED <path to directory>/*.txt
-```
+This bin file location should be specified as an absolute path or relative to the `Modelfile` location.
 
 ### PARAMETER
 
 The `PARAMETER` instruction defines a parameter that can be set when the model is run.
 
-```
+```modelfile
 PARAMETER <parameter> <parametervalue>
 ```
 
@@ -124,7 +112,8 @@ PARAMETER <parameter> <parametervalue>
 | repeat_last_n  | Sets how far back for the model to look back to prevent repetition. (Default: 64, 0 = disabled, -1 = num_ctx)                                                                                                                                           | int        | repeat_last_n 64     |
 | repeat_penalty | Sets how strongly to penalize repetitions. A higher value (e.g., 1.5) will penalize repetitions more strongly, while a lower value (e.g., 0.9) will be more lenient. (Default: 1.1)                                                                     | float      | repeat_penalty 1.1   |
 | temperature    | The temperature of the model. Increasing the temperature will make the model answer more creatively. (Default: 0.8)                                                                                                                                     | float      | temperature 0.7      |
-| stop           | Sets the stop sequences to use.                                                                                                                                                                                                                         | string     | stop "AI assistant:" |
+| seed           | Sets the random number seed to use for generation. Setting this to a specific number will make the model generate the same text for the same prompt. (Default: 0)                                                                                       | int        | seed 42              |
+| stop           | Sets the stop sequences to use. When this pattern is encountered the LLM will stop generating text and return. Multiple stop patterns may be set by specifying multiple separate `stop` parameters in a modelfile.                                      | string     | stop "AI assistant:" |
 | tfs_z          | Tail free sampling is used to reduce the impact of less probable tokens from the output. A higher value (e.g., 2.0) will reduce the impact more, while a value of 1.0 disables this setting. (default: 1)                                               | float      | tfs_z 1              |
 | num_predict    | Maximum number of tokens to predict when generating text. (Default: 128, -1 = infinite generation, -2 = fill context)                                                                                                                                   | int        | num_predict 42       |
 | top_k          | Reduces the probability of generating nonsense. A higher value (e.g. 100) will give more diverse answers, while a lower value (e.g. 10) will be more conservative. (Default: 40)                                                                        | int        | top_k 40             |
@@ -132,7 +121,7 @@ PARAMETER <parameter> <parametervalue>
 
 ### TEMPLATE
 
-`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.
+`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
 
@@ -142,7 +131,7 @@ PARAMETER <parameter> <parametervalue>
 | `{{ .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 """
 {{- if .First }}
 ### System:
@@ -162,7 +151,7 @@ SYSTEM """<system message>"""
 
 The `SYSTEM` instruction specifies the system prompt to be used in the template, if applicable.
 
-```
+```modelfile
 SYSTEM """<system message>"""
 ```
 
@@ -170,7 +159,7 @@ SYSTEM """<system message>"""
 
 The `ADAPTER` instruction specifies the LoRA adapter to apply to the base model. The value of this instruction should be an absolute path or a path relative to the Modelfile and the file must be in a GGML file format. The adapter should be tuned from the base model otherwise the behaviour is undefined.
 
-```
+```modelfile
 ADAPTER ./ollama-lora.bin
 ```
 
@@ -178,7 +167,7 @@ ADAPTER ./ollama-lora.bin
 
 The `LICENSE` instruction allows you to specify the legal license under which the model used with this Modelfile is shared or distributed.
 
-```
+```modelfile
 LICENSE """
 <license text>
 """
@@ -186,5 +175,5 @@ LICENSE """
 
 ## Notes
 
-- the **modelfile is not case sensitive**. In the examples, we use uppercase for instructions to make it easier to distinguish it from arguments.
+- 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.

docs/tutorials/langchainjs.md

@@ -23,13 +23,17 @@ const answer = await ollama.call(`why is the sky blue?`);
 console.log(answer);
 ```
 
-That will get us the same thing as if we ran `ollama run llama2 "why is the sky blue"` in the terminal. But we want to load a document from the web to ask a question against. **Cheerio** is a great library for ingesting a webpage, and **LangChain** uses it in their **CheerioWebBaseLoader**. So let's build that part of the app.
+That will get us the same thing as if we ran `ollama run llama2 "why is the sky blue"` in the terminal. But we want to load a document from the web to ask a question against. **Cheerio** is a great library for ingesting a webpage, and **LangChain** uses it in their **CheerioWebBaseLoader**. So let's install **Cheerio** and build that part of the app.
+
+```bash
+npm install cheerio 
+```
 
 ```javascript
 import { CheerioWebBaseLoader } from "langchain/document_loaders/web/cheerio";
 
 const loader = new CheerioWebBaseLoader("https://en.wikipedia.org/wiki/2023_Hawaii_wildfires");
-const data = loader.load();
+const data = await loader.load();
 ```
 
 That will load the document. Although this page is smaller than the Odyssey, it is certainly bigger than the context size for most LLMs. So we are going to need to split into smaller pieces, and then select just the pieces relevant to our question. This is a great use for a vector datastore. In this example, we will use the **MemoryVectorStore** that is part of **LangChain**. But there is one more thing we need to get the content into the datastore. We have to run an embeddings process that converts the tokens in the text into a series of vectors. And for that, we are going to use **Tensorflow**. There is a lot of stuff going on in this one. First, install the **Tensorflow** components that we need.

examples/.gitignore

@@ -0,0 +1,171 @@
+node_modules
+# OSX
+.DS_STORE
+
+# Models
+models/
+
+# Local Chroma db
+.chroma/
+db/
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/

examples/README.md

@@ -1,15 +1,3 @@
 # Examples
 
-This directory contains different examples of using Ollama
-
-To create a model:
-
-```
-ollama create example -f <example file>
-```
-
-To run a model:
-
-```
-ollama run example
-```
+This directory contains different examples of using Ollama.

examples/golang-simplegenerate/main.go

@@ -0,0 +1,27 @@
+package main
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"log"
+	"net/http"
+	"os"
+)
+
+func main() {
+	body := []byte(`{"model":"mistral"}`)
+	resp, err := http.Post("http://localhost:11434/api/generate", "application/json", bytes.NewBuffer(body))
+
+	if err != nil {
+		fmt.Print(err.Error())
+		os.Exit(1)
+	}
+
+	responseData, err := io.ReadAll(resp.Body)
+	if err != nil {
+		log.Fatal(err)
+	}
+	fmt.Println(string(responseData))
+
+}

examples/kubernetes/README.md

@@ -0,0 +1,36 @@
+# Deploy Ollama to Kubernetes
+
+## Prerequisites
+
+- Ollama: https://ollama.ai/download
+- Kubernetes cluster. This example will use Google Kubernetes Engine.
+
+## Steps
+
+1. Create the Ollama namespace, daemon set, and service
+
+    ```bash
+    kubectl apply -f cpu.yaml
+    ```
+
+1. Port forward the Ollama service to connect and use it locally
+
+    ```bash
+    kubectl -n ollama port-forward service/ollama 11434:80
+    ```
+
+1. Pull and run a model, for example `orca-mini:3b`
+
+    ```bash
+    ollama run orca-mini:3b
+    ```
+
+## (Optional) Hardware Acceleration
+
+Hardware acceleration in Kubernetes requires NVIDIA's [`k8s-device-plugin`](https://github.com/NVIDIA/k8s-device-plugin). Follow the link for more details.
+
+Once configured, create a GPU enabled Ollama deployment.
+
+```bash
+kubectl apply -f gpu.yaml
+```

examples/kubernetes/cpu.yaml

@@ -0,0 +1,42 @@
+---
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: ollama
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: ollama
+  namespace: ollama
+spec:
+  selector:
+    matchLabels:
+      name: ollama
+  template:
+    metadata:
+      labels:
+        name: ollama
+    spec:
+      containers:
+      - name: ollama
+        image: ollama/ollama:latest
+        ports:
+        - name: http
+          containerPort: 11434
+          protocol: TCP
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: ollama
+  namespace: ollama
+spec:
+  type: ClusterIP
+  selector:
+    name: ollama
+  ports:
+  - port: 80
+    name: http
+    targetPort: http
+    protocol: TCP

examples/kubernetes/gpu.yaml

@@ -0,0 +1,56 @@
+---
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: ollama
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: ollama
+  namespace: ollama
+spec:
+  strategy:
+    type: Recreate
+  selector:
+    matchLabels:
+      name: ollama
+  template:
+    metadata:
+      labels:
+        name: ollama
+    spec:
+      containers:
+      - name: ollama
+        image: ollama/ollama:latest
+        env:
+        - name: PATH
+          value: /usr/local/nvidia/bin:/usr/local/nvidia/lib64:/usr/bin:/usr/sbin:/bin:/sbin
+        - name: LD_LIBRARY_PATH
+          value: /usr/local/nvidia/lib64
+        ports:
+        - name: http
+          containerPort: 11434
+          protocol: TCP
+        resources:
+          limits:
+            nvidia.com/gpu: 1
+      tolerations:
+      - key: nvidia.com/gpu
+        operator: Exists
+        effect: NoSchedule
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: ollama
+  namespace: ollama
+spec:
+  type: ClusterIP
+  selector:
+    name: ollama
+  ports:
+  - port: 80
+    name: http
+    targetPort: http
+    protocol: TCP

examples/langchain-python-rag-privategpt/constants.py

@@ -6,7 +6,6 @@ PERSIST_DIRECTORY = os.environ.get('PERSIST_DIRECTORY', 'db')
 
 # Define the Chroma settings
 CHROMA_SETTINGS = Settings(
-        chroma_db_impl='duckdb+parquet',
         persist_directory=PERSIST_DIRECTORY,
         anonymized_telemetry=False
 )

examples/langchain-python-rag-privategpt/ingest.py

@@ -150,7 +150,7 @@ def main():
         print("Creating new vectorstore")
         texts = process_documents()
         print(f"Creating embeddings. May take some minutes...")
-        db = Chroma.from_documents(texts, embeddings, persist_directory=persist_directory, client_settings=CHROMA_SETTINGS)
+        db = Chroma.from_documents(texts, embeddings, persist_directory=persist_directory)
     db.persist()
     db = None
 

examples/langchain-python-rag-privategpt/privateGPT.py

@@ -4,6 +4,7 @@ from langchain.embeddings import HuggingFaceEmbeddings
 from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
 from langchain.vectorstores import Chroma
 from langchain.llms import Ollama
+import chromadb
 import os
 import argparse
 import time
@@ -22,7 +23,9 @@ def main():
     # Parse the command line arguments
     args = parse_arguments()
     embeddings = HuggingFaceEmbeddings(model_name=embeddings_model_name)
-    db = Chroma(persist_directory=persist_directory, embedding_function=embeddings, client_settings=CHROMA_SETTINGS)
+
+    db = Chroma(persist_directory=persist_directory, embedding_function=embeddings)
+
     retriever = db.as_retriever(search_kwargs={"k": target_source_chunks})
     # activate/deactivate the streaming StdOut callback for LLMs
     callbacks = [] if args.mute_stream else [StreamingStdOutCallbackHandler()]

examples/langchain-python-rag-privategpt/requirements.txt

@@ -0,0 +1,14 @@
+langchain==0.0.274
+gpt4all==1.0.8
+chromadb==0.4.7
+llama-cpp-python==0.1.81
+urllib3==2.0.4
+PyMuPDF==1.23.5
+python-dotenv==1.0.0
+unstructured==0.10.8
+extract-msg==0.45.0
+tabulate==0.9.0
+pandoc==2.3
+pypandoc==1.11
+tqdm==4.66.1
+sentence_transformers==2.2.2

examples/langchain-typescript-simple/README.md

@@ -0,0 +1,21 @@
+# LangChain
+
+This example is a basic "hello world" of using LangChain with Ollama using Node.js and Typescript.
+
+## Setup
+
+```shell
+npm install
+```
+
+## Run
+
+```shell
+ts-node main.ts
+```
+
+Running this example will print the response for "hello":
+
+```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?
+```

examples/langchain-typescript-simple/main.ts

@@ -0,0 +1,15 @@
+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 stream = await ollama.stream("Hello");
+
+  for await (const chunk of stream) {
+    process.stdout.write(chunk);
+  }
+}
+
+main();

examples/langchain-typescript-simple/package-lock.json

@@ -0,0 +1,997 @@
+{
+  "name": "with-langchain-typescript-simplegenerate",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {
+    "": {
+      "dependencies": {
+        "langchain": "^0.0.165"
+      },
+      "devDependencies": {
+        "typescript": "^5.2.2"
+      }
+    },
+    "node_modules/@anthropic-ai/sdk": {
+      "version": "0.6.2",
+      "resolved": "https://registry.npmjs.org/@anthropic-ai/sdk/-/sdk-0.6.2.tgz",
+      "integrity": "sha512-fB9PUj9RFT+XjkL+E9Ol864ZIJi+1P8WnbHspN3N3/GK2uSzjd0cbVIKTGgf4v3N8MwaQu+UWnU7C4BG/fap/g==",
+      "dependencies": {
+        "@types/node": "^18.11.18",
+        "@types/node-fetch": "^2.6.4",
+        "abort-controller": "^3.0.0",
+        "agentkeepalive": "^4.2.1",
+        "digest-fetch": "^1.3.0",
+        "form-data-encoder": "1.7.2",
+        "formdata-node": "^4.3.2",
+        "node-fetch": "^2.6.7"
+      }
+    },
+    "node_modules/@types/node": {
+      "version": "18.18.4",
+      "resolved": "https://registry.npmjs.org/@types/node/-/node-18.18.4.tgz",
+      "integrity": "sha512-t3rNFBgJRugIhackit2mVcLfF6IRc0JE4oeizPQL8Zrm8n2WY/0wOdpOPhdtG0V9Q2TlW/axbF1MJ6z+Yj/kKQ=="
+    },
+    "node_modules/@types/node-fetch": {
+      "version": "2.6.6",
+      "resolved": "https://registry.npmjs.org/@types/node-fetch/-/node-fetch-2.6.6.tgz",
+      "integrity": "sha512-95X8guJYhfqiuVVhRFxVQcf4hW/2bCuoPwDasMf/531STFoNoWTT7YDnWdXHEZKqAGUigmpG31r2FE70LwnzJw==",
+      "dependencies": {
+        "@types/node": "*",
+        "form-data": "^4.0.0"
+      }
+    },
+    "node_modules/@types/retry": {
+      "version": "0.12.0",
+      "resolved": "https://registry.npmjs.org/@types/retry/-/retry-0.12.0.tgz",
+      "integrity": "sha512-wWKOClTTiizcZhXnPY4wikVAwmdYHp8q6DmC+EJUzAMsycb7HB32Kh9RN4+0gExjmPmZSAQjgURXIGATPegAvA=="
+    },
+    "node_modules/@types/uuid": {
+      "version": "9.0.5",
+      "resolved": "https://registry.npmjs.org/@types/uuid/-/uuid-9.0.5.tgz",
+      "integrity": "sha512-xfHdwa1FMJ082prjSJpoEI57GZITiQz10r3vEJCHa2khEFQjKy91aWKz6+zybzssCvXUwE1LQWgWVwZ4nYUvHQ=="
+    },
+    "node_modules/abort-controller": {
+      "version": "3.0.0",
+      "resolved": "https://registry.npmjs.org/abort-controller/-/abort-controller-3.0.0.tgz",
+      "integrity": "sha512-h8lQ8tacZYnR3vNQTgibj+tODHI5/+l06Au2Pcriv/Gmet0eaj4TwWH41sO9wnHDiQsEj19q0drzdWdeAHtweg==",
+      "dependencies": {
+        "event-target-shim": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=6.5"
+      }
+    },
+    "node_modules/agentkeepalive": {
+      "version": "4.5.0",
+      "resolved": "https://registry.npmjs.org/agentkeepalive/-/agentkeepalive-4.5.0.tgz",
+      "integrity": "sha512-5GG/5IbQQpC9FpkRGsSvZI5QYeSCzlJHdpBQntCsuTOxhKD8lqKhrleg2Yi7yvMIf82Ycmmqln9U8V9qwEiJew==",
+      "dependencies": {
+        "humanize-ms": "^1.2.1"
+      },
+      "engines": {
+        "node": ">= 8.0.0"
+      }
+    },
+    "node_modules/ansi-styles": {
+      "version": "5.2.0",
+      "resolved": "https://registry.npmjs.org/ansi-styles/-/ansi-styles-5.2.0.tgz",
+      "integrity": "sha512-Cxwpt2SfTzTtXcfOlzGEee8O+c+MmUgGrNiBcXnuWxuFJHe6a5Hz7qwhwe5OgaSYI0IJvkLqWX1ASG+cJOkEiA==",
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-styles?sponsor=1"
+      }
+    },
+    "node_modules/argparse": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/argparse/-/argparse-2.0.1.tgz",
+      "integrity": "sha512-8+9WqebbFzpX9OR+Wa6O29asIogeRMzcGtAINdpMHHyAg10f05aSFVBbcEqGf/PXw1EjAZ+q2/bEBg3DvurK3Q=="
+    },
+    "node_modules/asynckit": {
+      "version": "0.4.0",
+      "resolved": "https://registry.npmjs.org/asynckit/-/asynckit-0.4.0.tgz",
+      "integrity": "sha512-Oei9OH4tRh0YqU3GxhX79dM/mwVgvbZJaSNaRk+bshkj0S5cfHcgYakreBjrHwatXKbz+IoIdYLxrKim2MjW0Q=="
+    },
+    "node_modules/base-64": {
+      "version": "0.1.0",
+      "resolved": "https://registry.npmjs.org/base-64/-/base-64-0.1.0.tgz",
+      "integrity": "sha512-Y5gU45svrR5tI2Vt/X9GPd3L0HNIKzGu202EjxrXMpuc2V2CiKgemAbUUsqYmZJvPtCXoUKjNZwBJzsNScUbXA=="
+    },
+    "node_modules/base64-js": {
+      "version": "1.5.1",
+      "resolved": "https://registry.npmjs.org/base64-js/-/base64-js-1.5.1.tgz",
+      "integrity": "sha512-AKpaYlHn8t4SVbOHCy+b5+KKgvR4vrsD8vbvrbiQJps7fKDTkjkDry6ji0rUJjC0kzbNePLwzxq8iypo41qeWA==",
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/feross"
+        },
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/feross"
+        },
+        {
+          "type": "consulting",
+          "url": "https://feross.org/support"
+        }
+      ]
+    },
+    "node_modules/binary-extensions": {
+      "version": "2.2.0",
+      "resolved": "https://registry.npmjs.org/binary-extensions/-/binary-extensions-2.2.0.tgz",
+      "integrity": "sha512-jDctJ/IVQbZoJykoeHbhXpOlNBqGNcwXJKJog42E5HDPUwQTSdjCHdihjj0DlnheQ7blbT6dHOafNAiS8ooQKA==",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/binary-search": {
+      "version": "1.3.6",
+      "resolved": "https://registry.npmjs.org/binary-search/-/binary-search-1.3.6.tgz",
+      "integrity": "sha512-nbE1WxOTTrUWIfsfZ4aHGYu5DOuNkbxGokjV6Z2kxfJK3uaAb8zNK1muzOeipoLHZjInT4Br88BHpzevc681xA=="
+    },
+    "node_modules/camelcase": {
+      "version": "6.3.0",
+      "resolved": "https://registry.npmjs.org/camelcase/-/camelcase-6.3.0.tgz",
+      "integrity": "sha512-Gmy6FhYlCY7uOElZUSbxo2UCDH8owEk996gkbrpsgGtrJLM3J7jGxl9Ic7Qwwj4ivOE5AWZWRMecDdF7hqGjFA==",
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/charenc": {
+      "version": "0.0.2",
+      "resolved": "https://registry.npmjs.org/charenc/-/charenc-0.0.2.tgz",
+      "integrity": "sha512-yrLQ/yVUFXkzg7EDQsPieE/53+0RlaWTs+wBrvW36cyilJ2SaDWfl4Yj7MtLTXleV9uEKefbAGUPv2/iWSooRA==",
+      "engines": {
+        "node": "*"
+      }
+    },
+    "node_modules/combined-stream": {
+      "version": "1.0.8",
+      "resolved": "https://registry.npmjs.org/combined-stream/-/combined-stream-1.0.8.tgz",
+      "integrity": "sha512-FQN4MRfuJeHf7cBbBMJFXhKSDq+2kAArBlmRBvcvFE5BB1HZKXtSFASDhdlz9zOYwxh8lDdnvmMOe/+5cdoEdg==",
+      "dependencies": {
+        "delayed-stream": "~1.0.0"
+      },
+      "engines": {
+        "node": ">= 0.8"
+      }
+    },
+    "node_modules/commander": {
+      "version": "10.0.1",
+      "resolved": "https://registry.npmjs.org/commander/-/commander-10.0.1.tgz",
+      "integrity": "sha512-y4Mg2tXshplEbSGzx7amzPwKKOCGuoSRP/CjEdwwk0FOGlUbq6lKuoyDZTNZkmxHdJtp54hdfY/JUrdL7Xfdug==",
+      "engines": {
+        "node": ">=14"
+      }
+    },
+    "node_modules/crypt": {
+      "version": "0.0.2",
+      "resolved": "https://registry.npmjs.org/crypt/-/crypt-0.0.2.tgz",
+      "integrity": "sha512-mCxBlsHFYh9C+HVpiEacem8FEBnMXgU9gy4zmNC+SXAZNB/1idgp/aulFJ4FgCi7GPEVbfyng092GqL2k2rmow==",
+      "engines": {
+        "node": "*"
+      }
+    },
+    "node_modules/decamelize": {
+      "version": "1.2.0",
+      "resolved": "https://registry.npmjs.org/decamelize/-/decamelize-1.2.0.tgz",
+      "integrity": "sha512-z2S+W9X73hAUUki+N+9Za2lBlun89zigOyGrsax+KUQ6wKW4ZoWpEYBkGhQjwAjjDCkWxhY0VKEhk8wzY7F5cA==",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/delayed-stream": {
+      "version": "1.0.0",
+      "resolved": "https://registry.npmjs.org/delayed-stream/-/delayed-stream-1.0.0.tgz",
+      "integrity": "sha512-ZySD7Nf91aLB0RxL4KGrKHBXl7Eds1DAmEdcoVawXnLD7SDhpNgtuII2aAkg7a7QS41jxPSZ17p4VdGnMHk3MQ==",
+      "engines": {
+        "node": ">=0.4.0"
+      }
+    },
+    "node_modules/digest-fetch": {
+      "version": "1.3.0",
+      "resolved": "https://registry.npmjs.org/digest-fetch/-/digest-fetch-1.3.0.tgz",
+      "integrity": "sha512-CGJuv6iKNM7QyZlM2T3sPAdZWd/p9zQiRNS9G+9COUCwzWFTs0Xp8NF5iePx7wtvhDykReiRRrSeNb4oMmB8lA==",
+      "dependencies": {
+        "base-64": "^0.1.0",
+        "md5": "^2.3.0"
+      }
+    },
+    "node_modules/event-target-shim": {
+      "version": "5.0.1",
+      "resolved": "https://registry.npmjs.org/event-target-shim/-/event-target-shim-5.0.1.tgz",
+      "integrity": "sha512-i/2XbnSz/uxRCU6+NdVJgKWDTM427+MqYbkQzD321DuCQJUqOuJKIA0IM2+W2xtYHdKOmZ4dR6fExsd4SXL+WQ==",
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/eventemitter3": {
+      "version": "4.0.7",
+      "resolved": "https://registry.npmjs.org/eventemitter3/-/eventemitter3-4.0.7.tgz",
+      "integrity": "sha512-8guHBZCwKnFhYdHr2ysuRWErTwhoN2X8XELRlrRwpmfeY2jjuUN4taQMsULKUVo1K4DvZl+0pgfyoysHxvmvEw=="
+    },
+    "node_modules/expr-eval": {
+      "version": "2.0.2",
+      "resolved": "https://registry.npmjs.org/expr-eval/-/expr-eval-2.0.2.tgz",
+      "integrity": "sha512-4EMSHGOPSwAfBiibw3ndnP0AvjDWLsMvGOvWEZ2F96IGk0bIVdjQisOHxReSkE13mHcfbuCiXw+G4y0zv6N8Eg=="
+    },
+    "node_modules/flat": {
+      "version": "5.0.2",
+      "resolved": "https://registry.npmjs.org/flat/-/flat-5.0.2.tgz",
+      "integrity": "sha512-b6suED+5/3rTpUBdG1gupIl8MPFCAMA0QXwmljLhvCUKcUvdE4gWky9zpuGCcXHOsz4J9wPGNWq6OKpmIzz3hQ==",
+      "bin": {
+        "flat": "cli.js"
+      }
+    },
+    "node_modules/form-data": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/form-data/-/form-data-4.0.0.tgz",
+      "integrity": "sha512-ETEklSGi5t0QMZuiXoA/Q6vcnxcLQP5vdugSpuAyi6SVGi2clPPp+xgEhuMaHC+zGgn31Kd235W35f7Hykkaww==",
+      "dependencies": {
+        "asynckit": "^0.4.0",
+        "combined-stream": "^1.0.8",
+        "mime-types": "^2.1.12"
+      },
+      "engines": {
+        "node": ">= 6"
+      }
+    },
+    "node_modules/form-data-encoder": {
+      "version": "1.7.2",
+      "resolved": "https://registry.npmjs.org/form-data-encoder/-/form-data-encoder-1.7.2.tgz",
+      "integrity": "sha512-qfqtYan3rxrnCk1VYaA4H+Ms9xdpPqvLZa6xmMgFvhO32x7/3J/ExcTd6qpxM0vH2GdMI+poehyBZvqfMTto8A=="
+    },
+    "node_modules/formdata-node": {
+      "version": "4.4.1",
+      "resolved": "https://registry.npmjs.org/formdata-node/-/formdata-node-4.4.1.tgz",
+      "integrity": "sha512-0iirZp3uVDjVGt9p49aTaqjk84TrglENEDuqfdlZQ1roC9CWlPk6Avf8EEnZNcAqPonwkG35x4n3ww/1THYAeQ==",
+      "dependencies": {
+        "node-domexception": "1.0.0",
+        "web-streams-polyfill": "4.0.0-beta.3"
+      },
+      "engines": {
+        "node": ">= 12.20"
+      }
+    },
+    "node_modules/humanize-ms": {
+      "version": "1.2.1",
+      "resolved": "https://registry.npmjs.org/humanize-ms/-/humanize-ms-1.2.1.tgz",
+      "integrity": "sha512-Fl70vYtsAFb/C06PTS9dZBo7ihau+Tu/DNCk/OyHhea07S+aeMWpFFkUaXRa8fI+ScZbEI8dfSxwY7gxZ9SAVQ==",
+      "dependencies": {
+        "ms": "^2.0.0"
+      }
+    },
+    "node_modules/is-any-array": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/is-any-array/-/is-any-array-2.0.1.tgz",
+      "integrity": "sha512-UtilS7hLRu++wb/WBAw9bNuP1Eg04Ivn1vERJck8zJthEvXCBEBpGR/33u/xLKWEQf95803oalHrVDptcAvFdQ=="
+    },
+    "node_modules/is-buffer": {
+      "version": "1.1.6",
+      "resolved": "https://registry.npmjs.org/is-buffer/-/is-buffer-1.1.6.tgz",
+      "integrity": "sha512-NcdALwpXkTm5Zvvbk7owOUSvVvBKDgKP5/ewfXEznmQFfs4ZRmanOeKBTjRVjka3QFoN6XJ+9F3USqfHqTaU5w=="
+    },
+    "node_modules/js-tiktoken": {
+      "version": "1.0.7",
+      "resolved": "https://registry.npmjs.org/js-tiktoken/-/js-tiktoken-1.0.7.tgz",
+      "integrity": "sha512-biba8u/clw7iesNEWLOLwrNGoBP2lA+hTaBLs/D45pJdUPFXyxD6nhcDVtADChghv4GgyAiMKYMiRx7x6h7Biw==",
+      "dependencies": {
+        "base64-js": "^1.5.1"
+      }
+    },
+    "node_modules/js-yaml": {
+      "version": "4.1.0",
+      "resolved": "https://registry.npmjs.org/js-yaml/-/js-yaml-4.1.0.tgz",
+      "integrity": "sha512-wpxZs9NoxZaJESJGIZTyDEaYpl0FKSA+FB9aJiyemKhMwkxQg63h4T1KJgUGHpTqPDNRcmmYLugrRjJlBtWvRA==",
+      "dependencies": {
+        "argparse": "^2.0.1"
+      },
+      "bin": {
+        "js-yaml": "bin/js-yaml.js"
+      }
+    },
+    "node_modules/jsonpointer": {
+      "version": "5.0.1",
+      "resolved": "https://registry.npmjs.org/jsonpointer/-/jsonpointer-5.0.1.tgz",
+      "integrity": "sha512-p/nXbhSEcu3pZRdkW1OfJhpsVtW1gd4Wa1fnQc9YLiTfAjn0312eMKimbdIQzuZl9aa9xUGaRlP9T/CJE/ditQ==",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/langchain": {
+      "version": "0.0.165",
+      "resolved": "https://registry.npmjs.org/langchain/-/langchain-0.0.165.tgz",
+      "integrity": "sha512-CpbNpjwaE+9lzjdw+pZz0VgnRrFivEgr7CVp9dDaAb5JpaJAA4V2v6uQ9ZPN+TSqupTQ79HFn2sfyZVEl2EG7Q==",
+      "dependencies": {
+        "@anthropic-ai/sdk": "^0.6.2",
+        "ansi-styles": "^5.0.0",
+        "binary-extensions": "^2.2.0",
+        "camelcase": "6",
+        "decamelize": "^1.2.0",
+        "expr-eval": "^2.0.2",
+        "flat": "^5.0.2",
+        "js-tiktoken": "^1.0.7",
+        "js-yaml": "^4.1.0",
+        "jsonpointer": "^5.0.1",
+        "langchainhub": "~0.0.6",
+        "langsmith": "~0.0.31",
+        "ml-distance": "^4.0.0",
+        "object-hash": "^3.0.0",
+        "openai": "~4.4.0",
+        "openapi-types": "^12.1.3",
+        "p-queue": "^6.6.2",
+        "p-retry": "4",
+        "uuid": "^9.0.0",
+        "yaml": "^2.2.1",
+        "zod": "^3.22.3",
+        "zod-to-json-schema": "^3.20.4"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "peerDependencies": {
+        "@aws-crypto/sha256-js": "^5.0.0",
+        "@aws-sdk/client-bedrock-runtime": "^3.422.0",
+        "@aws-sdk/client-dynamodb": "^3.310.0",
+        "@aws-sdk/client-kendra": "^3.352.0",
+        "@aws-sdk/client-lambda": "^3.310.0",
+        "@aws-sdk/client-s3": "^3.310.0",
+        "@aws-sdk/client-sagemaker-runtime": "^3.310.0",
+        "@aws-sdk/client-sfn": "^3.310.0",
+        "@aws-sdk/credential-provider-node": "^3.388.0",
+        "@azure/storage-blob": "^12.15.0",
+        "@clickhouse/client": "^0.0.14",
+        "@cloudflare/ai": "^1.0.12",
+        "@elastic/elasticsearch": "^8.4.0",
+        "@getmetal/metal-sdk": "*",
+        "@getzep/zep-js": "^0.7.0",
+        "@gomomento/sdk": "^1.23.0",
+        "@google-ai/generativelanguage": "^0.2.1",
+        "@google-cloud/storage": "^6.10.1",
+        "@huggingface/inference": "^1.5.1",
+        "@mozilla/readability": "*",
+        "@notionhq/client": "^2.2.10",
+        "@opensearch-project/opensearch": "*",
+        "@pinecone-database/pinecone": "^1.1.0",
+        "@planetscale/database": "^1.8.0",
+        "@qdrant/js-client-rest": "^1.2.0",
+        "@raycast/api": "^1.55.2",
+        "@smithy/eventstream-codec": "^2.0.5",
+        "@smithy/protocol-http": "^3.0.6",
+        "@smithy/signature-v4": "^2.0.10",
+        "@smithy/util-utf8": "^2.0.0",
+        "@supabase/postgrest-js": "^1.1.1",
+        "@supabase/supabase-js": "^2.10.0",
+        "@tensorflow-models/universal-sentence-encoder": "*",
+        "@tensorflow/tfjs-converter": "*",
+        "@tensorflow/tfjs-core": "*",
+        "@upstash/redis": "^1.20.6",
+        "@vercel/postgres": "^0.5.0",
+        "@writerai/writer-sdk": "^0.40.2",
+        "@xata.io/client": "^0.25.1",
+        "@xenova/transformers": "^2.5.4",
+        "@zilliz/milvus2-sdk-node": ">=2.2.7",
+        "apify-client": "^2.7.1",
+        "axios": "*",
+        "cassandra-driver": "^4.6.4",
+        "cheerio": "^1.0.0-rc.12",
+        "chromadb": "*",
+        "cohere-ai": ">=6.0.0",
+        "d3-dsv": "^2.0.0",
+        "epub2": "^3.0.1",
+        "faiss-node": "^0.3.0",
+        "fast-xml-parser": "^4.2.7",
+        "firebase-admin": "^11.9.0",
+        "google-auth-library": "^8.9.0",
+        "googleapis": "^126.0.1",
+        "hnswlib-node": "^1.4.2",
+        "html-to-text": "^9.0.5",
+        "ignore": "^5.2.0",
+        "ioredis": "^5.3.2",
+        "jsdom": "*",
+        "llmonitor": "*",
+        "lodash": "^4.17.21",
+        "mammoth": "*",
+        "mongodb": "^5.2.0",
+        "mysql2": "^3.3.3",
+        "neo4j-driver": "*",
+        "node-llama-cpp": "*",
+        "notion-to-md": "^3.1.0",
+        "pdf-parse": "1.1.1",
+        "peggy": "^3.0.2",
+        "pg": "^8.11.0",
+        "pg-copy-streams": "^6.0.5",
+        "pickleparser": "^0.1.0",
+        "playwright": "^1.32.1",
+        "portkey-ai": "^0.1.11",
+        "puppeteer": "^19.7.2",
+        "redis": "^4.6.4",
+        "replicate": "^0.18.0",
+        "sonix-speech-recognition": "^2.1.1",
+        "srt-parser-2": "^1.2.2",
+        "typeorm": "^0.3.12",
+        "typesense": "^1.5.3",
+        "usearch": "^1.1.1",
+        "vectordb": "^0.1.4",
+        "voy-search": "0.6.2",
+        "weaviate-ts-client": "^1.4.0",
+        "web-auth-library": "^1.0.3",
+        "youtube-transcript": "^1.0.6",
+        "youtubei.js": "^5.8.0"
+      },
+      "peerDependenciesMeta": {
+        "@aws-crypto/sha256-js": {
+          "optional": true
+        },
+        "@aws-sdk/client-bedrock-runtime": {
+          "optional": true
+        },
+        "@aws-sdk/client-dynamodb": {
+          "optional": true
+        },
+        "@aws-sdk/client-kendra": {
+          "optional": true
+        },
+        "@aws-sdk/client-lambda": {
+          "optional": true
+        },
+        "@aws-sdk/client-s3": {
+          "optional": true
+        },
+        "@aws-sdk/client-sagemaker-runtime": {
+          "optional": true
+        },
+        "@aws-sdk/client-sfn": {
+          "optional": true
+        },
+        "@aws-sdk/credential-provider-node": {
+          "optional": true
+        },
+        "@azure/storage-blob": {
+          "optional": true
+        },
+        "@clickhouse/client": {
+          "optional": true
+        },
+        "@cloudflare/ai": {
+          "optional": true
+        },
+        "@elastic/elasticsearch": {
+          "optional": true
+        },
+        "@getmetal/metal-sdk": {
+          "optional": true
+        },
+        "@getzep/zep-js": {
+          "optional": true
+        },
+        "@gomomento/sdk": {
+          "optional": true
+        },
+        "@google-ai/generativelanguage": {
+          "optional": true
+        },
+        "@google-cloud/storage": {
+          "optional": true
+        },
+        "@huggingface/inference": {
+          "optional": true
+        },
+        "@mozilla/readability": {
+          "optional": true
+        },
+        "@notionhq/client": {
+          "optional": true
+        },
+        "@opensearch-project/opensearch": {
+          "optional": true
+        },
+        "@pinecone-database/pinecone": {
+          "optional": true
+        },
+        "@planetscale/database": {
+          "optional": true
+        },
+        "@qdrant/js-client-rest": {
+          "optional": true
+        },
+        "@raycast/api": {
+          "optional": true
+        },
+        "@smithy/eventstream-codec": {
+          "optional": true
+        },
+        "@smithy/protocol-http": {
+          "optional": true
+        },
+        "@smithy/signature-v4": {
+          "optional": true
+        },
+        "@smithy/util-utf8": {
+          "optional": true
+        },
+        "@supabase/postgrest-js": {
+          "optional": true
+        },
+        "@supabase/supabase-js": {
+          "optional": true
+        },
+        "@tensorflow-models/universal-sentence-encoder": {
+          "optional": true
+        },
+        "@tensorflow/tfjs-converter": {
+          "optional": true
+        },
+        "@tensorflow/tfjs-core": {
+          "optional": true
+        },
+        "@upstash/redis": {
+          "optional": true
+        },
+        "@vercel/postgres": {
+          "optional": true
+        },
+        "@writerai/writer-sdk": {
+          "optional": true
+        },
+        "@xata.io/client": {
+          "optional": true
+        },
+        "@xenova/transformers": {
+          "optional": true
+        },
+        "@zilliz/milvus2-sdk-node": {
+          "optional": true
+        },
+        "apify-client": {
+          "optional": true
+        },
+        "axios": {
+          "optional": true
+        },
+        "cassandra-driver": {
+          "optional": true
+        },
+        "cheerio": {
+          "optional": true
+        },
+        "chromadb": {
+          "optional": true
+        },
+        "cohere-ai": {
+          "optional": true
+        },
+        "d3-dsv": {
+          "optional": true
+        },
+        "epub2": {
+          "optional": true
+        },
+        "faiss-node": {
+          "optional": true
+        },
+        "fast-xml-parser": {
+          "optional": true
+        },
+        "firebase-admin": {
+          "optional": true
+        },
+        "google-auth-library": {
+          "optional": true
+        },
+        "googleapis": {
+          "optional": true
+        },
+        "hnswlib-node": {
+          "optional": true
+        },
+        "html-to-text": {
+          "optional": true
+        },
+        "ignore": {
+          "optional": true
+        },
+        "ioredis": {
+          "optional": true
+        },
+        "jsdom": {
+          "optional": true
+        },
+        "llmonitor": {
+          "optional": true
+        },
+        "lodash": {
+          "optional": true
+        },
+        "mammoth": {
+          "optional": true
+        },
+        "mongodb": {
+          "optional": true
+        },
+        "mysql2": {
+          "optional": true
+        },
+        "neo4j-driver": {
+          "optional": true
+        },
+        "node-llama-cpp": {
+          "optional": true
+        },
+        "notion-to-md": {
+          "optional": true
+        },
+        "pdf-parse": {
+          "optional": true
+        },
+        "peggy": {
+          "optional": true
+        },
+        "pg": {
+          "optional": true
+        },
+        "pg-copy-streams": {
+          "optional": true
+        },
+        "pickleparser": {
+          "optional": true
+        },
+        "playwright": {
+          "optional": true
+        },
+        "portkey-ai": {
+          "optional": true
+        },
+        "puppeteer": {
+          "optional": true
+        },
+        "redis": {
+          "optional": true
+        },
+        "replicate": {
+          "optional": true
+        },
+        "sonix-speech-recognition": {
+          "optional": true
+        },
+        "srt-parser-2": {
+          "optional": true
+        },
+        "typeorm": {
+          "optional": true
+        },
+        "typesense": {
+          "optional": true
+        },
+        "usearch": {
+          "optional": true
+        },
+        "vectordb": {
+          "optional": true
+        },
+        "voy-search": {
+          "optional": true
+        },
+        "weaviate-ts-client": {
+          "optional": true
+        },
+        "web-auth-library": {
+          "optional": true
+        },
+        "youtube-transcript": {
+          "optional": true
+        },
+        "youtubei.js": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/langchainhub": {
+      "version": "0.0.6",
+      "resolved": "https://registry.npmjs.org/langchainhub/-/langchainhub-0.0.6.tgz",
+      "integrity": "sha512-SW6105T+YP1cTe0yMf//7kyshCgvCTyFBMTgH2H3s9rTAR4e+78DA/BBrUL/Mt4Q5eMWui7iGuAYb3pgGsdQ9w=="
+    },
+    "node_modules/langsmith": {
+      "version": "0.0.42",
+      "resolved": "https://registry.npmjs.org/langsmith/-/langsmith-0.0.42.tgz",
+      "integrity": "sha512-sFuN+e7E+pPBIRaRgFqZh/BRBWNHTZNAwi6uj4kydQawooCZYoJmM5snOkiQrhVSvAhgu6xFhLvmfvkPcKzD7w==",
+      "dependencies": {
+        "@types/uuid": "^9.0.1",
+        "commander": "^10.0.1",
+        "p-queue": "^6.6.2",
+        "p-retry": "4",
+        "uuid": "^9.0.0"
+      },
+      "bin": {
+        "langsmith": "dist/cli/main.cjs"
+      }
+    },
+    "node_modules/md5": {
+      "version": "2.3.0",
+      "resolved": "https://registry.npmjs.org/md5/-/md5-2.3.0.tgz",
+      "integrity": "sha512-T1GITYmFaKuO91vxyoQMFETst+O71VUPEU3ze5GNzDm0OWdP8v1ziTaAEPUr/3kLsY3Sftgz242A1SetQiDL7g==",
+      "dependencies": {
+        "charenc": "0.0.2",
+        "crypt": "0.0.2",
+        "is-buffer": "~1.1.6"
+      }
+    },
+    "node_modules/mime-db": {
+      "version": "1.52.0",
+      "resolved": "https://registry.npmjs.org/mime-db/-/mime-db-1.52.0.tgz",
+      "integrity": "sha512-sPU4uV7dYlvtWJxwwxHD0PuihVNiE7TyAbQ5SWxDCB9mUYvOgroQOwYQQOKPJ8CIbE+1ETVlOoK1UC2nU3gYvg==",
+      "engines": {
+        "node": ">= 0.6"
+      }
+    },
+    "node_modules/mime-types": {
+      "version": "2.1.35",
+      "resolved": "https://registry.npmjs.org/mime-types/-/mime-types-2.1.35.tgz",
+      "integrity": "sha512-ZDY+bPm5zTTF+YpCrAU9nK0UgICYPT0QtT1NZWFv4s++TNkcgVaT0g6+4R2uI4MjQjzysHB1zxuWL50hzaeXiw==",
+      "dependencies": {
+        "mime-db": "1.52.0"
+      },
+      "engines": {
+        "node": ">= 0.6"
+      }
+    },
+    "node_modules/ml-array-mean": {
+      "version": "1.1.6",
+      "resolved": "https://registry.npmjs.org/ml-array-mean/-/ml-array-mean-1.1.6.tgz",
+      "integrity": "sha512-MIdf7Zc8HznwIisyiJGRH9tRigg3Yf4FldW8DxKxpCCv/g5CafTw0RRu51nojVEOXuCQC7DRVVu5c7XXO/5joQ==",
+      "dependencies": {
+        "ml-array-sum": "^1.1.6"
+      }
+    },
+    "node_modules/ml-array-sum": {
+      "version": "1.1.6",
+      "resolved": "https://registry.npmjs.org/ml-array-sum/-/ml-array-sum-1.1.6.tgz",
+      "integrity": "sha512-29mAh2GwH7ZmiRnup4UyibQZB9+ZLyMShvt4cH4eTK+cL2oEMIZFnSyB3SS8MlsTh6q/w/yh48KmqLxmovN4Dw==",
+      "dependencies": {
+        "is-any-array": "^2.0.0"
+      }
+    },
+    "node_modules/ml-distance": {
+      "version": "4.0.1",
+      "resolved": "https://registry.npmjs.org/ml-distance/-/ml-distance-4.0.1.tgz",
+      "integrity": "sha512-feZ5ziXs01zhyFUUUeZV5hwc0f5JW0Sh0ckU1koZe/wdVkJdGxcP06KNQuF0WBTj8FttQUzcvQcpcrOp/XrlEw==",
+      "dependencies": {
+        "ml-array-mean": "^1.1.6",
+        "ml-distance-euclidean": "^2.0.0",
+        "ml-tree-similarity": "^1.0.0"
+      }
+    },
+    "node_modules/ml-distance-euclidean": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/ml-distance-euclidean/-/ml-distance-euclidean-2.0.0.tgz",
+      "integrity": "sha512-yC9/2o8QF0A3m/0IXqCTXCzz2pNEzvmcE/9HFKOZGnTjatvBbsn4lWYJkxENkA4Ug2fnYl7PXQxnPi21sgMy/Q=="
+    },
+    "node_modules/ml-tree-similarity": {
+      "version": "1.0.0",
+      "resolved": "https://registry.npmjs.org/ml-tree-similarity/-/ml-tree-similarity-1.0.0.tgz",
+      "integrity": "sha512-XJUyYqjSuUQkNQHMscr6tcjldsOoAekxADTplt40QKfwW6nd++1wHWV9AArl0Zvw/TIHgNaZZNvr8QGvE8wLRg==",
+      "dependencies": {
+        "binary-search": "^1.3.5",
+        "num-sort": "^2.0.0"
+      }
+    },
+    "node_modules/ms": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/ms/-/ms-2.1.3.tgz",
+      "integrity": "sha512-6FlzubTLZG3J2a/NVCAleEhjzq5oxgHyaCU9yYXvcLsvoVaHJq/s5xXI6/XXP6tz7R9xAOtHnSO/tXtF3WRTlA=="
+    },
+    "node_modules/node-domexception": {
+      "version": "1.0.0",
+      "resolved": "https://registry.npmjs.org/node-domexception/-/node-domexception-1.0.0.tgz",
+      "integrity": "sha512-/jKZoMpw0F8GRwl4/eLROPA3cfcXtLApP0QzLmUT/HuPCZWyB7IY9ZrMeKw2O/nFIqPQB3PVM9aYm0F312AXDQ==",
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/jimmywarting"
+        },
+        {
+          "type": "github",
+          "url": "https://paypal.me/jimmywarting"
+        }
+      ],
+      "engines": {
+        "node": ">=10.5.0"
+      }
+    },
+    "node_modules/node-fetch": {
+      "version": "2.7.0",
+      "resolved": "https://registry.npmjs.org/node-fetch/-/node-fetch-2.7.0.tgz",
+      "integrity": "sha512-c4FRfUm/dbcWZ7U+1Wq0AwCyFL+3nt2bEw05wfxSz+DWpWsitgmSgYmy2dQdWyKC1694ELPqMs/YzUSNozLt8A==",
+      "dependencies": {
+        "whatwg-url": "^5.0.0"
+      },
+      "engines": {
+        "node": "4.x || >=6.0.0"
+      },
+      "peerDependencies": {
+        "encoding": "^0.1.0"
+      },
+      "peerDependenciesMeta": {
+        "encoding": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/num-sort": {
+      "version": "2.1.0",
+      "resolved": "https://registry.npmjs.org/num-sort/-/num-sort-2.1.0.tgz",
+      "integrity": "sha512-1MQz1Ed8z2yckoBeSfkQHHO9K1yDRxxtotKSJ9yvcTUUxSvfvzEq5GwBrjjHEpMlq/k5gvXdmJ1SbYxWtpNoVg==",
+      "engines": {
+        "node": ">=8"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/object-hash": {
+      "version": "3.0.0",
+      "resolved": "https://registry.npmjs.org/object-hash/-/object-hash-3.0.0.tgz",
+      "integrity": "sha512-RSn9F68PjH9HqtltsSnqYC1XXoWe9Bju5+213R98cNGttag9q9yAOTzdbsqvIa7aNm5WffBZFpWYr2aWrklWAw==",
+      "engines": {
+        "node": ">= 6"
+      }
+    },
+    "node_modules/openai": {
+      "version": "4.4.0",
+      "resolved": "https://registry.npmjs.org/openai/-/openai-4.4.0.tgz",
+      "integrity": "sha512-JN0t628Kh95T0IrXl0HdBqnlJg+4Vq0Bnh55tio+dfCnyzHvMLiWyCM9m726MAJD2YkDU4/8RQB6rNbEq9ct2w==",
+      "dependencies": {
+        "@types/node": "^18.11.18",
+        "@types/node-fetch": "^2.6.4",
+        "abort-controller": "^3.0.0",
+        "agentkeepalive": "^4.2.1",
+        "digest-fetch": "^1.3.0",
+        "form-data-encoder": "1.7.2",
+        "formdata-node": "^4.3.2",
+        "node-fetch": "^2.6.7"
+      },
+      "bin": {
+        "openai": "bin/cli"
+      }
+    },
+    "node_modules/openapi-types": {
+      "version": "12.1.3",
+      "resolved": "https://registry.npmjs.org/openapi-types/-/openapi-types-12.1.3.tgz",
+      "integrity": "sha512-N4YtSYJqghVu4iek2ZUvcN/0aqH1kRDuNqzcycDxhOUpg7GdvLa2F3DgS6yBNhInhv2r/6I0Flkn7CqL8+nIcw=="
+    },
+    "node_modules/p-finally": {
+      "version": "1.0.0",
+      "resolved": "https://registry.npmjs.org/p-finally/-/p-finally-1.0.0.tgz",
+      "integrity": "sha512-LICb2p9CB7FS+0eR1oqWnHhp0FljGLZCWBE9aix0Uye9W8LTQPwMTYVGWQWIw9RdQiDg4+epXQODwIYJtSJaow==",
+      "engines": {
+        "node": ">=4"
+      }
+    },
+    "node_modules/p-queue": {
+      "version": "6.6.2",
+      "resolved": "https://registry.npmjs.org/p-queue/-/p-queue-6.6.2.tgz",
+      "integrity": "sha512-RwFpb72c/BhQLEXIZ5K2e+AhgNVmIejGlTgiB9MzZ0e93GRvqZ7uSi0dvRF7/XIXDeNkra2fNHBxTyPDGySpjQ==",
+      "dependencies": {
+        "eventemitter3": "^4.0.4",
+        "p-timeout": "^3.2.0"
+      },
+      "engines": {
+        "node": ">=8"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/p-retry": {
+      "version": "4.6.2",
+      "resolved": "https://registry.npmjs.org/p-retry/-/p-retry-4.6.2.tgz",
+      "integrity": "sha512-312Id396EbJdvRONlngUx0NydfrIQ5lsYu0znKVUzVvArzEIt08V1qhtyESbGVd1FGX7UKtiFp5uwKZdM8wIuQ==",
+      "dependencies": {
+        "@types/retry": "0.12.0",
+        "retry": "^0.13.1"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/p-timeout": {
+      "version": "3.2.0",
+      "resolved": "https://registry.npmjs.org/p-timeout/-/p-timeout-3.2.0.tgz",
+      "integrity": "sha512-rhIwUycgwwKcP9yTOOFK/AKsAopjjCakVqLHePO3CC6Mir1Z99xT+R63jZxAT5lFZLa2inS5h+ZS2GvR99/FBg==",
+      "dependencies": {
+        "p-finally": "^1.0.0"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/retry": {
+      "version": "0.13.1",
+      "resolved": "https://registry.npmjs.org/retry/-/retry-0.13.1.tgz",
+      "integrity": "sha512-XQBQ3I8W1Cge0Seh+6gjj03LbmRFWuoszgK9ooCpwYIrhhoO80pfq4cUkU5DkknwfOfFteRwlZ56PYOGYyFWdg==",
+      "engines": {
+        "node": ">= 4"
+      }
+    },
+    "node_modules/tr46": {
+      "version": "0.0.3",
+      "resolved": "https://registry.npmjs.org/tr46/-/tr46-0.0.3.tgz",
+      "integrity": "sha512-N3WMsuqV66lT30CrXNbEjx4GEwlow3v6rr4mCcv6prnfwhS01rkgyFdjPNBYd9br7LpXV1+Emh01fHnq2Gdgrw=="
+    },
+    "node_modules/typescript": {
+      "version": "5.2.2",
+      "resolved": "https://registry.npmjs.org/typescript/-/typescript-5.2.2.tgz",
+      "integrity": "sha512-mI4WrpHsbCIcwT9cF4FZvr80QUeKvsUsUvKDoR+X/7XHQH98xYD8YHZg7ANtz2GtZt/CBq2QJ0thkGJMHfqc1w==",
+      "dev": true,
+      "bin": {
+        "tsc": "bin/tsc",
+        "tsserver": "bin/tsserver"
+      },
+      "engines": {
+        "node": ">=14.17"
+      }
+    },
+    "node_modules/uuid": {
+      "version": "9.0.1",
+      "resolved": "https://registry.npmjs.org/uuid/-/uuid-9.0.1.tgz",
+      "integrity": "sha512-b+1eJOlsR9K8HJpow9Ok3fiWOWSIcIzXodvv0rQjVoOVNpWMpxf1wZNpt4y9h10odCNrqnYp1OBzRktckBe3sA==",
+      "funding": [
+        "https://github.com/sponsors/broofa",
+        "https://github.com/sponsors/ctavan"
+      ],
+      "bin": {
+        "uuid": "dist/bin/uuid"
+      }
+    },
+    "node_modules/web-streams-polyfill": {
+      "version": "4.0.0-beta.3",
+      "resolved": "https://registry.npmjs.org/web-streams-polyfill/-/web-streams-polyfill-4.0.0-beta.3.tgz",
+      "integrity": "sha512-QW95TCTaHmsYfHDybGMwO5IJIM93I/6vTRk+daHTWFPhwh+C8Cg7j7XyKrwrj8Ib6vYXe0ocYNrmzY4xAAN6ug==",
+      "engines": {
+        "node": ">= 14"
+      }
+    },
+    "node_modules/webidl-conversions": {
+      "version": "3.0.1",
+      "resolved": "https://registry.npmjs.org/webidl-conversions/-/webidl-conversions-3.0.1.tgz",
+      "integrity": "sha512-2JAn3z8AR6rjK8Sm8orRC0h/bcl/DqL7tRPdGZ4I1CjdF+EaMLmYxBHyXuKL849eucPFhvBoxMsflfOb8kxaeQ=="
+    },
+    "node_modules/whatwg-url": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/whatwg-url/-/whatwg-url-5.0.0.tgz",
+      "integrity": "sha512-saE57nupxk6v3HY35+jzBwYa0rKSy0XR8JSxZPwgLr7ys0IBzhGviA1/TUGJLmSVqs8pb9AnvICXEuOHLprYTw==",
+      "dependencies": {
+        "tr46": "~0.0.3",
+        "webidl-conversions": "^3.0.0"
+      }
+    },
+    "node_modules/yaml": {
+      "version": "2.3.2",
+      "resolved": "https://registry.npmjs.org/yaml/-/yaml-2.3.2.tgz",
+      "integrity": "sha512-N/lyzTPaJasoDmfV7YTrYCI0G/3ivm/9wdG0aHuheKowWQwGTsK0Eoiw6utmzAnI6pkJa0DUVygvp3spqqEKXg==",
+      "engines": {
+        "node": ">= 14"
+      }
+    },
+    "node_modules/zod": {
+      "version": "3.22.4",
+      "resolved": "https://registry.npmjs.org/zod/-/zod-3.22.4.tgz",
+      "integrity": "sha512-iC+8Io04lddc+mVqQ9AZ7OQ2MrUKGN+oIQyq1vemgt46jwCwLfhq7/pwnBnNXXXZb8VTVLKwp9EDkx+ryxIWmg==",
+      "funding": {
+        "url": "https://github.com/sponsors/colinhacks"
+      }
+    },
+    "node_modules/zod-to-json-schema": {
+      "version": "3.21.4",
+      "resolved": "https://registry.npmjs.org/zod-to-json-schema/-/zod-to-json-schema-3.21.4.tgz",
+      "integrity": "sha512-fjUZh4nQ1s6HMccgIeE0VP4QG/YRGPmyjO9sAh890aQKPEk3nqbfUXhMFaC+Dr5KvYBm8BCyvfpZf2jY9aGSsw==",
+      "peerDependencies": {
+        "zod": "^3.21.4"
+      }
+    }
+  }
+}

examples/langchain-typescript-simple/package.json

@@ -0,0 +1,8 @@
+{
+  "devDependencies": {
+    "typescript": "^5.2.2"
+  },
+  "dependencies": {
+    "langchain": "^0.0.165"
+  }
+}

examples/midjourney-prompter/Modelfile

@@ -1,8 +0,0 @@
-# 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 nous-hermes
-SYSTEM """
-Embrace your role as an AI-powered creative assistant, employing Midjourney to manifest compelling AI-generated art. I will outline a specific image concept, and in response, you must produce an exhaustive, multifaceted prompt for Midjourney, ensuring every detail of the original concept is represented in your instructions. Midjourney doesn't do well with text, so after the prompt, give me instructions that I can use to create the titles in a image editor.
-"""

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"
+```

examples/modelfile-devopsengineer/Modelfile

@@ -1,7 +1,7 @@
 # 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 llama2:13b
+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

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.
+
+````

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. 
+"""

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.
+```

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!
+```

examples/python-dockerit/Modelfile

@@ -1,4 +1,4 @@
-FROM llama2
+FROM mistral
 SYSTEM """
 You are an experienced Devops engineer focused on docker. When given specifications for a particular need or application you know the best way to host that within a docker container. For instance if someone tells you they want an nginx server to host files located at /web you will answer as follows
 

examples/python-rag-newssummary/README.md

@@ -0,0 +1,22 @@
+# News Summarizer
+
+This example goes through a series of steps:
+
+  1. You choose a topic area (e.g., "news", "NVidia", "music", etc.).
+  2. Gets the most recent articles on that topic from various sources.
+  3. Uses Ollama to summarize each article.
+  4. Creates chunks of sentences from each article.
+  5. Uses Sentence Transformers to generate embeddings for each of those chunks.
+  6. You enter a question regarding the summaries shown.
+  7. Uses Sentence Transformers to generate an embedding for that question.
+  8. Uses the embedded question to find the most similar chunks.
+  9. Feeds all that to Ollama to generate a good answer to your question based on these news articles.
+
+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.
+
+You can run the example like this:
+
+```bash
+pip install -r requirements.txt
+python summ.py
+```

examples/python-rag-newssummary/requirements.txt

@@ -0,0 +1,9 @@
+beautifulsoup4==4.12.2
+feedparser==6.0.10
+mattsollamatools==0.0.8
+newspaper3k==0.2.8
+nltk==3.8.1
+numpy==1.24.3
+Requests==2.31.0
+scikit_learn==1.3.0
+sentence_transformers==2.2.2

examples/python-rag-newssummary/summ.py

@@ -0,0 +1,86 @@
+import curses
+import json
+from utils import get_url_for_topic, topic_urls, menu, getUrls, get_summary, getArticleText, knn_search
+import requests
+from sentence_transformers import SentenceTransformer
+from mattsollamatools import chunker
+
+if __name__ == "__main__":
+    chosen_topic = curses.wrapper(menu)
+    print("Here is your news summary:\n")
+    urls = getUrls(chosen_topic, n=5)
+    model = SentenceTransformer('all-MiniLM-L6-v2')
+    allEmbeddings = []
+
+    for url in urls:
+      article={}
+      article['embeddings'] = []
+      article['url'] = url
+      text = getArticleText(url)
+      summary = get_summary(text)
+      chunks = chunker(text)  # Use the chunk_text function from web_utils
+      embeddings = model.encode(chunks)
+      for (chunk, embedding) in zip(chunks, embeddings):
+        item = {}
+        item['source'] = chunk
+        item['embedding'] = embedding.tolist()  # Convert NumPy array to list
+        item['sourcelength'] = len(chunk)
+        article['embeddings'].append(item)
+    
+      allEmbeddings.append(article)
+
+      print(f"{summary}\n")
+
+    
+    while True:
+      context = []
+      # Input a question from the user
+      question = input("Enter your question about the news, or type quit: ")
+
+      if question.lower() == 'quit':
+        break
+
+      # Embed the user's question
+      question_embedding = model.encode([question])
+
+      # Perform KNN search to find the best matches (indices and source text)
+      best_matches = knn_search(question_embedding, allEmbeddings, k=10)
+
+
+      sourcetext=""
+      for i, (index, source_text) in enumerate(best_matches, start=1):
+          sourcetext += f"{i}. Index: {index}, Source Text: {source_text}"
+
+      systemPrompt = f"Only use the following information to answer the question. Do not use anything else: {sourcetext}"
+
+      url = "http://localhost:11434/api/generate"
+
+      payload = {
+      "model": "mistral-openorca",
+      "prompt": question, 
+      "system": systemPrompt,
+      "stream": False, 
+      "context": context
+      }
+
+      # Convert the payload to a JSON string
+      payload_json = json.dumps(payload)
+
+      # Set the headers to specify JSON content
+      headers = {
+          "Content-Type": "application/json"
+      }
+
+      # Send the POST request
+      response = requests.post(url, data=payload_json, headers=headers)
+
+      # Check the response
+      if response.status_code == 200:
+          output = json.loads(response.text)
+          context = output['context']
+          print(output['response']+ "\n")
+          
+
+      else:
+          print(f"Request failed with status code {response.status_code}")
+

examples/python-rag-newssummary/utils.py

@@ -0,0 +1,108 @@
+import curses
+import feedparser
+import requests
+import unicodedata
+import json
+from newspaper import Article
+from bs4 import BeautifulSoup
+from nltk.tokenize import sent_tokenize, word_tokenize
+import numpy as np
+from sklearn.neighbors import NearestNeighbors
+from mattsollamatools import chunker
+
+# Create a dictionary to store topics and their URLs
+topic_urls = {
+    "Mac": "https://9to5mac.com/guides/mac/feed",
+    "News": "http://www.npr.org/rss/rss.php?id=1001",
+    "Nvidia": "https://nvidianews.nvidia.com/releases.xml",
+    "Raspberry Pi": "https://www.raspberrypi.com/news/feed/", 
+    "Music": "https://www.billboard.com/c/music/music-news/feed/"
+}
+
+# Use curses to create a menu of topics
+def menu(stdscr):
+    chosen_topic = get_url_for_topic(stdscr)  
+    url = topic_urls[chosen_topic] if chosen_topic in topic_urls else "Topic not found"
+    
+    stdscr.addstr(len(topic_urls) + 3, 0, f"Selected URL for {chosen_topic}: {url}")
+    stdscr.refresh()
+    
+    return chosen_topic
+
+# You have chosen a topic. Now return the url for that topic
+def get_url_for_topic(stdscr):
+    curses.curs_set(0)  # Hide the cursor
+    stdscr.clear()
+
+    stdscr.addstr(0, 0, "Choose a topic using the arrow keys (Press Enter to select):")
+
+    # Create a list of topics
+    topics = list(topic_urls.keys())
+    current_topic = 0
+
+    while True:
+        for i, topic in enumerate(topics):
+            if i == current_topic:
+                stdscr.addstr(i + 2, 2, f"> {topic}")
+            else:
+                stdscr.addstr(i + 2, 2, f"  {topic}")
+
+        stdscr.refresh()
+
+        key = stdscr.getch()
+
+        if key == curses.KEY_DOWN and current_topic < len(topics) - 1:
+            current_topic += 1
+        elif key == curses.KEY_UP and current_topic > 0:
+            current_topic -= 1
+        elif key == 10:  # Enter key
+            return topic_urls[topics[current_topic]]
+
+# Get the last N URLs from an RSS feed
+def getUrls(feed_url, n=20):
+    feed = feedparser.parse(feed_url)
+    entries = feed.entries[-n:]
+    urls = [entry.link for entry in entries]
+    return urls
+
+# Often there are a bunch of ads and menus on pages for a news article. This uses newspaper3k to get just the text of just the article.
+def getArticleText(url):
+  article = Article(url)
+  article.download()
+  article.parse()
+  return article.text
+
+def get_summary(text):
+  systemPrompt = "Write a concise summary of the text, return your responses with 5 lines that cover the key points of the text given."
+  prompt = text
+  
+  url = "http://localhost:11434/api/generate"
+
+  payload = {
+    "model": "mistral-openorca",
+    "prompt": prompt, 
+    "system": systemPrompt,
+    "stream": False
+  }
+  payload_json = json.dumps(payload)
+  headers = {"Content-Type": "application/json"}
+  response = requests.post(url, data=payload_json, headers=headers)
+
+  return json.loads(response.text)["response"]
+
+# Perform K-nearest neighbors (KNN) search
+def knn_search(question_embedding, embeddings, k=5):
+    X = np.array([item['embedding'] for article in embeddings for item in article['embeddings']])
+    source_texts = [item['source'] for article in embeddings for item in article['embeddings']]
+    
+    # Fit a KNN model on the embeddings
+    knn = NearestNeighbors(n_neighbors=k, metric='cosine')
+    knn.fit(X)
+    
+    # Find the indices and distances of the k-nearest neighbors
+    distances, indices = knn.kneighbors(question_embedding, n_neighbors=k)
+    
+    # Get the indices and source texts of the best matches
+    best_matches = [(indices[0][i], source_texts[indices[0][i]]) for i in range(k)]
+    
+    return best_matches

examples/python-simplegenerate/client.py

@@ -17,7 +17,7 @@ def generate(prompt, context):
     for line in r.iter_lines():
         body = json.loads(line)
         response_part = body.get('response', '')
-        # the response streams one token at a time, print that as we recieve it
+        # the response streams one token at a time, print that as we receive it
         print(response_part, end='', flush=True)
 
         if 'error' in body:
@@ -35,4 +35,4 @@ def main():
         print()
 
 if __name__ == "__main__":
-    main()
+    main()

examples/typescript-mentors/.gitignore

@@ -0,0 +1,2 @@
+node_modules
+package-lock.json

examples/typescript-mentors/README.md

@@ -0,0 +1,21 @@
+# Ask the Mentors
+
+This example demonstrates how one would create a set of 'mentors' you can have a conversation with. The mentors are generated using the `character-generator.ts` file. This will use **Stable Beluga 70b** to create a bio and list of verbal ticks and common phrases used by each person. Then `mentors.ts` will take a question, and choose three of the 'mentors' and start a conversation with them. Occasionally, they will talk to each other, and other times they will just deliver a set of monologues. It's fun to see what they do and say.
+
+## Usage
+
+```bash
+ts-node ./character-generator.ts "Lorne Greene"
+```
+
+This will create `lornegreene/Modelfile`. Now you can create a model with this command:
+
+```bash
+ollama create lornegreene -f lornegreene/Modelfile
+```
+
+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.
+
+```bash
+ts-node ./mentors.ts "What is a Jackalope?"
+```

examples/typescript-mentors/character-generator.ts

@@ -0,0 +1,26 @@
+import { Ollama } from 'ollama-node'
+import fs from 'fs';
+import path from 'path';
+
+async function characterGenerator() {
+  const character = process.argv[2];
+  console.log(`You are creating a character for ${character}.`);
+  const foldername = character.replace(/\s/g, '').toLowerCase();
+  const directory = path.join(__dirname, foldername);
+  if (!fs.existsSync(directory)) {
+    fs.mkdirSync(directory, { recursive: true });
+  }
+
+  const ollama = new Ollama();
+  ollama.setModel("stablebeluga2:70b-q4_K_M");
+  const bio = await ollama.generate(`create a bio of ${character} in a single long paragraph. Instead of saying '${character} is...' or '${character} was...' use language like 'You are...' or 'You were...'. Then create a paragraph describing the speaking mannerisms and style of ${character}. Don't include anything about how ${character} looked or what they sounded like, just focus on the words they said. Instead of saying '${character} would say...' use language like 'You should say...'. If you use quotes, always use single quotes instead of double quotes. If there are any specific words or phrases you used a lot, show how you used them. `);
+
+  const thecontents = `FROM llama2\nSYSTEM """\n${bio.response.replace(/(\r\n|\n|\r)/gm, " ").replace('would', 'should')} All answers to questions should be related back to what you are most known for.\n"""`;
+
+  fs.writeFile(path.join(directory, 'Modelfile'), thecontents, (err: any) => {
+    if (err) throw err;
+    console.log('The file has been saved!');
+  });
+}
+
+characterGenerator();

examples/typescript-mentors/mentors.ts

@@ -0,0 +1,59 @@
+import { Ollama } from 'ollama-node';
+
+const mentorCount = 3;
+const ollama = new Ollama();
+
+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);
+    mentors.splice(mentors.indexOf(mentor), 1);
+  }
+  return chosenMentors;
+}
+
+function getMentorFileName(mentor: string): string {
+  const model = mentor.toLowerCase().replace(/\s/g, '');
+  return `mattw/${model}`;
+}
+
+async function getSystemPrompt(mentor: string, isLast: boolean, question: string): Promise<string> {
+  ollama.setModel(getMentorFileName(mentor));
+  const info = await ollama.showModelInfo()
+  let SystemPrompt = info.system || '';
+  SystemPrompt += ` You should continue the conversation as if you were ${mentor} and acknowledge the people before you in the conversation. You should adopt their mannerisms and tone, but also not use language they wouldn't use. If they are not known to know about the concept in the question, don't offer an answer. Your answer should be no longer than 1 paragraph. And definitely try not to sound like anyone else. Don't repeat any slang or phrases already used. And if it is a question the original ${mentor} wouldn't have know the answer to, just say that you don't know, in the style of ${mentor}. And think about the time the person lived. Don't use terminology that they wouldn't have used.`
+
+  if (isLast) {
+    SystemPrompt += ` End your answer with something like I hope our answers help you out`;
+  } else {
+    SystemPrompt += ` Remember, this is a conversation, so you don't need a conclusion, but end your answer with a question related to the first question: "${question}".`;
+  }
+  return SystemPrompt;
+}
+
+async function main() {
+  const mentors = getMentors();
+  const question = process.argv[2];
+  let theConversation = `Here is the conversation so far.\nYou: ${question}\n`
+
+  for await (const mentor of mentors) {
+    const SystemPrompt = await getSystemPrompt(mentor, mentor === mentors[mentorCount - 1], question);
+    ollama.setModel(getMentorFileName(mentor));
+    ollama.setSystemPrompt(SystemPrompt);
+    let output = '';
+    process.stdout.write(`\n${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);
+      } else {
+        process.stdout.write('\n');
+      }
+    }
+    theConversation += `${mentor}: ${output}\n\n`
+  }
+}
+
+main();

examples/typescript-mentors/package.json

@@ -0,0 +1,7 @@
+{
+  "dependencies": {
+    "fs": "^0.0.1-security",
+    "ollama-node": "^0.0.3",
+    "path": "^0.12.7"
+  }
+}

format/bytes.go

@@ -0,0 +1,23 @@
+package format
+
+import "fmt"
+
+const (
+	Byte     = 1
+	KiloByte = Byte * 1000
+	MegaByte = KiloByte * 1000
+	GigaByte = MegaByte * 1000
+)
+
+func HumanBytes(b int64) string {
+	switch {
+	case b > GigaByte:
+		return fmt.Sprintf("%d GB", b/GigaByte)
+	case b > MegaByte:
+		return fmt.Sprintf("%d MB", b/MegaByte)
+	case b > KiloByte:
+		return fmt.Sprintf("%d KB", b/KiloByte)
+	default:
+		return fmt.Sprintf("%d B", b)
+	}
+}

format/format.go

@@ -0,0 +1,25 @@
+package format
+
+import (
+	"fmt"
+	"math"
+)
+
+const (
+	Thousand = 1000
+	Million  = Thousand * 1000
+	Billion  = Million * 1000
+)
+
+func HumanNumber(b uint64) string {
+	switch {
+	case b > Billion:
+		return fmt.Sprintf("%.0fB", math.Round(float64(b)/Billion))
+	case b > Million:
+		return fmt.Sprintf("%.0fM", math.Round(float64(b)/Million))
+	case b > Thousand:
+		return fmt.Sprintf("%.0fK", math.Round(float64(b)/Thousand))
+	default:
+		return fmt.Sprintf("%d", b)
+	}
+}

format/time.go

@@ -7,26 +7,14 @@ import (
 	"time"
 )
 
-// HumanDuration returns a human-readable approximation of a duration
-// (eg. "About a minute", "4 hours ago", etc.).
-// Modified version of github.com/docker/go-units.HumanDuration
-func HumanDuration(d time.Duration) string {
-	return HumanDurationWithCase(d, true)
-}
-
-// HumanDurationWithCase returns a human-readable approximation of a
-// duration (eg. "About a minute", "4 hours ago", etc.). but allows
-// you to specify whether the first word should be capitalized
-// (eg. "About" vs. "about")
-func HumanDurationWithCase(d time.Duration, useCaps bool) string {
+// humanDuration returns a human-readable approximation of a
+// duration (eg. "About a minute", "4 hours ago", etc.).
+func humanDuration(d time.Duration) string {
 	seconds := int(d.Seconds())
 
 	switch {
 	case seconds < 1:
-		if useCaps {
-			return "Less than a second"
-		}
-		return "less than a second"
+		return "Less than a second"
 	case seconds == 1:
 		return "1 second"
 	case seconds < 60:
@@ -36,10 +24,7 @@ func HumanDurationWithCase(d time.Duration, useCaps bool) string {
 	minutes := int(d.Minutes())
 	switch {
 	case minutes == 1:
-		if useCaps {
-			return "About a minute"
-		}
-		return "about a minute"
+		return "About a minute"
 	case minutes < 60:
 		return fmt.Sprintf("%d minutes", minutes)
 	}
@@ -47,10 +32,7 @@ func HumanDurationWithCase(d time.Duration, useCaps bool) string {
 	hours := int(math.Round(d.Hours()))
 	switch {
 	case hours == 1:
-		if useCaps {
-			return "About an hour"
-		}
-		return "about an hour"
+		return "About an hour"
 	case hours < 48:
 		return fmt.Sprintf("%d hours", hours)
 	case hours < 24*7*2:
@@ -65,77 +47,22 @@ func HumanDurationWithCase(d time.Duration, useCaps bool) string {
 }
 
 func HumanTime(t time.Time, zeroValue string) string {
-	return humanTimeWithCase(t, zeroValue, true)
+	return humanTime(t, zeroValue)
 }
 
 func HumanTimeLower(t time.Time, zeroValue string) string {
-	return humanTimeWithCase(t, zeroValue, false)
+	return strings.ToLower(humanTime(t, zeroValue))
 }
 
-func humanTimeWithCase(t time.Time, zeroValue string, useCaps bool) string {
+func humanTime(t time.Time, zeroValue string) string {
 	if t.IsZero() {
 		return zeroValue
 	}
 
 	delta := time.Since(t)
 	if delta < 0 {
-		return HumanDurationWithCase(-delta, useCaps) + " from now"
+		return humanDuration(-delta) + " from now"
 	}
-	return HumanDurationWithCase(delta, useCaps) + " ago"
-}
-
-// ExcatDuration returns a human readable hours/minutes/seconds or milliseconds format of a duration
-// the most precise level of duration is milliseconds
-func ExactDuration(d time.Duration) string {
-	if d.Seconds() < 1 {
-		if d.Milliseconds() == 1 {
-			return fmt.Sprintf("%d millisecond", d.Milliseconds())
-		}
-		return fmt.Sprintf("%d milliseconds", d.Milliseconds())
-	}
-
-	var readableDur strings.Builder
-
-	dur := d.String()
-
-	// split the default duration string format of 0h0m0s into something nicer to read
-	h := strings.Split(dur, "h")
-	if len(h) > 1 {
-		hours := h[0]
-		if hours == "1" {
-			readableDur.WriteString(fmt.Sprintf("%s hour ", hours))
-		} else {
-			readableDur.WriteString(fmt.Sprintf("%s hours ", hours))
-		}
-		dur = h[1]
-	}
-
-	m := strings.Split(dur, "m")
-	if len(m) > 1 {
-		mins := m[0]
-		switch mins {
-		case "0":
-			// skip
-		case "1":
-			readableDur.WriteString(fmt.Sprintf("%s minute ", mins))
-		default:
-			readableDur.WriteString(fmt.Sprintf("%s minutes ", mins))
-		}
-		dur = m[1]
-	}
-
-	s := strings.Split(dur, "s")
-	if len(s) > 0 {
-		sec := s[0]
-		switch sec {
-		case "0":
-			// skip
-		case "1":
-			readableDur.WriteString(fmt.Sprintf("%s second ", sec))
-		default:
-			readableDur.WriteString(fmt.Sprintf("%s seconds ", sec))
-		}
-	}
-
-	return strings.TrimSpace(readableDur.String())
+
+	return humanDuration(delta) + " ago"
 }

format/time_test.go

@@ -11,92 +11,25 @@ func assertEqual(t *testing.T, a interface{}, b interface{}) {
 	}
 }
 
-func TestHumanDuration(t *testing.T) {
-	day := 24 * time.Hour
-	week := 7 * day
-	month := 30 * day
-	year := 365 * day
-
-	assertEqual(t, "Less than a second", HumanDuration(450*time.Millisecond))
-	assertEqual(t, "Less than a second", HumanDurationWithCase(450*time.Millisecond, true))
-	assertEqual(t, "less than a second", HumanDurationWithCase(450*time.Millisecond, false))
-	assertEqual(t, "1 second", HumanDuration(1*time.Second))
-	assertEqual(t, "45 seconds", HumanDuration(45*time.Second))
-	assertEqual(t, "46 seconds", HumanDuration(46*time.Second))
-	assertEqual(t, "59 seconds", HumanDuration(59*time.Second))
-	assertEqual(t, "About a minute", HumanDuration(60*time.Second))
-	assertEqual(t, "About a minute", HumanDurationWithCase(1*time.Minute, true))
-	assertEqual(t, "about a minute", HumanDurationWithCase(1*time.Minute, false))
-	assertEqual(t, "3 minutes", HumanDuration(3*time.Minute))
-	assertEqual(t, "35 minutes", HumanDuration(35*time.Minute))
-	assertEqual(t, "35 minutes", HumanDuration(35*time.Minute+40*time.Second))
-	assertEqual(t, "45 minutes", HumanDuration(45*time.Minute))
-	assertEqual(t, "45 minutes", HumanDuration(45*time.Minute+40*time.Second))
-	assertEqual(t, "46 minutes", HumanDuration(46*time.Minute))
-	assertEqual(t, "59 minutes", HumanDuration(59*time.Minute))
-	assertEqual(t, "About an hour", HumanDuration(1*time.Hour))
-	assertEqual(t, "About an hour", HumanDurationWithCase(1*time.Hour+29*time.Minute, true))
-	assertEqual(t, "about an hour", HumanDurationWithCase(1*time.Hour+29*time.Minute, false))
-	assertEqual(t, "2 hours", HumanDuration(1*time.Hour+31*time.Minute))
-	assertEqual(t, "2 hours", HumanDuration(1*time.Hour+59*time.Minute))
-	assertEqual(t, "3 hours", HumanDuration(3*time.Hour))
-	assertEqual(t, "3 hours", HumanDuration(3*time.Hour+29*time.Minute))
-	assertEqual(t, "4 hours", HumanDuration(3*time.Hour+31*time.Minute))
-	assertEqual(t, "4 hours", HumanDuration(3*time.Hour+59*time.Minute))
-	assertEqual(t, "4 hours", HumanDuration(3*time.Hour+60*time.Minute))
-	assertEqual(t, "24 hours", HumanDuration(24*time.Hour))
-	assertEqual(t, "36 hours", HumanDuration(1*day+12*time.Hour))
-	assertEqual(t, "2 days", HumanDuration(2*day))
-	assertEqual(t, "7 days", HumanDuration(7*day))
-	assertEqual(t, "13 days", HumanDuration(13*day+5*time.Hour))
-	assertEqual(t, "2 weeks", HumanDuration(2*week))
-	assertEqual(t, "2 weeks", HumanDuration(2*week+4*day))
-	assertEqual(t, "3 weeks", HumanDuration(3*week))
-	assertEqual(t, "4 weeks", HumanDuration(4*week))
-	assertEqual(t, "4 weeks", HumanDuration(4*week+3*day))
-	assertEqual(t, "4 weeks", HumanDuration(1*month))
-	assertEqual(t, "6 weeks", HumanDuration(1*month+2*week))
-	assertEqual(t, "2 months", HumanDuration(2*month))
-	assertEqual(t, "2 months", HumanDuration(2*month+2*week))
-	assertEqual(t, "3 months", HumanDuration(3*month))
-	assertEqual(t, "3 months", HumanDuration(3*month+1*week))
-	assertEqual(t, "5 months", HumanDuration(5*month+2*week))
-	assertEqual(t, "13 months", HumanDuration(13*month))
-	assertEqual(t, "23 months", HumanDuration(23*month))
-	assertEqual(t, "24 months", HumanDuration(24*month))
-	assertEqual(t, "2 years", HumanDuration(24*month+2*week))
-	assertEqual(t, "3 years", HumanDuration(3*year+2*month))
-}
-
 func TestHumanTime(t *testing.T) {
 	now := time.Now()
 
 	t.Run("zero value", func(t *testing.T) {
 		assertEqual(t, HumanTime(time.Time{}, "never"), "never")
 	})
+
 	t.Run("time in the future", func(t *testing.T) {
 		v := now.Add(48 * time.Hour)
 		assertEqual(t, HumanTime(v, ""), "2 days from now")
 	})
+
 	t.Run("time in the past", func(t *testing.T) {
 		v := now.Add(-48 * time.Hour)
 		assertEqual(t, HumanTime(v, ""), "2 days ago")
 	})
-}
 
-func TestExactDuration(t *testing.T) {
-	assertEqual(t, "1 millisecond", ExactDuration(1*time.Millisecond))
-	assertEqual(t, "10 milliseconds", ExactDuration(10*time.Millisecond))
-	assertEqual(t, "1 second", ExactDuration(1*time.Second))
-	assertEqual(t, "10 seconds", ExactDuration(10*time.Second))
-	assertEqual(t, "1 minute", ExactDuration(1*time.Minute))
-	assertEqual(t, "10 minutes", ExactDuration(10*time.Minute))
-	assertEqual(t, "1 hour", ExactDuration(1*time.Hour))
-	assertEqual(t, "10 hours", ExactDuration(10*time.Hour))
-	assertEqual(t, "1 hour 1 second", ExactDuration(1*time.Hour+1*time.Second))
-	assertEqual(t, "1 hour 10 seconds", ExactDuration(1*time.Hour+10*time.Second))
-	assertEqual(t, "1 hour 1 minute", ExactDuration(1*time.Hour+1*time.Minute))
-	assertEqual(t, "1 hour 10 minutes", ExactDuration(1*time.Hour+10*time.Minute))
-	assertEqual(t, "1 hour 1 minute 1 second", ExactDuration(1*time.Hour+1*time.Minute+1*time.Second))
-	assertEqual(t, "10 hours 10 minutes 10 seconds", ExactDuration(10*time.Hour+10*time.Minute+10*time.Second))
+	t.Run("soon", func(t *testing.T) {
+		v := now.Add(800 * time.Millisecond)
+		assertEqual(t, HumanTime(v, ""), "Less than a second from now")
+	})
 }

go.mod

@@ -4,12 +4,13 @@ go 1.20
 
 require (
 	github.com/dustin/go-humanize v1.0.1
+	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/pdevine/readline v1.5.2
 	github.com/spf13/cobra v1.7.0
+	golang.org/x/sync v0.3.0
 )
 
 require github.com/rivo/uniseg v0.2.0 // indirect
@@ -38,13 +39,12 @@ require (
 	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
 	github.com/ugorji/go/codec v1.2.11 // indirect
 	golang.org/x/arch v0.3.0 // indirect
-	golang.org/x/crypto v0.10.0
+	golang.org/x/crypto v0.14.0
 	golang.org/x/exp v0.0.0-20230817173708-d852ddb80c63
-	golang.org/x/net v0.10.0 // indirect
-	golang.org/x/sys v0.11.0 // indirect
-	golang.org/x/term v0.10.0
-	golang.org/x/text v0.10.0 // indirect
-	gonum.org/v1/gonum v0.13.0
+	golang.org/x/net v0.17.0 // indirect
+	golang.org/x/sys v0.13.0 // indirect
+	golang.org/x/term v0.13.0
+	golang.org/x/text v0.13.0 // indirect
 	google.golang.org/protobuf v1.30.0 // indirect
 	gopkg.in/yaml.v3 v3.0.1 // indirect
 )

go.sum

@@ -4,10 +4,6 @@ github.com/bytedance/sonic v1.9.1/go.mod h1:i736AoUSYt75HyZLoJW9ERYxcy6eaN6h4BZX
 github.com/chenzhuoyu/base64x v0.0.0-20211019084208-fb5309c8db06/go.mod h1:DH46F32mSOjUmXrMHnKwZdA8wcEefY7UVqBKYGjpdQY=
 github.com/chenzhuoyu/base64x v0.0.0-20221115062448-fe3a3abad311 h1:qSGYFH7+jGhDF8vLC+iwCD4WpbV1EBDSzWkJODFLams=
 github.com/chenzhuoyu/base64x v0.0.0-20221115062448-fe3a3abad311/go.mod h1:b583jCggY9gE99b6G5LEC39OIiVsWj+R97kbl5odCEk=
-github.com/chzyer/logex v1.2.1 h1:XHDu3E6q+gdHgsdTPH6ImJMIp436vR6MPtH8gP05QzM=
-github.com/chzyer/logex v1.2.1/go.mod h1:JLbx6lG2kDbNRFnfkgvh4eRJRPX1QCoOIWomwysCBrQ=
-github.com/chzyer/test v1.0.0 h1:p3BQDXSxOhOG0P9z6/hGnII4LGiEPOYBhs8asl/fC04=
-github.com/chzyer/test v1.0.0/go.mod h1:2JlltgoNkt4TW/z9V/IzDdFaMTM2JPIi26O1pF38GC8=
 github.com/cpuguy83/go-md2man/v2 v2.0.2/go.mod h1:tgQtvFlXSQOSOSIRvRPT7W67SCa46tRHOmNcaadrF8o=
 github.com/creack/pty v1.1.9/go.mod h1:oKZEueFk5CKHvIhNR5MUki03XCEU+Q6VDXinZuGJ33E=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -15,6 +11,8 @@ github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/dustin/go-humanize v1.0.1 h1:GzkhY7T5VNhEkwH0PVJgjz+fX1rhBrR7pRT3mDkpeCY=
 github.com/dustin/go-humanize v1.0.1/go.mod h1:Mu1zIs6XwVuF/gI1OepvI0qD18qycQx+mFykh5fBlto=
+github.com/emirpasic/gods v1.18.1 h1:FXtiHYKDGKCW2KzwZKx0iC0PQmdlorYgdFG9jPXJ1Bc=
+github.com/emirpasic/gods v1.18.1/go.mod h1:8tpGGwCnJ5H4r6BWwaV6OrWmMoPhUl5jm/FMNAnJvWQ=
 github.com/gabriel-vasile/mimetype v1.4.2 h1:w5qFW6JKBz9Y393Y4q372O9A7cUSequkh1Q7OhCmWKU=
 github.com/gabriel-vasile/mimetype v1.4.2/go.mod h1:zApsH/mKG4w07erKIaJPFiX0Tsq9BFQgN3qGY5GnNgA=
 github.com/gin-contrib/cors v1.4.0 h1:oJ6gwtUl3lqV0WEIwM/LxPF1QZ5qe2lGWdY2+bz7y0g=
@@ -78,8 +76,6 @@ github.com/olekukonko/tablewriter v0.0.5 h1:P2Ga83D34wi1o9J6Wh1mRuqd4mF/x/lgBS7N
 github.com/olekukonko/tablewriter v0.0.5/go.mod h1:hPp6KlRPjbx+hW8ykQs1w3UBbZlj6HuIJcUGPhkA7kY=
 github.com/pbnjay/memory v0.0.0-20210728143218-7b4eea64cf58 h1:onHthvaw9LFnH4t2DcNVpwGmV9E1BkGknEliJkfwQj0=
 github.com/pbnjay/memory v0.0.0-20210728143218-7b4eea64cf58/go.mod h1:DXv8WO4yhMYhSNPKjeNKa5WY9YCIEBRbNzFFPJbWO6Y=
-github.com/pdevine/readline v1.5.2 h1:oz6Y5GdTmhPG+08hhxcAvtHitSANWuA2100Sppb38xI=
-github.com/pdevine/readline v1.5.2/go.mod h1:na/LbuE5PYwxI7GyopWdIs3U8HVe89lYlNTFTXH3wOw=
 github.com/pelletier/go-toml/v2 v2.0.1/go.mod h1:r9LEWfGN8R5k0VXJ+0BkIe7MYkRdwZOjgMj2KwnJFUo=
 github.com/pelletier/go-toml/v2 v2.0.8 h1:0ctb6s9mE31h0/lhu+J6OPmVeDxJn+kYnJc2jZR9tGQ=
 github.com/pelletier/go-toml/v2 v2.0.8/go.mod h1:vuYfssBdrU2XDZ9bYydBu6t+6a6PYNcZljzZR9VXg+4=
@@ -118,33 +114,32 @@ golang.org/x/arch v0.0.0-20210923205945-b76863e36670/go.mod h1:5om86z9Hs0C8fWVUu
 golang.org/x/arch v0.3.0 h1:02VY4/ZcO/gBOH6PUaoiptASxtXU10jazRCP865E97k=
 golang.org/x/arch v0.3.0/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
 golang.org/x/crypto v0.0.0-20210711020723-a769d52b0f97/go.mod h1:GvvjBRRGRdwPK5ydBHafDWAxML/pGHZbMvKqRZ5+Abc=
-golang.org/x/crypto v0.10.0 h1:LKqV2xt9+kDzSTfOhx4FrkEBcMrAgHSYgzywV9zcGmM=
-golang.org/x/crypto v0.10.0/go.mod h1:o4eNf7Ede1fv+hwOwZsTHl9EsPFO6q6ZvYR8vYfY45I=
+golang.org/x/crypto v0.14.0 h1:wBqGXzWJW6m1XrIKlAH0Hs1JJ7+9KBwnIO8v66Q9cHc=
+golang.org/x/crypto v0.14.0/go.mod h1:MVFd36DqK4CsrnJYDkBA3VC4m2GkXAM0PvzMCn4JQf4=
 golang.org/x/exp v0.0.0-20230817173708-d852ddb80c63 h1:m64FZMko/V45gv0bNmrNYoDEq8U5YUhetc9cBWKS1TQ=
 golang.org/x/exp v0.0.0-20230817173708-d852ddb80c63/go.mod h1:0v4NqG35kSWCMzLaMeX+IQrlSnVE/bqGSyC2cz/9Le8=
 golang.org/x/net v0.0.0-20210226172049-e18ecbb05110/go.mod h1:m0MpNAwzfU5UDzcl9v0D8zg8gWTRqZa9RBIspLL5mdg=
-golang.org/x/net v0.10.0 h1:X2//UzNDwYmtCLn7To6G58Wr6f5ahEAQgKNzv9Y951M=
-golang.org/x/net v0.10.0/go.mod h1:0qNGK6F8kojg2nk9dLZ2mShWaEBan6FAoqfSigmmuDg=
+golang.org/x/net v0.17.0 h1:pVaXccu2ozPjCXewfr1S7xza/zcXTity9cCdXQYSjIM=
+golang.org/x/net v0.17.0/go.mod h1:NxSsAGuq816PNPmqtQdLE42eU2Fs7NoRIZrHJAlaCOE=
+golang.org/x/sync v0.3.0 h1:ftCYgMx6zT/asHUrPw8BLLscYtGznsLAnjq5RH9P66E=
+golang.org/x/sync v0.3.0/go.mod h1:FU7BRWz2tNW+3quACPkgCx/L+uEAv1htQ0V83Z9Rj+Y=
 golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20210615035016-665e8c7367d1/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20210630005230-0f9fa26af87c/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20210806184541-e5e7981a1069/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.0.0-20220310020820-b874c991c1a5/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220704084225-05e143d24a9e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.11.0 h1:eG7RXZHdqOJ1i+0lgLgCpSXAp6M3LYlAo6osgSi0xOM=
-golang.org/x/sys v0.11.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.13.0 h1:Af8nKPmuFypiUBjVoU9V20FiaFXOcuZI21p0ycVYYGE=
+golang.org/x/sys v0.13.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
-golang.org/x/term v0.10.0 h1:3R7pNqamzBraeqj/Tj8qt1aQ2HpmlC+Cx/qL/7hn4/c=
-golang.org/x/term v0.10.0/go.mod h1:lpqdcUyK/oCiQxvxVrppt5ggO2KCZ5QblwqPnfZ6d5o=
+golang.org/x/term v0.13.0 h1:bb+I9cTfFazGW51MZqBVmZy7+JEJMouUHTUSKVQLBek=
+golang.org/x/term v0.13.0/go.mod h1:LTmsnFJwVN6bCy1rVCoS+qHT1HhALEFxKncY3WNNh4U=
 golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
 golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
-golang.org/x/text v0.10.0 h1:UpjohKhiEgNc0CSauXmwYftY1+LlaC75SJwh0SgCX58=
-golang.org/x/text v0.10.0/go.mod h1:TvPlkZtksWOMsz7fbANvkp4WM8x/WCo/om8BMLbz+aE=
+golang.org/x/text v0.13.0 h1:ablQoSUd0tRdKxZewP80B+BaqeKJuVhuRxj/dkrun3k=
+golang.org/x/text v0.13.0/go.mod h1:TvPlkZtksWOMsz7fbANvkp4WM8x/WCo/om8BMLbz+aE=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
-gonum.org/v1/gonum v0.13.0 h1:a0T3bh+7fhRyqeNbiC3qVHYmkiQgit3wnNan/2c0HMM=
-gonum.org/v1/gonum v0.13.0/go.mod h1:/WPYRckkfWrhWefxyYTfrTtQR0KH4iyHNuzxqXAKyAU=
 google.golang.org/protobuf v1.26.0-rc.1/go.mod h1:jlhhOSvTdKEhbULTjvd4ARK9grFBp09yW+WbY/TyQbw=
 google.golang.org/protobuf v1.28.0/go.mod h1:HV8QOd/L58Z+nl8r43ehVNZIU/HEI6OcFqwMG9pJV4I=
 google.golang.org/protobuf v1.30.0 h1:kPPoIgf3TsEvrm0PFe15JQ+570QVxYzEvvHqChK+cng=

llm/falcon.go

@@ -1,7 +1,5 @@
 package llm
 
-const ModelFamilyFalcon = "falcon"
-
 const (
 	falconModelType7B   = 32
 	falconModelType40B  = 60
@@ -17,6 +15,6 @@ func falconModelType(numLayer uint32) string {
 	case 80:
 		return "180B"
 	default:
-		return "Unknown"
+		return "unknown"
 	}
 }

llm/ggml.go

@@ -69,7 +69,7 @@ func fileType(fileType uint32) string {
 	case fileTypeQ6_K:
 		return "Q6_K"
 	default:
-		return "Unknown"
+		return "unknown"
 	}
 }
 
@@ -175,7 +175,8 @@ const (
 	// Magic constant for `ggla` files (LoRA adapter).
 	FILE_MAGIC_GGLA = 0x67676C61
 	// Magic constant for `gguf` files (versioned, gguf)
-	FILE_MAGIC_GGUF = 0x46554747
+	FILE_MAGIC_GGUF_LE = 0x46554747
+	FILE_MAGIC_GGUF_BE = 0x47475546
 )
 
 func DecodeGGML(r io.ReadSeeker) (*GGML, error) {
@@ -191,8 +192,10 @@ func DecodeGGML(r io.ReadSeeker) (*GGML, error) {
 		ggml.container = &containerGGJT{}
 	case FILE_MAGIC_GGLA:
 		ggml.container = &containerLORA{}
-	case FILE_MAGIC_GGUF:
-		ggml.container = &containerGGUF{}
+	case FILE_MAGIC_GGUF_LE:
+		ggml.container = &containerGGUF{bo: binary.LittleEndian}
+	case FILE_MAGIC_GGUF_BE:
+		ggml.container = &containerGGUF{bo: binary.BigEndian}
 	default:
 		return nil, errors.New("invalid file magic")
 	}

llm/gguf.go

@@ -3,12 +3,15 @@ package llm
 import (
 	"bytes"
 	"encoding/binary"
-	"errors"
 	"fmt"
 	"io"
+
+	"github.com/jmorganca/ollama/format"
 )
 
 type containerGGUF struct {
+	bo binary.ByteOrder
+
 	Version uint32
 
 	V1 struct {
@@ -20,6 +23,8 @@ type containerGGUF struct {
 		NumTensor uint64
 		NumKV     uint64
 	}
+
+	parameters uint64
 }
 
 func (c *containerGGUF) Name() string {
@@ -27,15 +32,13 @@ func (c *containerGGUF) Name() string {
 }
 
 func (c *containerGGUF) Decode(r io.Reader) (model, error) {
-	binary.Read(r, binary.LittleEndian, &c.Version)
+	binary.Read(r, c.bo, &c.Version)
 
 	switch c.Version {
 	case 1:
-		binary.Read(r, binary.LittleEndian, &c.V1)
-	case 2:
-		binary.Read(r, binary.LittleEndian, &c.V2)
+		binary.Read(r, c.bo, &c.V1)
 	default:
-		return nil, errors.New("invalid version")
+		binary.Read(r, c.bo, &c.V2)
 	}
 
 	model := newGGUFModel(c)
@@ -76,6 +79,14 @@ func newGGUFModel(container *containerGGUF) *ggufModel {
 	}
 }
 
+func (llm *ggufModel) NumTensor() uint64 {
+	if llm.Version == 1 {
+		return uint64(llm.V1.NumTensor)
+	}
+
+	return llm.V2.NumTensor
+}
+
 func (llm *ggufModel) NumKV() uint64 {
 	if llm.Version == 1 {
 		return uint64(llm.V1.NumKV)
@@ -94,6 +105,10 @@ func (llm *ggufModel) ModelFamily() string {
 }
 
 func (llm *ggufModel) ModelType() string {
+	if llm.parameters > 0 {
+		return format.HumanNumber(llm.parameters)
+	}
+
 	switch llm.ModelFamily() {
 	case "llama":
 		if blocks, ok := llm.kv["llama.block_count"].(uint32); ok {
@@ -109,9 +124,13 @@ func (llm *ggufModel) ModelType() string {
 		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"
+	return "unknown"
 }
 
 func (llm *ggufModel) FileType() string {
@@ -120,17 +139,13 @@ func (llm *ggufModel) FileType() string {
 		return fileType(t)
 	}
 
-	return "Unknown"
+	return "unknown"
 }
 
 func (llm *ggufModel) Decode(r io.Reader) error {
-	read := llm.readString
-	if llm.Version == 1 {
-		read = llm.readStringV1
-	}
-
+	// decode key-values
 	for i := 0; uint64(i) < llm.NumKV(); i++ {
-		k, err := read(r)
+		k, err := llm.readString(r)
 		if err != nil {
 			return err
 		}
@@ -162,24 +177,14 @@ func (llm *ggufModel) Decode(r io.Reader) error {
 		case ggufTypeBool:
 			v = llm.readBool(r)
 		case ggufTypeString:
-			fn := llm.readString
-			if llm.Version == 1 {
-				fn = llm.readStringV1
-			}
-
-			s, err := fn(r)
+			s, err := llm.readString(r)
 			if err != nil {
 				return err
 			}
 
 			v = s
 		case ggufTypeArray:
-			fn := llm.readArray
-			if llm.Version == 1 {
-				fn = llm.readArrayV1
-			}
-
-			a, err := fn(r)
+			a, err := llm.readArray(r)
 			if err != nil {
 				return err
 			}
@@ -192,6 +197,25 @@ func (llm *ggufModel) Decode(r io.Reader) error {
 		llm.kv[k] = v
 	}
 
+	// decode tensors
+	for i := 0; uint64(i) < llm.NumTensor(); i++ {
+		if _, err := llm.readString(r); err != nil {
+			return err
+		}
+
+		dimensions := llm.readU32(r)
+
+		var elements uint64 = 1
+		for i := 0; uint32(i) < dimensions; i++ {
+			elements *= llm.readU64(r)
+		}
+
+		llm.readU32(r) // type
+		llm.readU64(r) // offset
+
+		llm.parameters += elements
+	}
+
 	return nil
 }
 
@@ -205,75 +229,75 @@ func (llm *ggufModel) NumLayers() int64 {
 	return int64(v)
 }
 
-func (ggufModel) readU8(r io.Reader) uint8 {
+func (llm ggufModel) readU8(r io.Reader) uint8 {
 	var u8 uint8
-	binary.Read(r, binary.LittleEndian, &u8)
+	binary.Read(r, llm.bo, &u8)
 	return u8
 }
 
-func (ggufModel) readI8(r io.Reader) int8 {
+func (llm ggufModel) readI8(r io.Reader) int8 {
 	var i8 int8
-	binary.Read(r, binary.LittleEndian, &i8)
+	binary.Read(r, llm.bo, &i8)
 	return i8
 }
 
-func (ggufModel) readU16(r io.Reader) uint16 {
+func (llm ggufModel) readU16(r io.Reader) uint16 {
 	var u16 uint16
-	binary.Read(r, binary.LittleEndian, &u16)
+	binary.Read(r, llm.bo, &u16)
 	return u16
 }
 
-func (ggufModel) readI16(r io.Reader) int16 {
+func (llm ggufModel) readI16(r io.Reader) int16 {
 	var i16 int16
-	binary.Read(r, binary.LittleEndian, &i16)
+	binary.Read(r, llm.bo, &i16)
 	return i16
 }
 
-func (ggufModel) readU32(r io.Reader) uint32 {
+func (llm ggufModel) readU32(r io.Reader) uint32 {
 	var u32 uint32
-	binary.Read(r, binary.LittleEndian, &u32)
+	binary.Read(r, llm.bo, &u32)
 	return u32
 }
 
-func (ggufModel) readI32(r io.Reader) int32 {
+func (llm ggufModel) readI32(r io.Reader) int32 {
 	var i32 int32
-	binary.Read(r, binary.LittleEndian, &i32)
+	binary.Read(r, llm.bo, &i32)
 	return i32
 }
 
-func (ggufModel) readU64(r io.Reader) uint64 {
+func (llm ggufModel) readU64(r io.Reader) uint64 {
 	var u64 uint64
-	binary.Read(r, binary.LittleEndian, &u64)
+	binary.Read(r, llm.bo, &u64)
 	return u64
 }
 
-func (ggufModel) readI64(r io.Reader) int64 {
+func (llm ggufModel) readI64(r io.Reader) int64 {
 	var i64 int64
-	binary.Read(r, binary.LittleEndian, &i64)
+	binary.Read(r, llm.bo, &i64)
 	return i64
 }
 
-func (ggufModel) readF32(r io.Reader) float32 {
+func (llm ggufModel) readF32(r io.Reader) float32 {
 	var f32 float32
-	binary.Read(r, binary.LittleEndian, &f32)
+	binary.Read(r, llm.bo, &f32)
 	return f32
 }
 
-func (ggufModel) readF64(r io.Reader) float64 {
+func (llm ggufModel) readF64(r io.Reader) float64 {
 	var f64 float64
-	binary.Read(r, binary.LittleEndian, &f64)
+	binary.Read(r, llm.bo, &f64)
 	return f64
 }
 
-func (ggufModel) readBool(r io.Reader) bool {
+func (llm ggufModel) readBool(r io.Reader) bool {
 	var b bool
-	binary.Read(r, binary.LittleEndian, &b)
+	binary.Read(r, llm.bo, &b)
 	return b
 }
 
-func (ggufModel) readStringV1(r io.Reader) (string, error) {
+func (llm ggufModel) readStringV1(r io.Reader) (string, error) {
 	var nameLength uint32
-	binary.Read(r, binary.LittleEndian, &nameLength)
+	binary.Read(r, llm.bo, &nameLength)
 
 	var b bytes.Buffer
 	if _, err := io.CopyN(&b, r, int64(nameLength)); err != nil {
@@ -287,8 +311,12 @@ func (ggufModel) readStringV1(r io.Reader) (string, error) {
 }
 
 func (llm ggufModel) readString(r io.Reader) (string, error) {
+	if llm.Version == 1 {
+		return llm.readStringV1(r)
+	}
+
 	var nameLength uint64
-	binary.Read(r, binary.LittleEndian, &nameLength)
+	binary.Read(r, llm.bo, &nameLength)
 
 	var b bytes.Buffer
 	if _, err := io.CopyN(&b, r, int64(nameLength)); err != nil {
@@ -336,6 +364,10 @@ func (llm *ggufModel) readArrayV1(r io.Reader) (arr []any, err error) {
 }
 
 func (llm *ggufModel) readArray(r io.Reader) (arr []any, err error) {
+	if llm.Version == 1 {
+		return llm.readArrayV1(r)
+	}
+
 	atype := llm.readU32(r)
 	n := llm.readU64(r)
 

llm/llama.cpp/generate_darwin_amd64.go

@@ -9,8 +9,11 @@ package llm
 //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_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-remove-warm-up-logging.patch
+//go:generate git -C gguf apply ../patches/0001-update-default-log-target.patch
+//go:generate git -C gguf apply ../patches/0001-metal-handle-ggml_scale-for-n-4-0-close-3754.patch
 //go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DCMAKE_SYSTEM_PROCESSOR=x86_64 -DCMAKE_OSX_ARCHITECTURES=x86_64 -DCMAKE_OSX_DEPLOYMENT_TARGET=11.0
 //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

llm/llama.cpp/generate_darwin_arm64.go

@@ -9,8 +9,11 @@ package llm
 //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-remove-warm-up-logging.patch
+//go:generate git -C gguf apply ../patches/0001-update-default-log-target.patch
+//go:generate git -C gguf apply ../patches/0001-metal-handle-ggml_scale-for-n-4-0-close-3754.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

llm/llama.cpp/generate_linux.go

@@ -9,14 +9,18 @@ package llm
 //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-remove-warm-up-logging.patch
-//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_K_QUANTS=on
+//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 cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
+//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
 //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

llm/llama.cpp/generate_windows.go

@@ -7,8 +7,10 @@ package llm
 //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-remove-warm-up-logging.patch
-//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_K_QUANTS=on
+//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

llm/llama.cpp/patches/0001-metal-handle-ggml_scale-for-n-4-0-close-3754.patch

@@ -0,0 +1,91 @@
+From 469c9addef75893e6be12edda852d12e840bf064 Mon Sep 17 00:00:00 2001
+From: Georgi Gerganov <ggerganov@gmail.com>
+Date: Tue, 24 Oct 2023 09:46:50 +0300
+Subject: [PATCH 1/2] metal : handle ggml_scale for n%4 != 0 (close #3754)
+
+ggml-ci
+---
+ ggml-metal.m     | 18 +++++++++++++-----
+ ggml-metal.metal | 10 +++++++++-
+ 2 files changed, 22 insertions(+), 6 deletions(-)
+
+diff --git a/ggml-metal.m b/ggml-metal.m
+index c908106..c1901dc 100644
+--- a/ggml-metal.m
++++ b/ggml-metal.m
+@@ -62,6 +62,7 @@
+     GGML_METAL_DECL_KERNEL(mul);
+     GGML_METAL_DECL_KERNEL(mul_row); // TODO: avoid this extra kernel, instead extend the "mul" kernel to support broadcast
+     GGML_METAL_DECL_KERNEL(scale);
++    GGML_METAL_DECL_KERNEL(scale_4);
+     GGML_METAL_DECL_KERNEL(silu);
+     GGML_METAL_DECL_KERNEL(relu);
+     GGML_METAL_DECL_KERNEL(gelu);
+@@ -249,6 +250,7 @@ static void ggml_metal_log(enum ggml_log_level level, const char* format, ...){
+         GGML_METAL_ADD_KERNEL(mul);
+         GGML_METAL_ADD_KERNEL(mul_row);
+         GGML_METAL_ADD_KERNEL(scale);
++        GGML_METAL_ADD_KERNEL(scale_4);
+         GGML_METAL_ADD_KERNEL(silu);
+         GGML_METAL_ADD_KERNEL(relu);
+         GGML_METAL_ADD_KERNEL(gelu);
+@@ -347,6 +349,7 @@ void ggml_metal_free(struct ggml_metal_context * ctx) {
+     GGML_METAL_DEL_KERNEL(mul);
+     GGML_METAL_DEL_KERNEL(mul_row);
+     GGML_METAL_DEL_KERNEL(scale);
++    GGML_METAL_DEL_KERNEL(scale_4);
+     GGML_METAL_DEL_KERNEL(silu);
+     GGML_METAL_DEL_KERNEL(relu);
+     GGML_METAL_DEL_KERNEL(gelu);
+@@ -923,15 +926,20 @@ void ggml_metal_graph_compute(
+ 
+                             const float scale = *(const float *) src1->data;
+ 
+-                            [encoder setComputePipelineState:ctx->pipeline_scale];
++                            int64_t n = ggml_nelements(dst);
++
++                            if (n % 4 == 0) {
++                                n /= 4;
++                                [encoder setComputePipelineState:ctx->pipeline_scale_4];
++                            } else {
++                                [encoder setComputePipelineState:ctx->pipeline_scale];
++                            }
++
+                             [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
+                             [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
+                             [encoder setBytes:&scale length:sizeof(scale) atIndex:2];
+ 
+-                            const int64_t n = ggml_nelements(dst);
+-                            GGML_ASSERT(n % 4 == 0);
+-
+-                            [encoder dispatchThreadgroups:MTLSizeMake(n/4, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
++                            [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
+                         } break;
+                     case GGML_OP_UNARY:
+                         switch (ggml_get_unary_op(gf->nodes[i])) {
+diff --git a/ggml-metal.metal b/ggml-metal.metal
+index 69fc713..f4b4605 100644
+--- a/ggml-metal.metal
++++ b/ggml-metal.metal
+@@ -125,9 +125,17 @@ kernel void kernel_mul_row(
+ }
+ 
+ kernel void kernel_scale(
++        device const float * src0,
++        device       float * dst,
++        constant     float & scale,
++        uint tpig[[thread_position_in_grid]]) {
++    dst[tpig] = src0[tpig] * scale;
++}
++
++kernel void kernel_scale_4(
+         device const float4 * src0,
+         device       float4 * dst,
+-        constant     float & scale,
++        constant     float  & scale,
+         uint tpig[[thread_position_in_grid]]) {
+     dst[tpig] = src0[tpig] * scale;
+ }
+-- 
+2.39.3 (Apple Git-145)
+

llm/llama.cpp/patches/0001-remove-warm-up-logging.patch

@@ -1,25 +0,0 @@
-From 07993bdc35345b67b27aa649a7c099ad42d80c4c Mon Sep 17 00:00:00 2001
-From: Michael Yang <mxyng@pm.me>
-Date: Thu, 21 Sep 2023 14:43:21 -0700
-Subject: [PATCH] remove warm up logging
-
----
- common/common.cpp | 2 --
- 1 file changed, 2 deletions(-)
-
-diff --git a/common/common.cpp b/common/common.cpp
-index 2597ba0..b56549b 100644
---- a/common/common.cpp
-+++ b/common/common.cpp
-@@ -780,8 +780,6 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
-     }
- 
-     {
--        LOG("warming up the model with an empty run\n");
--
-         const std::vector<llama_token> tmp = { llama_token_bos(lctx), llama_token_eos(lctx), };
-         llama_eval(lctx, tmp.data(), std::min(tmp.size(), (size_t) params.n_batch), 0, params.n_threads);
-         llama_reset_timings(lctx);
--- 
-2.42.0
-

llm/llama.cpp/patches/0001-update-default-log-target.patch

@@ -0,0 +1,25 @@
+From 6465fec6290f0a7f5d4d0fbe6bcf634e4810dde6 Mon Sep 17 00:00:00 2001
+From: Michael Yang <mxyng@pm.me>
+Date: Mon, 23 Oct 2023 10:39:34 -0700
+Subject: [PATCH] default log stderr
+
+---
+ common/log.h | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/common/log.h b/common/log.h
+index b8953fd..25522cd 100644
+--- a/common/log.h
++++ b/common/log.h
+@@ -90,7 +90,7 @@
+ //  }
+ //
+ #ifndef LOG_TARGET
+-    #define LOG_TARGET log_handler()
++    #define LOG_TARGET nullptr
+ #endif
+ 
+ #ifndef LOG_TEE_TARGET
+-- 
+2.42.0
+

llm/llama.go

@@ -20,54 +20,85 @@ import (
 	"runtime"
 	"strconv"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/jmorganca/ollama/api"
+	"github.com/jmorganca/ollama/format"
 )
 
+const jsonGrammar = `
+root   ::= object
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\"" (
+    [^"\\] |
+    "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+# Optional space: by convention, applied in this grammar after literal chars when allowed
+ws ::= ([ \t\n] ws)?
+`
+
 //go:embed llama.cpp/*/build/*/bin/*
 var llamaCppEmbed embed.FS
 
 type ModelRunner struct {
-	Path string // path to the model runner executable
+	Path        string // path to the model runner executable
+	Accelerated bool
 }
 
 func chooseRunners(workDir, runnerType string) []ModelRunner {
 	buildPath := path.Join("llama.cpp", runnerType, "build")
-	var runners []string
+	var runners []ModelRunner
 
 	// set the runners based on the OS
 	// IMPORTANT: the order of the runners in the array is the priority order
 	switch runtime.GOOS {
 	case "darwin":
-		runners = []string{
-			path.Join(buildPath, "metal", "bin", "server"),
-			path.Join(buildPath, "cpu", "bin", "server"),
+		runners = []ModelRunner{
+			{Path: path.Join(buildPath, "metal", "bin", "ollama-runner")},
+			{Path: path.Join(buildPath, "cpu", "bin", "ollama-runner")},
 		}
 	case "linux":
-		runners = []string{
-			path.Join(buildPath, "cuda", "bin", "server"),
-			path.Join(buildPath, "cpu", "bin", "server"),
+		runners = []ModelRunner{
+			{Path: path.Join(buildPath, "cuda", "bin", "ollama-runner"), Accelerated: true},
+			{Path: path.Join(buildPath, "cpu", "bin", "ollama-runner")},
 		}
 	case "windows":
 		// TODO: select windows GPU runner here when available
-		runners = []string{
-			path.Join(buildPath, "cpu", "bin", "Release", "server.exe"),
+		runners = []ModelRunner{
+			{Path: path.Join(buildPath, "cpu", "bin", "Release", "ollama-runner.exe")},
 		}
 	default:
 		log.Printf("unknown OS, running on CPU: %s", runtime.GOOS)
-		runners = []string{
-			path.Join(buildPath, "cpu", "bin", "server"),
+		runners = []ModelRunner{
+			{Path: path.Join(buildPath, "cpu", "bin", "ollama-runner")},
 		}
 	}
 
 	runnerAvailable := false // if no runner files are found in the embed, this flag will cause a fast fail
 	for _, r := range runners {
 		// find all the files in the runner's bin directory
-		files, err := fs.Glob(llamaCppEmbed, path.Join(path.Dir(r), "*"))
+		files, err := fs.Glob(llamaCppEmbed, path.Join(path.Dir(r.Path), "*"))
 		if err != nil {
 			// this is expected, ollama may be compiled without all runners packed in
-			log.Printf("%s runner not found: %v", r, err)
+			log.Printf("%s runner not found: %v", r.Path, err)
 			continue
 		}
 
@@ -114,7 +145,10 @@ func chooseRunners(workDir, runnerType string) []ModelRunner {
 	localRunnersByPriority := []ModelRunner{}
 	for _, r := range runners {
 		// clean the ModelRunner paths so that they match the OS we are running on
-		localRunnersByPriority = append(localRunnersByPriority, ModelRunner{Path: filepath.Clean(path.Join(workDir, r))})
+		localRunnersByPriority = append(localRunnersByPriority, ModelRunner{
+			Path:        filepath.Clean(path.Join(workDir, r.Path)),
+			Accelerated: r.Accelerated,
+		})
 	}
 
 	return localRunnersByPriority
@@ -143,7 +177,7 @@ func llamaModelType(numLayer uint32) string {
 	case 80:
 		return "65B"
 	default:
-		return "Unknown"
+		return "unknown"
 	}
 }
 
@@ -177,9 +211,12 @@ type llamaHyperparameters struct {
 }
 
 type Running struct {
-	Port   int
-	Cmd    *exec.Cmd
-	Cancel context.CancelFunc
+	Port          int
+	Cmd           *exec.Cmd
+	Cancel        context.CancelFunc
+	exitOnce      sync.Once
+	exitCh        chan error // channel to receive the exit status of the subprocess
+	*StatusWriter            // captures error messages from the llama runner process
 }
 
 type llama struct {
@@ -187,63 +224,105 @@ type llama struct {
 	Running
 }
 
-var errNoGPU = errors.New("nvidia-smi command failed")
+var (
+	errNvidiaSMI     = errors.New("nvidia-smi command failed")
+	errAvailableVRAM = errors.New("not enough VRAM available, falling back to CPU only")
+)
 
-// CheckVRAM returns the available VRAM in MiB on Linux machines with NVIDIA GPUs
+// CheckVRAM returns the free VRAM in bytes on Linux machines with NVIDIA GPUs
 func CheckVRAM() (int64, error) {
-	cmd := exec.Command("nvidia-smi", "--query-gpu=memory.total", "--format=csv,noheader,nounits")
+	cmd := exec.Command("nvidia-smi", "--query-gpu=memory.free", "--format=csv,noheader,nounits")
 	var stdout bytes.Buffer
 	cmd.Stdout = &stdout
 	err := cmd.Run()
 	if err != nil {
-		return 0, errNoGPU
+		return 0, errNvidiaSMI
 	}
 
-	var total int64
+	var freeMiB int64
 	scanner := bufio.NewScanner(&stdout)
 	for scanner.Scan() {
 		line := scanner.Text()
+		if strings.Contains(line, "[Insufficient Permissions]") {
+			return 0, fmt.Errorf("GPU support may not enabled, check you have installed GPU drivers and have the necessary permissions to run nvidia-smi")
+		}
+
 		vram, err := strconv.ParseInt(strings.TrimSpace(line), 10, 64)
 		if err != nil {
 			return 0, fmt.Errorf("failed to parse available VRAM: %v", err)
 		}
 
-		total += vram
+		freeMiB += vram
 	}
 
-	return total, nil
+	freeBytes := freeMiB * 1024 * 1024
+	if freeBytes < 2*format.GigaByte {
+		log.Printf("less than 2 GB VRAM available")
+		return 0, errAvailableVRAM
+	}
+
+	return freeBytes, nil
 }
 
 func NumGPU(numLayer, fileSizeBytes int64, opts api.Options) int {
 	if opts.NumGPU != -1 {
 		return opts.NumGPU
 	}
-	n := 1 // default to enable metal on macOS
 	if runtime.GOOS == "linux" {
-		vramMib, err := CheckVRAM()
+		freeBytes, err := CheckVRAM()
 		if err != nil {
-			if err.Error() != "nvidia-smi command failed" {
+			if !errors.Is(err, errNvidiaSMI) {
 				log.Print(err.Error())
 			}
 			// nvidia driver not installed or no nvidia GPU found
 			return 0
 		}
 
-		totalVramBytes := int64(vramMib) * 1024 * 1024 // 1 MiB = 1024^2 bytes
-
-		// Calculate bytes per layer
-		// TODO: this is a rough heuristic, better would be to calculate this based on number of layers and context size
+		/*
+		 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 := fileSizeBytes / numLayer
 
-		// set n to the max number of layers we can fit in VRAM
-		return int(totalVramBytes / bytesPerLayer)
+		// 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(freeBytes/bytesPerLayer) * 3 / 4
+		log.Printf("%d MB VRAM available, loading up to %d GPU layers", freeBytes/(1024*1024), layers)
 
-		log.Printf("%d MiB VRAM available, loading up to %d GPU layers", vramMib, n)
+		return layers
 	}
 	// default to enable metal on macOS
 	return 1
 }
 
+// StatusWriter is a writer that captures error messages from the llama runner process
+type StatusWriter struct {
+	ErrCh      chan error
+	LastErrMsg string
+}
+
+func NewStatusWriter() *StatusWriter {
+	return &StatusWriter{
+		ErrCh: make(chan error, 1),
+	}
+}
+
+func (w *StatusWriter) Write(b []byte) (int, error) {
+	var errMsg string
+	if _, after, ok := bytes.Cut(b, []byte("error:")); ok {
+		errMsg = string(bytes.TrimSpace(after))
+	} else if _, after, ok := bytes.Cut(b, []byte("CUDA error")); ok {
+		errMsg = string(bytes.TrimSpace(after))
+	}
+
+	if errMsg != "" {
+		w.LastErrMsg = errMsg
+		w.ErrCh <- fmt.Errorf("llama runner: %s", errMsg)
+	}
+
+	return os.Stderr.Write(b)
+}
+
 func newLlama(model string, adapters []string, runners []ModelRunner, numLayers int64, opts api.Options) (*llama, error) {
 	fileInfo, err := os.Stat(model)
 	if err != nil {
@@ -254,16 +333,23 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 		return nil, errors.New("ollama supports only one lora adapter, but multiple were provided")
 	}
 
+	numGPU := NumGPU(numLayers, fileInfo.Size(), opts)
 	params := []string{
 		"--model", model,
 		"--ctx-size", fmt.Sprintf("%d", opts.NumCtx),
-		"--rope-freq-base", fmt.Sprintf("%f", opts.RopeFrequencyBase),
-		"--rope-freq-scale", fmt.Sprintf("%f", opts.RopeFrequencyScale),
 		"--batch-size", fmt.Sprintf("%d", opts.NumBatch),
-		"--n-gpu-layers", fmt.Sprintf("%d", NumGPU(numLayers, fileInfo.Size(), opts)),
+		"--n-gpu-layers", fmt.Sprintf("%d", numGPU),
 		"--embedding",
 	}
 
+	if opts.RopeFrequencyBase > 0 {
+		params = append(params, "--rope-freq-base", fmt.Sprintf("%f", opts.RopeFrequencyBase))
+	}
+
+	if opts.RopeFrequencyScale > 0 {
+		params = append(params, "--rope-freq-scale", fmt.Sprintf("%f", opts.RopeFrequencyScale))
+	}
+
 	if opts.NumGQA > 0 {
 		params = append(params, "--gqa", fmt.Sprintf("%d", opts.NumGQA))
 	}
@@ -290,8 +376,15 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 		params = append(params, "--numa")
 	}
 
+	var runnerErr error
+
 	// start the llama.cpp server with a retry in case the port is already in use
 	for _, runner := range runners {
+		if runner.Accelerated && numGPU == 0 {
+			log.Printf("skipping accelerated runner because num_gpu=0")
+			continue
+		}
+
 		if _, err := os.Stat(runner.Path); err != nil {
 			log.Printf("llama runner not found: %v", err)
 			continue
@@ -304,11 +397,20 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 			runner.Path,
 			append(params, "--port", strconv.Itoa(port))...,
 		)
-		cmd.Env = append(os.Environ(), fmt.Sprintf("LD_LIBRARY_PATH=%s", filepath.Dir(runner.Path)))
-		cmd.Stdout = os.Stderr
-		cmd.Stderr = os.Stderr
 
-		llm := &llama{Options: opts, Running: Running{Port: port, Cmd: cmd, Cancel: cancel}}
+		var libraryPaths []string
+		if libraryPath, ok := os.LookupEnv("LD_LIBRARY_PATH"); ok {
+			libraryPaths = append(libraryPaths, libraryPath)
+		}
+
+		libraryPaths = append(libraryPaths, filepath.Dir(runner.Path))
+
+		cmd.Env = append(os.Environ(), fmt.Sprintf("LD_LIBRARY_PATH=%s", strings.Join(libraryPaths, ":")))
+		cmd.Stdout = os.Stderr
+		statusWriter := NewStatusWriter()
+		cmd.Stderr = statusWriter
+
+		llm := &llama{Options: opts, Running: Running{Port: port, Cmd: cmd, Cancel: cancel, exitCh: make(chan error)}}
 
 		log.Print("starting llama runner")
 		if err := llm.Cmd.Start(); err != nil {
@@ -316,19 +418,36 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 			continue
 		}
 
-		// monitor the command, it is blocking, so if it exits we need to capture that
+		// monitor the llama runner process and signal when it exits
 		go func() {
-			err := llm.Cmd.Wait() // this will block until the command exits
-			if err != nil {
-				log.Printf("llama runner exited with error: %v", err)
-			} else {
-				log.Printf("llama runner exited")
+			err := llm.Cmd.Wait()
+			// default to printing the exit message of the command process, it will probably just say 'exit staus 1'
+			errMsg := err.Error()
+			// try to set a better error message if llama runner logs captured an error
+			if statusWriter.LastErrMsg != "" {
+				errMsg = statusWriter.LastErrMsg
 			}
+			log.Println(errMsg)
+			// llm.Cmd.Wait() can only be called once, use this exit channel to signal that the process has exited
+			llm.exitOnce.Do(func() {
+				close(llm.exitCh)
+			})
 		}()
 
 		if err := waitForServer(llm); err != nil {
 			log.Printf("error starting llama runner: %v", err)
 			llm.Close()
+
+			// default the runnerErr to the error returned by the most recent llama runner process
+			runnerErr = err
+
+			// capture the error directly from the runner process, if any
+			select {
+			case runnerErr = <-statusWriter.ErrCh:
+			default:
+				// the runner process probably timed out
+			}
+
 			// try again
 			continue
 		}
@@ -337,147 +456,125 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 		return llm, nil
 	}
 
+	if runnerErr != nil {
+		// this is the error returned from the llama runner process that failed most recently
+		return nil, runnerErr
+	}
+
 	return nil, fmt.Errorf("failed to start a llama runner")
 }
 
 func waitForServer(llm *llama) error {
-	// wait for the server to start responding
 	start := time.Now()
-	expiresAt := time.Now().Add(2 * time.Minute) // be generous with timeout, large models can take a while to load
+	expiresAt := time.Now().Add(3 * time.Minute) // be generous with timeout, large models can take a while to load
 	ticker := time.NewTicker(200 * time.Millisecond)
+	defer ticker.Stop()
 
 	log.Print("waiting for llama runner to start responding")
-	for range ticker.C {
-		if time.Now().After(expiresAt) {
-			return fmt.Errorf("llama runner did not start within alloted time, retrying")
-		}
-
-		// check if the server process has terminated
-		if llm.Cmd.ProcessState != nil && llm.Cmd.ProcessState.Exited() {
+	for {
+		select {
+		case <-llm.exitCh:
+			// failed to start subprocess
 			return fmt.Errorf("llama runner process has terminated")
-		}
+		case <-ticker.C:
+			if time.Now().After(expiresAt) {
+				// timeout
+				return fmt.Errorf("timed out waiting for llama runner to start")
+			}
 
-		if err := llm.Ping(context.Background()); err == nil {
-			break
+			if err := llm.Ping(context.Background()); err == nil {
+				// success
+				log.Printf("llama runner started in %f seconds", time.Since(start).Seconds())
+				return nil
+			}
 		}
 	}
-
-	log.Printf("llama runner started in %f seconds", time.Since(start).Seconds())
-	return nil
 }
 
 func (llm *llama) Close() {
+	// signal the sub-process to terminate
 	llm.Cancel()
+
+	// wait for the command to exit to prevent race conditions with the next run
+	<-llm.exitCh
+
+	if llm.StatusWriter != nil && llm.StatusWriter.LastErrMsg != "" {
+		log.Printf("llama runner stopped with error: %v", llm.StatusWriter.LastErrMsg)
+	} else {
+		log.Print("llama runner stopped successfully")
+	}
 }
 
 func (llm *llama) SetOptions(opts api.Options) {
 	llm.Options = opts
 }
 
-type GenerationSettings struct {
-	FrequencyPenalty float64       `json:"frequency_penalty"`
-	IgnoreEOS        bool          `json:"ignore_eos"`
-	LogitBias        []interface{} `json:"logit_bias"`
-	Mirostat         int           `json:"mirostat"`
-	MirostatEta      float64       `json:"mirostat_eta"`
-	MirostatTau      float64       `json:"mirostat_tau"`
-	Model            string        `json:"model"`
-	NCtx             int           `json:"n_ctx"`
-	NKeep            int           `json:"n_keep"`
-	NPredict         int           `json:"n_predict"`
-	NProbs           int           `json:"n_probs"`
-	PenalizeNl       bool          `json:"penalize_nl"`
-	PresencePenalty  float64       `json:"presence_penalty"`
-	RepeatLastN      int           `json:"repeat_last_n"`
-	RepeatPenalty    float64       `json:"repeat_penalty"`
-	Seed             uint32        `json:"seed"`
-	Stop             []string      `json:"stop"`
-	Stream           bool          `json:"stream"`
-	Temp             float64       `json:"temp"`
-	TfsZ             float64       `json:"tfs_z"`
-	TopK             int           `json:"top_k"`
-	TopP             float64       `json:"top_p"`
-	TypicalP         float64       `json:"typical_p"`
-}
-
-type Timings struct {
-	PredictedN  int     `json:"predicted_n"`
-	PredictedMS float64 `json:"predicted_ms"`
-	PromptN     int     `json:"prompt_n"`
-	PromptMS    float64 `json:"prompt_ms"`
-}
-
-type Prediction struct {
+type prediction struct {
 	Content string `json:"content"`
 	Model   string `json:"model"`
 	Prompt  string `json:"prompt"`
 	Stop    bool   `json:"stop"`
 
-	Timings `json:"timings"`
+	Timings struct {
+		PredictedN  int     `json:"predicted_n"`
+		PredictedMS float64 `json:"predicted_ms"`
+		PromptN     int     `json:"prompt_n"`
+		PromptMS    float64 `json:"prompt_ms"`
+	}
 }
 
-type PredictRequest struct {
-	Stream           bool            `json:"stream"`
-	NPredict         int             `json:"n_predict,omitempty"`
-	TopK             int             `json:"top_k,omitempty"`
-	TopP             float32         `json:"top_p,omitempty"`
-	TfsZ             float32         `json:"tfs_z,omitempty"`
-	TypicalP         float32         `json:"typical_p,omitempty"`
-	RepeatLastN      int             `json:"repeat_last_n,omitempty"`
-	Temperature      float32         `json:"temperature,omitempty"`
-	RepeatPenalty    float32         `json:"repeat_penalty,omitempty"`
-	PresencePenalty  float32         `json:"presence_penalty,omitempty"`
-	FrequencyPenalty float32         `json:"frequency_penalty,omitempty"`
-	Mirostat         int             `json:"mirostat,omitempty"`
-	MirostatTau      float32         `json:"mirostat_tau,omitempty"`
-	MirostatEta      float32         `json:"mirostat_eta,omitempty"`
-	PenalizeNl       bool            `json:"penalize_nl,omitempty"`
-	NKeep            int             `json:"n_keep,omitempty"`
-	Seed             int             `json:"seed,omitempty"`
-	Prompt           string          `json:"prompt,omitempty"`
-	NProbs           int             `json:"n_probs,omitempty"`
-	LogitBias        map[int]float32 `json:"logit_bias,omitempty"`
-	IgnoreEos        bool            `json:"ignore_eos,omitempty"`
-	Stop             []string        `json:"stop,omitempty"`
-}
+const maxBufferSize = 512 * format.KiloByte
 
-func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string, fn func(api.GenerateResponse)) error {
+func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string, format string, fn func(api.GenerateResponse)) error {
 	prevConvo, err := llm.Decode(ctx, prevContext)
 	if err != nil {
 		return err
 	}
 
+	// Remove leading spaces from prevConvo if present
+	prevConvo = strings.TrimPrefix(prevConvo, " ")
+
 	var nextContext strings.Builder
 	nextContext.WriteString(prevConvo)
 	nextContext.WriteString(prompt)
 
-	endpoint := fmt.Sprintf("http://127.0.0.1:%d/completion", llm.Port)
-	predReq := PredictRequest{
-		Prompt:           nextContext.String(),
-		Stream:           true,
-		NPredict:         llm.NumPredict,
-		NKeep:            llm.NumKeep,
-		Temperature:      llm.Temperature,
-		TopK:             llm.TopK,
-		TopP:             llm.TopP,
-		TfsZ:             llm.TFSZ,
-		TypicalP:         llm.TypicalP,
-		RepeatLastN:      llm.RepeatLastN,
-		RepeatPenalty:    llm.RepeatPenalty,
-		PresencePenalty:  llm.PresencePenalty,
-		FrequencyPenalty: llm.FrequencyPenalty,
-		Mirostat:         llm.Mirostat,
-		MirostatTau:      llm.MirostatTau,
-		MirostatEta:      llm.MirostatEta,
-		PenalizeNl:       llm.PenalizeNewline,
-		Stop:             llm.Stop,
-	}
-	data, err := json.Marshal(predReq)
-	if err != nil {
-		return fmt.Errorf("error marshaling data: %v", err)
+	request := map[string]any{
+		"prompt":            nextContext.String(),
+		"stream":            true,
+		"n_predict":         llm.NumPredict,
+		"n_keep":            llm.NumKeep,
+		"temperature":       llm.Temperature,
+		"top_k":             llm.TopK,
+		"top_p":             llm.TopP,
+		"tfs_z":             llm.TFSZ,
+		"typical_p":         llm.TypicalP,
+		"repeat_last_n":     llm.RepeatLastN,
+		"repeat_penalty":    llm.RepeatPenalty,
+		"presence_penalty":  llm.PresencePenalty,
+		"frequency_penalty": llm.FrequencyPenalty,
+		"mirostat":          llm.Mirostat,
+		"mirostat_tau":      llm.MirostatTau,
+		"mirostat_eta":      llm.MirostatEta,
+		"penalize_nl":       llm.PenalizeNewline,
+		"seed":              llm.Seed,
+		"stop":              llm.Stop,
 	}
 
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, bytes.NewBuffer(data))
+	if format == "json" {
+		request["grammar"] = jsonGrammar
+	}
+
+	// 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: %v", err)
+	}
+
+	endpoint := fmt.Sprintf("http://127.0.0.1:%d/completion", llm.Port)
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, buffer)
 	if err != nil {
 		return fmt.Errorf("error creating POST request: %v", err)
 	}
@@ -499,22 +596,23 @@ func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string,
 	}
 
 	scanner := bufio.NewScanner(resp.Body)
+	// increase the buffer size to avoid running out of space
+	buf := make([]byte, 0, maxBufferSize)
+	scanner.Buffer(buf, maxBufferSize)
 	for scanner.Scan() {
 		select {
 		case <-ctx.Done():
 			// This handles the request cancellation
 			return ctx.Err()
 		default:
-			line := scanner.Text()
-			if line == "" {
+			line := scanner.Bytes()
+			if len(line) == 0 {
 				continue
 			}
 
-			// Read data from the server-side event stream
-			if strings.HasPrefix(line, "data: ") {
-				evt := line[6:]
-				var p Prediction
-				if err := json.Unmarshal([]byte(evt), &p); err != nil {
+			if evt, ok := bytes.CutPrefix(line, []byte("data: ")); ok {
+				var p prediction
+				if err := json.Unmarshal(evt, &p); err != nil {
 					return fmt.Errorf("error unmarshaling llm prediction response: %v", err)
 				}
 
@@ -532,10 +630,10 @@ func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string,
 					fn(api.GenerateResponse{
 						Done:               true,
 						Context:            embd,
-						PromptEvalCount:    p.PromptN,
-						PromptEvalDuration: parseDurationMs(p.PromptMS),
-						EvalCount:          p.PredictedN,
-						EvalDuration:       parseDurationMs(p.PredictedMS),
+						PromptEvalCount:    p.Timings.PromptN,
+						PromptEvalDuration: parseDurationMs(p.Timings.PromptMS),
+						EvalCount:          p.Timings.PredictedN,
+						EvalDuration:       parseDurationMs(p.Timings.PredictedMS),
 					})
 
 					return nil
@@ -545,6 +643,14 @@ func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string,
 	}
 
 	if err := scanner.Err(); err != nil {
+		if strings.Contains(err.Error(), "unexpected EOF") {
+			// this means the llama runner subprocess crashed
+			llm.Close()
+			if llm.StatusWriter != nil && llm.StatusWriter.LastErrMsg != "" {
+				return fmt.Errorf("llama runner exited: %v", llm.StatusWriter.LastErrMsg)
+			}
+			return fmt.Errorf("llama runner exited, you may not have enough available memory to run this model")
+		}
 		return fmt.Errorf("error reading llm response: %v", err)
 	}
 
@@ -641,9 +747,6 @@ func (llm *llama) Decode(ctx context.Context, tokens []int) (string, error) {
 		return "", fmt.Errorf("unmarshal encode response: %w", err)
 	}
 
-	// decoded content contains a leading whitespace
-	decoded.Content, _ = strings.CutPrefix(decoded.Content, "")
-
 	return decoded.Content, nil
 }
 

llm/llm.go

@@ -5,14 +5,16 @@ import (
 	"fmt"
 	"log"
 	"os"
+	"runtime"
 
 	"github.com/pbnjay/memory"
 
 	"github.com/jmorganca/ollama/api"
+	"github.com/jmorganca/ollama/format"
 )
 
 type LLM interface {
-	Predict(context.Context, []int, string, func(api.GenerateResponse)) error
+	Predict(context.Context, []int, string, string, func(api.GenerateResponse)) error
 	Embedding(context.Context, string) ([]float64, error)
 	Encode(context.Context, string) ([]int, error)
 	Decode(context.Context, []int) (string, error)
@@ -37,60 +39,56 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
 		return nil, err
 	}
 
-	switch ggml.FileType() {
-	case "Q8_0":
-		if ggml.Name() != "gguf" && opts.NumGPU != 0 {
-			// GGML Q8_0 do not support Metal API and will
-			// cause the runner to segmentation fault so disable GPU
-			log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
-			opts.NumGPU = 0
+	if runtime.GOOS == "darwin" {
+		switch ggml.FileType() {
+		case "Q8_0":
+			if ggml.Name() != "gguf" && opts.NumGPU != 0 {
+				// GGML Q8_0 do not support Metal API and will
+				// cause the runner to segmentation fault so disable GPU
+				log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
+				opts.NumGPU = 0
+			}
+		case "F32", "Q5_0", "Q5_1":
+			if opts.NumGPU != 0 {
+				// F32, Q5_0, Q5_1, and Q8_0 do not support Metal API and will
+				// cause the runner to segmentation fault so disable GPU
+				log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
+				opts.NumGPU = 0
+			}
 		}
-	case "F32", "Q5_0", "Q5_1":
-		if opts.NumGPU != 0 {
-			// F32, Q5_0, Q5_1, and Q8_0 do not support Metal API and will
-			// cause the runner to segmentation fault so disable GPU
-			log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
-			opts.NumGPU = 0
-		}
-	}
 
-	totalResidentMemory := memory.TotalMemory()
-	switch ggml.ModelType() {
-	case "3B", "7B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 16*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 16GB of memory")
-		} else if totalResidentMemory < 8*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 8GB of memory")
+		var requiredMemory int64
+		var f16Multiplier int64 = 2
+
+		switch ggml.ModelType() {
+		case "3B", "7B":
+			requiredMemory = 8 * format.GigaByte
+		case "13B":
+			requiredMemory = 16 * format.GigaByte
+		case "30B", "34B", "40B":
+			requiredMemory = 32 * format.GigaByte
+		case "65B", "70B":
+			requiredMemory = 64 * format.GigaByte
+		case "180B":
+			requiredMemory = 128 * format.GigaByte
+			f16Multiplier = 4
 		}
-	case "13B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 32*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 32GB of memory")
-		} else if totalResidentMemory < 16*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 16GB of memory")
-		}
-	case "30B", "34B", "40B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 64*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 64GB of memory")
-		} else if totalResidentMemory < 32*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 32GB of memory")
-		}
-	case "65B", "70B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 128*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 128GB of memory")
-		} else if totalResidentMemory < 64*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 64GB of memory")
-		}
-	case "180B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 512*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 512GB of memory")
-		} else if totalResidentMemory < 128*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 128GB of memory")
+
+		systemMemory := int64(memory.TotalMemory())
+
+		if ggml.FileType() == "F16" && requiredMemory*f16Multiplier > systemMemory {
+			return nil, fmt.Errorf("F16 model requires at least %s of total memory", format.HumanBytes(requiredMemory))
+		} else if requiredMemory > systemMemory {
+			return nil, fmt.Errorf("model requires at least %s of total memory", format.HumanBytes(requiredMemory))
 		}
 	}
 
 	switch ggml.Name() {
 	case "gguf":
-		opts.NumGQA = 0 // TODO: remove this when llama.cpp runners differ enough to need separate newLlama functions
+		// TODO: gguf will load these options automatically from the model binary
+		opts.NumGQA = 0
+		opts.RopeFrequencyBase = 0.0
+		opts.RopeFrequencyScale = 0.0
 		return newLlama(model, adapters, chooseRunners(workDir, "gguf"), ggml.NumLayers(), opts)
 	case "ggml", "ggmf", "ggjt", "ggla":
 		return newLlama(model, adapters, chooseRunners(workDir, "ggml"), ggml.NumLayers(), opts)

llm/starcoder.go

@@ -0,0 +1,23 @@
+package llm
+
+const (
+	starCoderModelType1B  = 24
+	starCoderModelType3B  = 36
+	starCoderModelType7B  = 42
+	starCoderModelType15B = 40
+)
+
+func starCoderModelType(numLayer uint32) string {
+	switch numLayer {
+	case 24:
+		return "1B"
+	case 36:
+		return "3B"
+	case 42:
+		return "7B"
+	case 40:
+		return "15B"
+	default:
+		return "unknown"
+	}
+}

parser/parser.go

@@ -40,7 +40,7 @@ func Parse(reader io.Reader) ([]Command, error) {
 			command.Args = string(fields[1])
 			// copy command for validation
 			modelCommand = command
-		case "LICENSE", "TEMPLATE", "SYSTEM", "PROMPT", "EMBED", "ADAPTER":
+		case "LICENSE", "TEMPLATE", "SYSTEM", "PROMPT", "ADAPTER":
 			command.Name = string(bytes.ToLower(fields[0]))
 			command.Args = string(fields[1])
 		case "PARAMETER":
@@ -51,6 +51,8 @@ func Parse(reader io.Reader) ([]Command, error) {
 
 			command.Name = string(fields[0])
 			command.Args = string(fields[1])
+		case "EMBED":
+			return nil, fmt.Errorf("deprecated command: EMBED is no longer supported, use the /embed API endpoint instead")
 		default:
 			if !bytes.HasPrefix(fields[0], []byte("#")) {
 				// log a warning for unknown commands

progressbar/progressbar.go

@@ -291,7 +291,7 @@ func OptionShowDescriptionAtLineEnd() Option {
 	}
 }
 
-var defaultTheme = Theme{Saucer: "█", SaucerPadding: " ", BarStart: "|", BarEnd: "|"}
+var defaultTheme = Theme{Saucer: "█", SaucerPadding: " ", BarStart: "▕", BarEnd: "▏"}
 
 // NewOptions constructs a new instance of ProgressBar, with any options you specify
 func NewOptions(max int, options ...Option) *ProgressBar {

readline/buffer.go

@@ -0,0 +1,372 @@
+package readline
+
+import (
+	"fmt"
+	"os"
+
+	"github.com/emirpasic/gods/lists/arraylist"
+	"golang.org/x/term"
+)
+
+type Buffer struct {
+	Pos       int
+	Buf       *arraylist.List
+	Prompt    *Prompt
+	LineWidth int
+	Width     int
+	Height    int
+}
+
+func NewBuffer(prompt *Prompt) (*Buffer, error) {
+	fd := int(os.Stdout.Fd())
+	width, height, err := term.GetSize(fd)
+	if err != nil {
+		fmt.Println("Error getting size:", err)
+		return nil, err
+	}
+
+	lwidth := width - len(prompt.Prompt)
+	if prompt.UseAlt {
+		lwidth = width - len(prompt.AltPrompt)
+	}
+
+	b := &Buffer{
+		Pos:       0,
+		Buf:       arraylist.New(),
+		Prompt:    prompt,
+		Width:     width,
+		Height:    height,
+		LineWidth: lwidth,
+	}
+
+	return b, nil
+}
+
+func (b *Buffer) MoveLeft() {
+	if b.Pos > 0 {
+		if b.Pos%b.LineWidth == 0 {
+			fmt.Printf(CursorUp + CursorBOL + cursorRightN(b.Width))
+		} else {
+			fmt.Print(CursorLeft)
+		}
+		b.Pos -= 1
+	}
+}
+
+func (b *Buffer) MoveLeftWord() {
+	if b.Pos > 0 {
+		var foundNonspace bool
+		for {
+			v, _ := b.Buf.Get(b.Pos - 1)
+			if v == ' ' {
+				if foundNonspace {
+					break
+				}
+			} else {
+				foundNonspace = true
+			}
+			b.MoveLeft()
+
+			if b.Pos == 0 {
+				break
+			}
+		}
+	}
+}
+
+func (b *Buffer) MoveRight() {
+	if b.Pos < b.Size() {
+		b.Pos += 1
+		if b.Pos%b.LineWidth == 0 {
+			fmt.Printf(CursorDown + CursorBOL + cursorRightN(b.PromptSize()))
+		} else {
+			fmt.Print(CursorRight)
+		}
+	}
+}
+
+func (b *Buffer) MoveRightWord() {
+	if b.Pos < b.Size() {
+		for {
+			b.MoveRight()
+			v, _ := b.Buf.Get(b.Pos)
+			if v == ' ' {
+				break
+			}
+
+			if b.Pos == b.Size() {
+				break
+			}
+		}
+	}
+}
+
+func (b *Buffer) MoveToStart() {
+	if b.Pos > 0 {
+		currLine := b.Pos / b.LineWidth
+		if currLine > 0 {
+			for cnt := 0; cnt < currLine; cnt++ {
+				fmt.Print(CursorUp)
+			}
+		}
+		fmt.Printf(CursorBOL + cursorRightN(b.PromptSize()))
+		b.Pos = 0
+	}
+}
+
+func (b *Buffer) MoveToEnd() {
+	if b.Pos < b.Size() {
+		currLine := b.Pos / b.LineWidth
+		totalLines := b.Size() / b.LineWidth
+		if currLine < totalLines {
+			for cnt := 0; cnt < totalLines-currLine; cnt++ {
+				fmt.Print(CursorDown)
+			}
+			remainder := b.Size() % b.LineWidth
+			fmt.Printf(CursorBOL + cursorRightN(b.PromptSize()+remainder))
+		} else {
+			fmt.Print(cursorRightN(b.Size() - b.Pos))
+		}
+
+		b.Pos = b.Size()
+	}
+}
+
+func (b *Buffer) Size() int {
+	return b.Buf.Size()
+}
+
+func min(n, m int) int {
+	if n > m {
+		return m
+	}
+	return n
+}
+
+func (b *Buffer) PromptSize() int {
+	if b.Prompt.UseAlt {
+		return len(b.Prompt.AltPrompt)
+	}
+	return len(b.Prompt.Prompt)
+}
+
+func (b *Buffer) Add(r rune) {
+	if b.Pos == b.Buf.Size() {
+		fmt.Printf("%c", r)
+		b.Buf.Add(r)
+		b.Pos += 1
+		if b.Pos > 0 && b.Pos%b.LineWidth == 0 {
+			fmt.Printf("\n%s", b.Prompt.AltPrompt)
+		}
+	} else {
+		fmt.Printf("%c", r)
+		b.Buf.Insert(b.Pos, r)
+		b.Pos += 1
+		if b.Pos > 0 && b.Pos%b.LineWidth == 0 {
+			fmt.Printf("\n%s", b.Prompt.AltPrompt)
+		}
+		b.drawRemaining()
+	}
+}
+
+func (b *Buffer) drawRemaining() {
+	var place int
+	remainingText := b.StringN(b.Pos)
+	if b.Pos > 0 {
+		place = b.Pos % b.LineWidth
+	}
+	fmt.Print(CursorHide)
+
+	// render the rest of the current line
+	currLine := remainingText[:min(b.LineWidth-place, len(remainingText))]
+	if len(currLine) > 0 {
+		fmt.Printf(ClearToEOL + currLine)
+		fmt.Print(cursorLeftN(len(currLine)))
+	} else {
+		fmt.Print(ClearToEOL)
+	}
+
+	// render the other lines
+	if len(remainingText) > len(currLine) {
+		remaining := []rune(remainingText[len(currLine):])
+		var totalLines int
+		for i, c := range remaining {
+			if i%b.LineWidth == 0 {
+				fmt.Printf("\n%s", b.Prompt.AltPrompt)
+				totalLines += 1
+			}
+			fmt.Printf("%c", c)
+		}
+		fmt.Print(ClearToEOL)
+		fmt.Print(cursorUpN(totalLines))
+		fmt.Printf(CursorBOL + cursorRightN(b.Width-len(currLine)))
+	}
+
+	fmt.Print(CursorShow)
+}
+
+func (b *Buffer) Remove() {
+	if b.Buf.Size() > 0 && b.Pos > 0 {
+		if b.Pos%b.LineWidth == 0 {
+			// if the user backspaces over the word boundary, do this magic to clear the line
+			// and move to the end of the previous line
+			fmt.Printf(CursorBOL + ClearToEOL)
+			fmt.Printf(CursorUp + CursorBOL + cursorRightN(b.Width) + " " + CursorLeft)
+		} else {
+			fmt.Printf(CursorLeft + " " + CursorLeft)
+		}
+
+		var eraseExtraLine bool
+		if (b.Size()-1)%b.LineWidth == 0 {
+			eraseExtraLine = true
+		}
+
+		b.Pos -= 1
+		b.Buf.Remove(b.Pos)
+
+		if b.Pos < b.Size() {
+			b.drawRemaining()
+			// this erases a line which is left over when backspacing in the middle of a line and there
+			// are trailing characters which go over the line width boundary
+			if eraseExtraLine {
+				remainingLines := (b.Size() - b.Pos) / b.LineWidth
+				fmt.Printf(cursorDownN(remainingLines+1) + CursorBOL + ClearToEOL)
+				place := b.Pos % b.LineWidth
+				fmt.Printf(cursorUpN(remainingLines+1) + cursorRightN(place+len(b.Prompt.Prompt)))
+			}
+		}
+	}
+}
+
+func (b *Buffer) Delete() {
+	if b.Size() > 0 && b.Pos < b.Size() {
+		b.Buf.Remove(b.Pos)
+		b.drawRemaining()
+		if b.Size()%b.LineWidth == 0 {
+			if b.Pos != b.Size() {
+				remainingLines := (b.Size() - b.Pos) / b.LineWidth
+				fmt.Printf(cursorDownN(remainingLines) + CursorBOL + ClearToEOL)
+				place := b.Pos % b.LineWidth
+				fmt.Printf(cursorUpN(remainingLines) + cursorRightN(place+len(b.Prompt.Prompt)))
+			}
+		}
+	}
+}
+
+func (b *Buffer) DeleteBefore() {
+	if b.Pos > 0 {
+		for cnt := b.Pos - 1; cnt >= 0; cnt-- {
+			b.Remove()
+		}
+	}
+}
+
+func (b *Buffer) DeleteRemaining() {
+	if b.Size() > 0 && b.Pos < b.Size() {
+		charsToDel := b.Size() - b.Pos
+		for cnt := 0; cnt < charsToDel; cnt++ {
+			b.Delete()
+		}
+	}
+}
+
+func (b *Buffer) DeleteWord() {
+	if b.Buf.Size() > 0 && b.Pos > 0 {
+		var foundNonspace bool
+		for {
+			v, _ := b.Buf.Get(b.Pos - 1)
+			if v == ' ' {
+				if !foundNonspace {
+					b.Remove()
+				} else {
+					break
+				}
+			} else {
+				foundNonspace = true
+				b.Remove()
+			}
+
+			if b.Pos == 0 {
+				break
+			}
+		}
+	}
+}
+
+func (b *Buffer) ClearScreen() {
+	fmt.Printf(ClearScreen + CursorReset + b.Prompt.Prompt)
+	if b.IsEmpty() {
+		ph := b.Prompt.Placeholder
+		fmt.Printf(ColorGrey + ph + cursorLeftN(len(ph)) + ColorDefault)
+	} else {
+		currPos := b.Pos
+		b.Pos = 0
+		b.drawRemaining()
+		fmt.Printf(CursorReset + cursorRightN(len(b.Prompt.Prompt)))
+		if currPos > 0 {
+			targetLine := currPos / b.LineWidth
+			if targetLine > 0 {
+				for cnt := 0; cnt < targetLine; cnt++ {
+					fmt.Print(CursorDown)
+				}
+			}
+			remainder := currPos % b.LineWidth
+			if remainder > 0 {
+				fmt.Print(cursorRightN(remainder))
+			}
+			if currPos%b.LineWidth == 0 {
+				fmt.Printf(CursorBOL + b.Prompt.AltPrompt)
+			}
+		}
+		b.Pos = currPos
+	}
+}
+
+func (b *Buffer) IsEmpty() bool {
+	return b.Buf.Empty()
+}
+
+func (b *Buffer) Replace(r []rune) {
+	b.Pos = 0
+	b.Buf.Clear()
+	fmt.Printf(ClearLine + CursorBOL + b.Prompt.Prompt)
+	for _, c := range r {
+		b.Add(c)
+	}
+}
+
+func (b *Buffer) String() string {
+	return b.StringN(0)
+}
+
+func (b *Buffer) StringN(n int) string {
+	return b.StringNM(n, 0)
+}
+
+func (b *Buffer) StringNM(n, m int) string {
+	var s string
+	if m == 0 {
+		m = b.Size()
+	}
+	for cnt := n; cnt < m; cnt++ {
+		c, _ := b.Buf.Get(cnt)
+		s += string(c.(rune))
+	}
+	return s
+}
+
+func cursorLeftN(n int) string {
+	return fmt.Sprintf(CursorLeftN, n)
+}
+
+func cursorRightN(n int) string {
+	return fmt.Sprintf(CursorRightN, n)
+}
+
+func cursorUpN(n int) string {
+	return fmt.Sprintf(CursorUpN, n)
+}
+
+func cursorDownN(n int) string {
+	return fmt.Sprintf(CursorDownN, n)
+}