log probs working

api/client.go

@@ -129,7 +129,7 @@ func (c *Client) do(ctx context.Context, method, path string, reqData, respData
 	return nil
 }
 
-const maxBufferSize = 512 * format.KiloByte
+const maxBufferSize = 1024 * format.KiloByte
 
 func (c *Client) stream(ctx context.Context, method, path string, data any, fn func([]byte) error) error {
 	var buf *bytes.Buffer

api/types.go

@@ -80,6 +80,8 @@ type GenerateRequest struct {
 	// Options lists model-specific options. For example, temperature can be
 	// set through this field, if the model supports it.
 	Options map[string]interface{} `json:"options"`
+
+	ReturnLogits bool `json:"return_logits,omitempty"`
 }
 
 // ChatRequest describes a request sent by [Client.Chat].
@@ -105,6 +107,8 @@ type ChatRequest struct {
 
 	// Options lists model-specific options.
 	Options map[string]interface{} `json:"options"`
+
+	ReturnLogits bool `json:"return_logits,omitempty"`
 }
 
 type Tools []Tool
@@ -185,10 +189,12 @@ func (t *ToolFunction) String() string {
 // ChatResponse is the response returned by [Client.Chat]. Its fields are
 // similar to [GenerateResponse].
 type ChatResponse struct {
-	Model      string    `json:"model"`
-	CreatedAt  time.Time `json:"created_at"`
-	Message    Message   `json:"message"`
-	DoneReason string    `json:"done_reason,omitempty"`
+	Model       string         `json:"model"`
+	CreatedAt   time.Time      `json:"created_at"`
+	Message     Message        `json:"message"`
+	DoneReason  string         `json:"done_reason,omitempty"`
+	Logits      []float32      `json:"logits"`
+	TopLogprobs []TokenLogprob `json:"top_logprobs"`
 
 	Done bool `json:"done"`
 
@@ -204,6 +210,11 @@ type Metrics struct {
 	EvalDuration       time.Duration `json:"eval_duration,omitempty"`
 }
 
+type TokenLogprob struct {
+	Text    string  `json:"text"`
+	Logprob float32 `json:"logprob"`
+}
+
 // Options specified in [GenerateRequest].  If you add a new option here, also
 // add it to the API docs.
 type Options struct {
@@ -450,6 +461,8 @@ type GenerateResponse struct {
 	Context []int `json:"context,omitempty"`
 
 	Metrics
+
+	Logits []float32 `json:"logits"`
 }
 
 // ModelDetails provides details about a model.

llama/llama.go

@@ -260,6 +260,31 @@ func (c *Context) GetEmbeddingsIth(i int) []float32 {
 	return unsafe.Slice((*float32)(embeddings), c.Model().NEmbd())
 }
 
+// GetLogits returns the logits from the last decode operation.
+// The returned slice has length equal to the vocabulary size.
+func (c *Context) GetLogits() []float32 {
+	logits := unsafe.Pointer(C.llama_get_logits(c.c))
+	if logits == nil {
+		return nil
+	}
+
+	// Get the number of vocabulary tokens to determine array size
+	vocabSize := c.Model().NumVocab()
+	return unsafe.Slice((*float32)(logits), vocabSize)
+}
+
+func (m *Model) Detokenize(tokens []int) (string, error) {
+	var text string
+	for _, token := range tokens {
+		piece := m.TokenToPiece(token)
+		if piece == "" {
+			return "", fmt.Errorf("failed to convert token %d to piece", token)
+		}
+		text += piece
+	}
+	return text, nil
+}
+
 type ModelParams struct {
 	NumGpuLayers int
 	MainGpu      int

llama/runner/runner.go

@@ -8,6 +8,7 @@ import (
 	"fmt"
 	"log"
 	"log/slog"
+	"math"
 	"net"
 	"net/http"
 	"os"
@@ -59,7 +60,7 @@ type Sequence struct {
 	crossAttention bool
 
 	// channel to send responses over
-	responses chan string
+	responses chan CompletionResponse
 
 	// channel to stop decoding (such as if the remote connection is closed)
 	quit chan bool
@@ -88,6 +89,15 @@ type Sequence struct {
 	startGenerationTime time.Time
 	numDecoded          int
 	numPromptInputs     int
+
+	// New flag we need to add to Sequence struct
+	returnLogits bool
+
+	// Using our new GetLogits() method
+	logits []float32
+
+	// Add new channel for logits
+	logitsOut chan []float32
 }
 
 type NewSequenceParams struct {
@@ -96,6 +106,7 @@ type NewSequenceParams struct {
 	numKeep        int
 	samplingParams *llama.SamplingParams
 	embedding      bool
+	returnLogits   bool
 }
 
 func (s *Server) NewSequence(prompt string, images []ImageData, params NewSequenceParams) (*Sequence, error) {
@@ -149,13 +160,15 @@ func (s *Server) NewSequence(prompt string, images []ImageData, params NewSequen
 		startProcessingTime: startTime,
 		numPredict:          params.numPredict,
 		pendingResponses:    make([]string, 0),
-		responses:           make(chan string, 100),
+		responses:           make(chan CompletionResponse, 100),
 		quit:                make(chan bool, 1),
 		embedding:           make(chan []float32, 1),
 		samplingCtx:         sc,
 		embeddingOnly:       params.embedding,
 		stop:                params.stop,
 		numKeep:             params.numKeep,
+		returnLogits:        params.returnLogits,
+		logitsOut:           make(chan []float32, 100),
 	}, nil
 }
 
@@ -274,25 +287,34 @@ func (s *Server) allNil() bool {
 }
 
 func flushPending(seq *Sequence) bool {
-	joined := strings.Join(seq.pendingResponses, "")
-	seq.pendingResponses = []string{}
+	if len(seq.pendingResponses) == 0 {
+		return true
+	}
 
+	content := strings.Join(seq.pendingResponses, "")
 	// Check if there are any partial UTF-8 characters remaining.
 	// We already check and queue as we are generating but some may
 	// still make it here:
 	// - Sequence is ending, e.g. generation limit has been hit
 	// - Invalid characters in the middle of a string
 	// This is a stricter check to ensure we never output invalid Unicode.
-	for !utf8.ValidString(joined) {
-		joined = joined[:len(joined)-1]
+	for !utf8.ValidString(content) {
+		content = content[:len(content)-1]
+	}
+	seq.pendingResponses = nil
+
+	resp := CompletionResponse{
+		Content: content,
 	}
 
-	if len(joined) == 0 {
-		return true
+	// Add logits if requested and available
+	if seq.returnLogits && seq.logits != nil {
+		resp.Logits = seq.logits
+		seq.logits = nil
 	}
 
 	select {
-	case seq.responses <- joined:
+	case seq.responses <- resp:
 		return true
 	case <-seq.quit:
 		return false
@@ -476,7 +498,14 @@ func (s *Server) processBatch(tokenBatch *llama.Batch, embedBatch *llama.Batch)
 			continue
 		}
 
-		// sample a token
+		// Before sampling:
+		if seq.returnLogits { // New flag we need to add to Sequence struct
+			logits := s.lc.GetLogits()
+			seq.logits = make([]float32, len(logits))
+			copy(seq.logits, logits)
+		}
+
+		// Then sample token
 		token := seq.samplingCtx.Sample(s.lc, seq.iBatch)
 		seq.samplingCtx.Accept(token, true)
 		piece := s.model.TokenToPiece(token)
@@ -572,10 +601,11 @@ type ImageData struct {
 }
 
 type CompletionRequest struct {
-	Prompt      string      `json:"prompt"`
-	Images      []ImageData `json:"image_data"`
-	Grammar     string      `json:"grammar"`
-	CachePrompt bool        `json:"cache_prompt"`
+	Prompt       string      `json:"prompt"`
+	Images       []ImageData `json:"image_data"`
+	Grammar      string      `json:"grammar"`
+	CachePrompt  bool        `json:"cache_prompt"`
+	ReturnLogits bool        `json:"return_logits,omitempty"` // defaults to false
 
 	Options
 }
@@ -588,8 +618,10 @@ type Timings struct {
 }
 
 type CompletionResponse struct {
-	Content string `json:"content"`
-	Stop    bool   `json:"stop"`
+	Content string    `json:"content"`
+	Logits  []float32 `json:"logits,omitempty"`
+	Tokens  []string  `json:"tokens,omitempty"`
+	Stop    bool      `json:"stop"`
 
 	Model        string  `json:"model,omitempty"`
 	Prompt       string  `json:"prompt,omitempty"`
@@ -637,12 +669,14 @@ func (s *Server) completion(w http.ResponseWriter, r *http.Request) {
 	samplingParams.Seed = uint32(req.Seed)
 	samplingParams.Grammar = req.Grammar
 
+	slog.Info("completion request", "return_logits", req.ReturnLogits)
 	seq, err := s.NewSequence(req.Prompt, req.Images, NewSequenceParams{
 		numPredict:     req.NumPredict,
 		stop:           req.Stop,
 		numKeep:        req.NumKeep,
 		samplingParams: &samplingParams,
 		embedding:      false,
+		returnLogits:   req.ReturnLogits,
 	})
 	if err != nil {
 		http.Error(w, fmt.Sprintf("Failed to create new sequence: %v", err), http.StatusInternalServerError)
@@ -692,9 +726,8 @@ func (s *Server) completion(w http.ResponseWriter, r *http.Request) {
 			return
 		case content, ok := <-seq.responses:
 			if ok {
-				if err := json.NewEncoder(w).Encode(&CompletionResponse{
-					Content: content,
-				}); err != nil {
+				// slog.Info("content", "content", content.Content)
+				if err := json.NewEncoder(w).Encode(&content); err != nil {
 					http.Error(w, fmt.Sprintf("failed to encode response: %v", err), http.StatusInternalServerError)
 					close(seq.quit)
 					return
@@ -1003,3 +1036,76 @@ func Execute(args []string) error {
 	cancel()
 	return nil
 }
+
+// // Helper function to get top K logits and convert to log probabilities
+// func getTopLogits(logits []float32, k int, model *llama.Model) []api.LogProbs {
+// 	if k <= 0 {
+// 		return nil
+// 	}
+
+// 	// Convert logits to probabilities using softmax
+// 	probs := softmax(logits)
+
+// 	// Create slice of index/probability pairs
+// 	pairs := make([]struct {
+// 		token int
+// 		prob  float32
+// 	}, len(probs))
+
+// 	for i, p := range probs {
+// 		pairs[i] = struct {
+// 			token int
+// 			prob  float32
+// 		}{i, p}
+// 	}
+
+// 	// Sort by probability (descending)
+// 	sort.Slice(pairs, func(i, j int) bool {
+// 		return pairs[i].prob > pairs[j].prob
+// 	})
+
+// 	// Take top K
+// 	k = min(k, len(pairs))
+// 	result := make([]api.LogProbs, k)
+
+// 	for i := 0; i < k; i++ {
+// 		result[i] = api.LogProbs{
+// 			TopLogprobs: []api.TokenLogprob{
+// 				{
+// 					Token:   model.TokenToPiece(pairs[i].token),
+// 					Logprob: float32(math.Log(float64(pairs[i].prob))),
+// 				},
+// 			},
+// 		}
+// 	}
+
+// 	return result
+// }
+
+// Helper function to compute softmax
+func softmax(logits []float32) []float32 {
+	probs := make([]float32, len(logits))
+
+	// Find max for numerical stability
+	max := float32(math.Inf(-1))
+	for _, l := range logits {
+		if l > max {
+			max = l
+		}
+	}
+
+	// Compute exp(x - max) and sum
+	sum := float32(0)
+	for i, l := range logits {
+		ex := float32(math.Exp(float64(l - max)))
+		probs[i] = ex
+		sum += ex
+	}
+
+	// Normalize
+	for i := range probs {
+		probs[i] /= sum
+	}
+
+	return probs
+}

llm/server.go

@@ -633,7 +633,8 @@ number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
 ws ::= ([ \t\n] ws)?
 `
 
-const maxBufferSize = 512 * format.KiloByte
+// TODO: change back to 512 * format.KiloByte
+const maxBufferSize = 2048 * format.KiloByte
 
 type ImageData struct {
 	Data          []byte `json:"data"`
@@ -642,11 +643,12 @@ type ImageData struct {
 }
 
 type completion struct {
-	Content      string `json:"content"`
-	Model        string `json:"model"`
-	Prompt       string `json:"prompt"`
-	Stop         bool   `json:"stop"`
-	StoppedLimit bool   `json:"stopped_limit"`
+	Content      string    `json:"content"`
+	Model        string    `json:"model"`
+	Prompt       string    `json:"prompt"`
+	Stop         bool      `json:"stop"`
+	StoppedLimit bool      `json:"stopped_limit"`
+	Logits       []float32 `json:"logits,omitempty"`
 
 	Timings struct {
 		PredictedN  int     `json:"predicted_n"`
@@ -657,10 +659,11 @@ type completion struct {
 }
 
 type CompletionRequest struct {
-	Prompt  string
-	Format  json.RawMessage
-	Images  []ImageData
-	Options *api.Options
+	Prompt       string
+	Format       json.RawMessage
+	Images       []ImageData
+	Options      *api.Options
+	ReturnLogits bool
 }
 
 type CompletionResponse struct {
@@ -671,6 +674,7 @@ type CompletionResponse struct {
 	PromptEvalDuration time.Duration
 	EvalCount          int
 	EvalDuration       time.Duration
+	Logits             []float32
 }
 
 func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn func(CompletionResponse)) error {
@@ -696,6 +700,7 @@ func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn fu
 		"seed":              req.Options.Seed,
 		"stop":              req.Options.Stop,
 		"image_data":        req.Images,
+		"return_logits":     req.ReturnLogits,
 		"cache_prompt":      true,
 	}
 
@@ -821,6 +826,7 @@ func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn fu
 			if c.Content != "" {
 				fn(CompletionResponse{
 					Content: c.Content,
+					Logits:  c.Logits,
 				})
 			}
 
@@ -837,6 +843,7 @@ func (s *llmServer) Completion(ctx context.Context, req CompletionRequest, fn fu
 					PromptEvalDuration: parseDurationMs(c.Timings.PromptMS),
 					EvalCount:          c.Timings.PredictedN,
 					EvalDuration:       parseDurationMs(c.Timings.PredictedMS),
+					Logits:             c.Logits,
 				})
 				return nil
 			}

server/routes.go

@@ -19,6 +19,7 @@ import (
 	"os/signal"
 	"path/filepath"
 	"slices"
+	"sort"
 	"strings"
 	"syscall"
 	"time"
@@ -295,10 +296,11 @@ func (s *Server) GenerateHandler(c *gin.Context) {
 		var sb strings.Builder
 		defer close(ch)
 		if err := r.Completion(c.Request.Context(), llm.CompletionRequest{
-			Prompt:  prompt,
-			Images:  images,
-			Format:  req.Format,
-			Options: opts,
+			Prompt:       prompt,
+			Images:       images,
+			Format:       req.Format,
+			Options:      opts,
+			ReturnLogits: false,
 		}, func(cr llm.CompletionResponse) {
 			res := api.GenerateResponse{
 				Model:      req.Model,
@@ -312,6 +314,7 @@ func (s *Server) GenerateHandler(c *gin.Context) {
 					EvalCount:          cr.EvalCount,
 					EvalDuration:       cr.EvalDuration,
 				},
+				Logits: cr.Logits,
 			}
 
 			if _, err := sb.WriteString(cr.Content); err != nil {
@@ -1547,26 +1550,32 @@ func (s *Server) ChatHandler(c *gin.Context) {
 		var sb strings.Builder
 		var toolCallIndex int = 0
 		if err := r.Completion(c.Request.Context(), llm.CompletionRequest{
-			Prompt:  prompt,
-			Images:  images,
-			Format:  req.Format,
-			Options: opts,
-		}, func(r llm.CompletionResponse) {
+			Prompt:       prompt,
+			Images:       images,
+			Format:       req.Format,
+			Options:      opts,
+			ReturnLogits: true,
+		}, func(cr llm.CompletionResponse) {
 			res := api.ChatResponse{
 				Model:      req.Model,
 				CreatedAt:  time.Now().UTC(),
-				Message:    api.Message{Role: "assistant", Content: r.Content},
-				Done:       r.Done,
-				DoneReason: r.DoneReason,
+				Message:    api.Message{Role: "assistant", Content: cr.Content},
+				Done:       cr.Done,
+				DoneReason: cr.DoneReason,
+				Logits:     []float32{},
 				Metrics: api.Metrics{
-					PromptEvalCount:    r.PromptEvalCount,
-					PromptEvalDuration: r.PromptEvalDuration,
-					EvalCount:          r.EvalCount,
-					EvalDuration:       r.EvalDuration,
+					PromptEvalCount:    cr.PromptEvalCount,
+					PromptEvalDuration: cr.PromptEvalDuration,
+					EvalCount:          cr.EvalCount,
+					EvalDuration:       cr.EvalDuration,
 				},
 			}
 
-			if r.Done {
+			topK := int(3)
+			logits := make([]float32, len(cr.Logits))
+			copy(logits, cr.Logits)
+			res.TopLogprobs = getTopKLogProbs(c.Request.Context(), r, logits, topK)
+			if cr.Done {
 				res.TotalDuration = time.Since(checkpointStart)
 				res.LoadDuration = checkpointLoaded.Sub(checkpointStart)
 			}
@@ -1582,7 +1591,7 @@ func (s *Server) ChatHandler(c *gin.Context) {
 			// Streaming tool calls:
 			// If tools are recognized, use a flag to track the sending of a tool downstream
 			// This ensures that content is cleared from the message on the last chunk sent
-			sb.WriteString(r.Content)
+			sb.WriteString(cr.Content)
 			if toolCalls, ok := m.parseToolCalls(sb.String()); ok {
 				res.Message.ToolCalls = toolCalls
 				for i := range toolCalls {
@@ -1595,7 +1604,7 @@ func (s *Server) ChatHandler(c *gin.Context) {
 				return
 			}
 
-			if r.Done {
+			if cr.Done {
 				// Send any remaining content if no tool calls were detected
 				if toolCallIndex == 0 {
 					res.Message.Content = sb.String()
@@ -1645,6 +1654,48 @@ func (s *Server) ChatHandler(c *gin.Context) {
 	streamResponse(c, ch)
 }
 
+func getTopKLogProbs(ctx context.Context, s llm.LlamaServer, logits []float32, topK int) []api.TokenLogprob {
+	// Calculate softmax denominator first (log sum exp trick for numerical stability)
+	maxLogit := float32(math.Inf(-1))
+	for _, logit := range logits {
+		if logit > maxLogit {
+			maxLogit = logit
+		}
+	}
+
+	var sumExp float32
+	for _, logit := range logits {
+		sumExp += float32(math.Exp(float64(logit - maxLogit)))
+	}
+	logSumExp := float32(math.Log(float64(sumExp))) + maxLogit
+
+	// Calculate log probs and track top K
+	logProbs := make([]api.TokenLogprob, len(logits))
+	for i, logit := range logits {
+		text, err := s.Detokenize(ctx, []int{i})
+		if err != nil {
+			slog.Error("detokenize error for logprob", "error", err)
+			continue
+		}
+
+		logProbs[i] = api.TokenLogprob{
+			Text:    text,
+			Logprob: logit - logSumExp,
+		}
+	}
+
+	// Sort by logprob descending and take top K
+	sort.Slice(logProbs, func(i, j int) bool {
+		return logProbs[i].Logprob > logProbs[j].Logprob
+	})
+
+	if len(logProbs) > topK {
+		logProbs = logProbs[:topK]
+	}
+
+	return logProbs
+}
+
 func handleScheduleError(c *gin.Context, name string, err error) {
 	switch {
 	case errors.Is(err, errCapabilities), errors.Is(err, errRequired):