openai.go 26.4 KB
Newer Older
1
2
3
4
5
// openai package provides middleware for partial compatibility with the OpenAI REST API
package openai

import (
	"bytes"
6
	"encoding/base64"
7
	"encoding/json"
Michael Yang's avatar
lint  
Michael Yang committed
8
	"errors"
9
10
	"fmt"
	"io"
royjhan's avatar
royjhan committed
11
	"log/slog"
12
13
	"math/rand"
	"net/http"
14
	"strings"
15
16
17
	"time"

	"github.com/gin-gonic/gin"
Michael Yang's avatar
lint  
Michael Yang committed
18

19
	"github.com/ollama/ollama/api"
20
	"github.com/ollama/ollama/types/model"
21
22
)

23
24
var finishReasonToolCalls = "tool_calls"

25
type Error struct {
26
27
28
29
	Message string  `json:"message"`
	Type    string  `json:"type"`
	Param   any     `json:"param"`
	Code    *string `json:"code"`
30
31
32
33
34
35
36
}

type ErrorResponse struct {
	Error Error `json:"error"`
}

type Message struct {
37
38
39
40
41
42
	Role       string     `json:"role"`
	Content    any        `json:"content"`
	Reasoning  string     `json:"reasoning,omitempty"`
	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`
	Name       string     `json:"name,omitempty"`
	ToolCallID string     `json:"tool_call_id,omitempty"`
43
44
45
46
47
48
49
50
51
52
53
54
55
56
}

type Choice struct {
	Index        int     `json:"index"`
	Message      Message `json:"message"`
	FinishReason *string `json:"finish_reason"`
}

type ChunkChoice struct {
	Index        int     `json:"index"`
	Delta        Message `json:"delta"`
	FinishReason *string `json:"finish_reason"`
}

57
58
59
60
61
62
type CompleteChunkChoice struct {
	Text         string  `json:"text"`
	Index        int     `json:"index"`
	FinishReason *string `json:"finish_reason"`
}

63
64
65
66
67
68
69
type Usage struct {
	PromptTokens     int `json:"prompt_tokens"`
	CompletionTokens int `json:"completion_tokens"`
	TotalTokens      int `json:"total_tokens"`
}

type ResponseFormat struct {
70
71
72
73
74
	Type       string      `json:"type"`
	JsonSchema *JsonSchema `json:"json_schema,omitempty"`
}

type JsonSchema struct {
75
	Schema json.RawMessage `json:"schema"`
76
77
}

78
type EmbedRequest struct {
79
80
81
	Input      any    `json:"input"`
	Model      string `json:"model"`
	Dimensions int    `json:"dimensions,omitempty"`
82
83
}

84
85
86
87
type StreamOptions struct {
	IncludeUsage bool `json:"include_usage"`
}

Michael Yang's avatar
Michael Yang committed
88
89
90
91
type Reasoning struct {
	Effort *string `json:"effort,omitempty"`
}

92
93
94
95
type ChatCompletionRequest struct {
	Model            string          `json:"model"`
	Messages         []Message       `json:"messages"`
	Stream           bool            `json:"stream"`
96
	StreamOptions    *StreamOptions  `json:"stream_options"`
97
98
99
100
101
	MaxTokens        *int            `json:"max_tokens"`
	Seed             *int            `json:"seed"`
	Stop             any             `json:"stop"`
	Temperature      *float64        `json:"temperature"`
	FrequencyPenalty *float64        `json:"frequency_penalty"`
102
	PresencePenalty  *float64        `json:"presence_penalty"`
103
104
	TopP             *float64        `json:"top_p"`
	ResponseFormat   *ResponseFormat `json:"response_format"`
royjhan's avatar
royjhan committed
105
	Tools            []api.Tool      `json:"tools"`
Michael Yang's avatar
Michael Yang committed
106
	Reasoning        *Reasoning      `json:"reasoning,omitempty"`
107
	ReasoningEffort  *string         `json:"reasoning_effort,omitempty"`
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
}

type ChatCompletion struct {
	Id                string   `json:"id"`
	Object            string   `json:"object"`
	Created           int64    `json:"created"`
	Model             string   `json:"model"`
	SystemFingerprint string   `json:"system_fingerprint"`
	Choices           []Choice `json:"choices"`
	Usage             Usage    `json:"usage,omitempty"`
}

type ChatCompletionChunk struct {
	Id                string        `json:"id"`
	Object            string        `json:"object"`
	Created           int64         `json:"created"`
	Model             string        `json:"model"`
	SystemFingerprint string        `json:"system_fingerprint"`
	Choices           []ChunkChoice `json:"choices"`
127
	Usage             *Usage        `json:"usage,omitempty"`
128
129
}

130
131
// TODO (https://github.com/ollama/ollama/issues/5259): support []string, []int and [][]int
type CompletionRequest struct {
132
133
134
135
136
137
138
139
140
141
142
143
	Model            string         `json:"model"`
	Prompt           string         `json:"prompt"`
	FrequencyPenalty float32        `json:"frequency_penalty"`
	MaxTokens        *int           `json:"max_tokens"`
	PresencePenalty  float32        `json:"presence_penalty"`
	Seed             *int           `json:"seed"`
	Stop             any            `json:"stop"`
	Stream           bool           `json:"stream"`
	StreamOptions    *StreamOptions `json:"stream_options"`
	Temperature      *float32       `json:"temperature"`
	TopP             float32        `json:"top_p"`
	Suffix           string         `json:"suffix"`
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
}

type Completion struct {
	Id                string                `json:"id"`
	Object            string                `json:"object"`
	Created           int64                 `json:"created"`
	Model             string                `json:"model"`
	SystemFingerprint string                `json:"system_fingerprint"`
	Choices           []CompleteChunkChoice `json:"choices"`
	Usage             Usage                 `json:"usage,omitempty"`
}

type CompletionChunk struct {
	Id                string                `json:"id"`
	Object            string                `json:"object"`
	Created           int64                 `json:"created"`
	Choices           []CompleteChunkChoice `json:"choices"`
	Model             string                `json:"model"`
	SystemFingerprint string                `json:"system_fingerprint"`
163
	Usage             *Usage                `json:"usage,omitempty"`
164
165
}

royjhan's avatar
royjhan committed
166
167
type ToolCall struct {
	ID       string `json:"id"`
168
	Index    int    `json:"index"`
royjhan's avatar
royjhan committed
169
170
171
172
173
174
175
	Type     string `json:"type"`
	Function struct {
		Name      string `json:"name"`
		Arguments string `json:"arguments"`
	} `json:"function"`
}

176
177
178
179
180
181
182
type Model struct {
	Id      string `json:"id"`
	Object  string `json:"object"`
	Created int64  `json:"created"`
	OwnedBy string `json:"owned_by"`
}

183
184
185
186
187
188
type Embedding struct {
	Object    string    `json:"object"`
	Embedding []float32 `json:"embedding"`
	Index     int       `json:"index"`
}

189
190
191
192
193
type ListCompletion struct {
	Object string  `json:"object"`
	Data   []Model `json:"data"`
}

194
type EmbeddingList struct {
195
196
197
198
199
200
201
202
203
	Object string         `json:"object"`
	Data   []Embedding    `json:"data"`
	Model  string         `json:"model"`
	Usage  EmbeddingUsage `json:"usage,omitempty"`
}

type EmbeddingUsage struct {
	PromptTokens int `json:"prompt_tokens"`
	TotalTokens  int `json:"total_tokens"`
204
205
}

206
207
208
209
210
211
212
213
214
215
216
217
218
219
func NewError(code int, message string) ErrorResponse {
	var etype string
	switch code {
	case http.StatusBadRequest:
		etype = "invalid_request_error"
	case http.StatusNotFound:
		etype = "not_found_error"
	default:
		etype = "api_error"
	}

	return ErrorResponse{Error{Type: etype, Message: message}}
}

220
221
222
223
224
225
226
227
func toUsage(r api.ChatResponse) Usage {
	return Usage{
		PromptTokens:     r.PromptEvalCount,
		CompletionTokens: r.EvalCount,
		TotalTokens:      r.PromptEvalCount + r.EvalCount,
	}
}

royjhan's avatar
royjhan committed
228
229
230
231
232
233
234
235
236
func toolCallId() string {
	const letterBytes = "abcdefghijklmnopqrstuvwxyz0123456789"
	b := make([]byte, 8)
	for i := range b {
		b[i] = letterBytes[rand.Intn(len(letterBytes))]
	}
	return "call_" + strings.ToLower(string(b))
}

237
238
239
func toToolCalls(tc []api.ToolCall) []ToolCall {
	toolCalls := make([]ToolCall, len(tc))
	for i, tc := range tc {
royjhan's avatar
royjhan committed
240
241
242
		toolCalls[i].ID = toolCallId()
		toolCalls[i].Type = "function"
		toolCalls[i].Function.Name = tc.Function.Name
243
		toolCalls[i].Index = tc.Function.Index
royjhan's avatar
royjhan committed
244
245
246
247
248
249
250
251
252

		args, err := json.Marshal(tc.Function.Arguments)
		if err != nil {
			slog.Error("could not marshall function arguments to json", "error", err)
			continue
		}

		toolCalls[i].Function.Arguments = string(args)
	}
253
254
	return toolCalls
}
royjhan's avatar
royjhan committed
255

256
257
func toChatCompletion(id string, r api.ChatResponse) ChatCompletion {
	toolCalls := toToolCalls(r.Message.ToolCalls)
258
259
260
261
262
263
264
	return ChatCompletion{
		Id:                id,
		Object:            "chat.completion",
		Created:           r.CreatedAt.Unix(),
		Model:             r.Model,
		SystemFingerprint: "fp_ollama",
		Choices: []Choice{{
265
			Index:   0,
Michael Yang's avatar
Michael Yang committed
266
			Message: Message{Role: r.Message.Role, Content: r.Message.Content, ToolCalls: toolCalls, Reasoning: r.Message.Thinking},
267
			FinishReason: func(reason string) *string {
268
269
270
				if len(toolCalls) > 0 {
					reason = "tool_calls"
				}
271
272
273
274
275
				if len(reason) > 0 {
					return &reason
				}
				return nil
			}(r.DoneReason),
276
		}},
277
		Usage: toUsage(r),
278
279
280
	}
}

281
func toChunk(id string, r api.ChatResponse, toolCallSent bool) ChatCompletionChunk {
282
	toolCalls := toToolCalls(r.Message.ToolCalls)
283
284
285
286
287
288
	return ChatCompletionChunk{
		Id:                id,
		Object:            "chat.completion.chunk",
		Created:           time.Now().Unix(),
		Model:             r.Model,
		SystemFingerprint: "fp_ollama",
289
290
		Choices: []ChunkChoice{{
			Index: 0,
Michael Yang's avatar
Michael Yang committed
291
			Delta: Message{Role: "assistant", Content: r.Message.Content, ToolCalls: toolCalls, Reasoning: r.Message.Thinking},
292
293
			FinishReason: func(reason string) *string {
				if len(reason) > 0 {
Michael Yang's avatar
Michael Yang committed
294
					if toolCallSent || len(toolCalls) > 0 {
295
296
						return &finishReasonToolCalls
					}
297
298
299
300
301
					return &reason
				}
				return nil
			}(r.DoneReason),
		}},
302
303
304
	}
}

305
306
307
308
309
310
311
312
func toUsageGenerate(r api.GenerateResponse) Usage {
	return Usage{
		PromptTokens:     r.PromptEvalCount,
		CompletionTokens: r.EvalCount,
		TotalTokens:      r.PromptEvalCount + r.EvalCount,
	}
}

313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
func toCompletion(id string, r api.GenerateResponse) Completion {
	return Completion{
		Id:                id,
		Object:            "text_completion",
		Created:           r.CreatedAt.Unix(),
		Model:             r.Model,
		SystemFingerprint: "fp_ollama",
		Choices: []CompleteChunkChoice{{
			Text:  r.Response,
			Index: 0,
			FinishReason: func(reason string) *string {
				if len(reason) > 0 {
					return &reason
				}
				return nil
			}(r.DoneReason),
		}},
330
		Usage: toUsageGenerate(r),
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
	}
}

func toCompleteChunk(id string, r api.GenerateResponse) CompletionChunk {
	return CompletionChunk{
		Id:                id,
		Object:            "text_completion",
		Created:           time.Now().Unix(),
		Model:             r.Model,
		SystemFingerprint: "fp_ollama",
		Choices: []CompleteChunkChoice{{
			Text:  r.Response,
			Index: 0,
			FinishReason: func(reason string) *string {
				if len(reason) > 0 {
					return &reason
				}
				return nil
			}(r.DoneReason),
		}},
	}
}

354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
func toListCompletion(r api.ListResponse) ListCompletion {
	var data []Model
	for _, m := range r.Models {
		data = append(data, Model{
			Id:      m.Name,
			Object:  "model",
			Created: m.ModifiedAt.Unix(),
			OwnedBy: model.ParseName(m.Name).Namespace,
		})
	}

	return ListCompletion{
		Object: "list",
		Data:   data,
	}
}

371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
func toEmbeddingList(model string, r api.EmbedResponse) EmbeddingList {
	if r.Embeddings != nil {
		var data []Embedding
		for i, e := range r.Embeddings {
			data = append(data, Embedding{
				Object:    "embedding",
				Embedding: e,
				Index:     i,
			})
		}

		return EmbeddingList{
			Object: "list",
			Data:   data,
			Model:  model,
386
387
388
389
			Usage: EmbeddingUsage{
				PromptTokens: r.PromptEvalCount,
				TotalTokens:  r.PromptEvalCount,
			},
390
391
392
393
394
395
		}
	}

	return EmbeddingList{}
}

396
397
398
399
400
401
402
403
404
func toModel(r api.ShowResponse, m string) Model {
	return Model{
		Id:      m,
		Object:  "model",
		Created: r.ModifiedAt.Unix(),
		OwnedBy: model.ParseName(m).Namespace,
	}
}

405
func fromChatRequest(r ChatCompletionRequest) (*api.ChatRequest, error) {
406
407
	var messages []api.Message
	for _, msg := range r.Messages {
408
409
410
411
412
413
414
		toolName := ""
		if strings.ToLower(msg.Role) == "tool" {
			toolName = msg.Name
			if toolName == "" && msg.ToolCallID != "" {
				toolName = nameFromToolCallID(r.Messages, msg.ToolCallID)
			}
		}
415
416
		switch content := msg.Content.(type) {
		case string:
417
418
419
420
			toolCalls, err := fromCompletionToolCall(msg.ToolCalls)
			if err != nil {
				return nil, err
			}
421
			messages = append(messages, api.Message{Role: msg.Role, Content: content, Thinking: msg.Reasoning, ToolCalls: toolCalls, ToolName: toolName})
422
423
424
425
		case []any:
			for _, c := range content {
				data, ok := c.(map[string]any)
				if !ok {
Michael Yang's avatar
lint  
Michael Yang committed
426
					return nil, errors.New("invalid message format")
427
428
429
430
431
				}
				switch data["type"] {
				case "text":
					text, ok := data["text"].(string)
					if !ok {
Michael Yang's avatar
lint  
Michael Yang committed
432
						return nil, errors.New("invalid message format")
433
					}
434
					messages = append(messages, api.Message{Role: msg.Role, Content: text})
435
436
437
438
				case "image_url":
					var url string
					if urlMap, ok := data["image_url"].(map[string]any); ok {
						if url, ok = urlMap["url"].(string); !ok {
Michael Yang's avatar
lint  
Michael Yang committed
439
							return nil, errors.New("invalid message format")
440
441
442
						}
					} else {
						if url, ok = data["image_url"].(string); !ok {
Michael Yang's avatar
lint  
Michael Yang committed
443
							return nil, errors.New("invalid message format")
444
445
446
						}
					}

447
					types := []string{"jpeg", "jpg", "png", "webp"}
448
449
450
451
452
453
454
455
456
457
458
					valid := false
					for _, t := range types {
						prefix := "data:image/" + t + ";base64,"
						if strings.HasPrefix(url, prefix) {
							url = strings.TrimPrefix(url, prefix)
							valid = true
							break
						}
					}

					if !valid {
Michael Yang's avatar
lint  
Michael Yang committed
459
						return nil, errors.New("invalid image input")
460
461
462
463
					}

					img, err := base64.StdEncoding.DecodeString(url)
					if err != nil {
Michael Yang's avatar
lint  
Michael Yang committed
464
						return nil, errors.New("invalid message format")
465
					}
466
467

					messages = append(messages, api.Message{Role: msg.Role, Images: []api.ImageData{img}})
468
				default:
Michael Yang's avatar
lint  
Michael Yang committed
469
					return nil, errors.New("invalid message format")
470
471
				}
			}
472
473
474
475
476
477
478
479
			// since we might have added multiple messages above, if we have tools
			// calls we'll add them to the last message
			if len(messages) > 0 && len(msg.ToolCalls) > 0 {
				toolCalls, err := fromCompletionToolCall(msg.ToolCalls)
				if err != nil {
					return nil, err
				}
				messages[len(messages)-1].ToolCalls = toolCalls
480
481
482
				if toolName != "" {
					messages[len(messages)-1].ToolName = toolName
				}
483
				messages[len(messages)-1].Thinking = msg.Reasoning
484
			}
485
		default:
486
			// content is only optional if tool calls are present
royjhan's avatar
royjhan committed
487
488
489
490
491
492
493
494
495
			if msg.ToolCalls == nil {
				return nil, fmt.Errorf("invalid message content type: %T", content)
			}

			toolCalls := make([]api.ToolCall, len(msg.ToolCalls))
			for i, tc := range msg.ToolCalls {
				toolCalls[i].Function.Name = tc.Function.Name
				err := json.Unmarshal([]byte(tc.Function.Arguments), &toolCalls[i].Function.Arguments)
				if err != nil {
Michael Yang's avatar
lint  
Michael Yang committed
496
					return nil, errors.New("invalid tool call arguments")
royjhan's avatar
royjhan committed
497
498
				}
			}
499
			messages = append(messages, api.Message{Role: msg.Role, Thinking: msg.Reasoning, ToolCalls: toolCalls})
500
		}
501
502
	}

503
	options := make(map[string]any)
504
505
506
507

	switch stop := r.Stop.(type) {
	case string:
		options["stop"] = []string{stop}
508
	case []any:
509
510
511
512
513
514
515
516
517
518
519
520
521
522
		var stops []string
		for _, s := range stop {
			if str, ok := s.(string); ok {
				stops = append(stops, str)
			}
		}
		options["stop"] = stops
	}

	if r.MaxTokens != nil {
		options["num_predict"] = *r.MaxTokens
	}

	if r.Temperature != nil {
523
		options["temperature"] = *r.Temperature
524
525
526
527
528
529
530
531
532
	} else {
		options["temperature"] = 1.0
	}

	if r.Seed != nil {
		options["seed"] = *r.Seed
	}

	if r.FrequencyPenalty != nil {
533
		options["frequency_penalty"] = *r.FrequencyPenalty
534
535
536
	}

	if r.PresencePenalty != nil {
537
		options["presence_penalty"] = *r.PresencePenalty
538
539
540
541
542
543
544
545
	}

	if r.TopP != nil {
		options["top_p"] = *r.TopP
	} else {
		options["top_p"] = 1.0
	}

546
547
548
549
550
551
552
553
	var format json.RawMessage
	if r.ResponseFormat != nil {
		switch strings.ToLower(strings.TrimSpace(r.ResponseFormat.Type)) {
		// Support the old "json_object" type for OpenAI compatibility
		case "json_object":
			format = json.RawMessage(`"json"`)
		case "json_schema":
			if r.ResponseFormat.JsonSchema != nil {
554
				format = r.ResponseFormat.JsonSchema.Schema
555
556
			}
		}
557
558
	}

Michael Yang's avatar
Michael Yang committed
559
560
561
562
563
	var think *api.ThinkValue
	if r.Reasoning != nil {
		think = &api.ThinkValue{
			Value: *r.Reasoning.Effort,
		}
564
565
566
567
	} else if r.ReasoningEffort != nil {
		think = &api.ThinkValue{
			Value: *r.ReasoningEffort,
		}
Michael Yang's avatar
Michael Yang committed
568
569
	}

570
	return &api.ChatRequest{
571
572
573
574
575
		Model:    r.Model,
		Messages: messages,
		Format:   format,
		Options:  options,
		Stream:   &r.Stream,
royjhan's avatar
royjhan committed
576
		Tools:    r.Tools,
Michael Yang's avatar
Michael Yang committed
577
		Think:    think,
578
	}, nil
579
580
}

581
582
583
584
585
586
587
588
589
590
591
592
593
594
func nameFromToolCallID(messages []Message, toolCallID string) string {
	// iterate backwards to be more resilient to duplicate tool call IDs (this
	// follows "last one wins")
	for i := len(messages) - 1; i >= 0; i-- {
		msg := messages[i]
		for _, tc := range msg.ToolCalls {
			if tc.ID == toolCallID {
				return tc.Function.Name
			}
		}
	}
	return ""
}

595
596
597
598
599
600
601
602
603
604
605
606
607
func fromCompletionToolCall(toolCalls []ToolCall) ([]api.ToolCall, error) {
	apiToolCalls := make([]api.ToolCall, len(toolCalls))
	for i, tc := range toolCalls {
		apiToolCalls[i].Function.Name = tc.Function.Name
		err := json.Unmarshal([]byte(tc.Function.Arguments), &apiToolCalls[i].Function.Arguments)
		if err != nil {
			return nil, errors.New("invalid tool call arguments")
		}
	}

	return apiToolCalls, nil
}

608
609
610
611
612
613
func fromCompleteRequest(r CompletionRequest) (api.GenerateRequest, error) {
	options := make(map[string]any)

	switch stop := r.Stop.(type) {
	case string:
		options["stop"] = []string{stop}
614
615
616
617
618
619
620
621
	case []any:
		var stops []string
		for _, s := range stop {
			if str, ok := s.(string); ok {
				stops = append(stops, str)
			} else {
				return api.GenerateRequest{}, fmt.Errorf("invalid type for 'stop' field: %T", s)
			}
622
		}
623
		options["stop"] = stops
624
625
626
627
628
629
630
	}

	if r.MaxTokens != nil {
		options["num_predict"] = *r.MaxTokens
	}

	if r.Temperature != nil {
631
		options["temperature"] = *r.Temperature
632
633
634
635
636
637
638
639
	} else {
		options["temperature"] = 1.0
	}

	if r.Seed != nil {
		options["seed"] = *r.Seed
	}

640
	options["frequency_penalty"] = r.FrequencyPenalty
641

642
	options["presence_penalty"] = r.PresencePenalty
643
644
645
646
647
648
649
650
651
652
653
654

	if r.TopP != 0.0 {
		options["top_p"] = r.TopP
	} else {
		options["top_p"] = 1.0
	}

	return api.GenerateRequest{
		Model:   r.Model,
		Prompt:  r.Prompt,
		Options: options,
		Stream:  &r.Stream,
655
		Suffix:  r.Suffix,
656
657
658
	}, nil
}

659
660
661
662
663
type BaseWriter struct {
	gin.ResponseWriter
}

type ChatWriter struct {
664
665
666
	stream        bool
	streamOptions *StreamOptions
	id            string
667
	toolCallSent  bool
668
	BaseWriter
669
670
}

671
type CompleteWriter struct {
672
673
674
	stream        bool
	streamOptions *StreamOptions
	id            string
675
676
677
	BaseWriter
}

678
679
680
681
682
683
684
685
686
type ListWriter struct {
	BaseWriter
}

type RetrieveWriter struct {
	BaseWriter
	model string
}

687
688
689
690
691
type EmbedWriter struct {
	BaseWriter
	model string
}

692
func (w *BaseWriter) writeError(data []byte) (int, error) {
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
	var serr api.StatusError
	err := json.Unmarshal(data, &serr)
	if err != nil {
		return 0, err
	}

	w.ResponseWriter.Header().Set("Content-Type", "application/json")
	err = json.NewEncoder(w.ResponseWriter).Encode(NewError(http.StatusInternalServerError, serr.Error()))
	if err != nil {
		return 0, err
	}

	return len(data), nil
}

708
func (w *ChatWriter) writeResponse(data []byte) (int, error) {
709
710
711
712
713
714
715
716
	var chatResponse api.ChatResponse
	err := json.Unmarshal(data, &chatResponse)
	if err != nil {
		return 0, err
	}

	// chat chunk
	if w.stream {
717
		c := toChunk(w.id, chatResponse, w.toolCallSent)
718
		d, err := json.Marshal(c)
719
720
721
		if err != nil {
			return 0, err
		}
722
723
724
		if !w.toolCallSent && len(c.Choices) > 0 && len(c.Choices[0].Delta.ToolCalls) > 0 {
			w.toolCallSent = true
		}
725
726
727
728
729
730
731
732

		w.ResponseWriter.Header().Set("Content-Type", "text/event-stream")
		_, err = w.ResponseWriter.Write([]byte(fmt.Sprintf("data: %s\n\n", d)))
		if err != nil {
			return 0, err
		}

		if chatResponse.Done {
733
734
735
736
737
738
739
740
741
742
743
744
745
			if w.streamOptions != nil && w.streamOptions.IncludeUsage {
				u := toUsage(chatResponse)
				c.Usage = &u
				c.Choices = []ChunkChoice{}
				d, err := json.Marshal(c)
				if err != nil {
					return 0, err
				}
				_, err = w.ResponseWriter.Write([]byte(fmt.Sprintf("data: %s\n\n", d)))
				if err != nil {
					return 0, err
				}
			}
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
			_, err = w.ResponseWriter.Write([]byte("data: [DONE]\n\n"))
			if err != nil {
				return 0, err
			}
		}

		return len(data), nil
	}

	// chat completion
	w.ResponseWriter.Header().Set("Content-Type", "application/json")
	err = json.NewEncoder(w.ResponseWriter).Encode(toChatCompletion(w.id, chatResponse))
	if err != nil {
		return 0, err
	}

	return len(data), nil
}

765
func (w *ChatWriter) Write(data []byte) (int, error) {
766
767
	code := w.ResponseWriter.Status()
	if code != http.StatusOK {
768
		return w.writeError(data)
769
770
771
772
773
	}

	return w.writeResponse(data)
}

774
775
776
777
778
779
780
781
782
func (w *CompleteWriter) writeResponse(data []byte) (int, error) {
	var generateResponse api.GenerateResponse
	err := json.Unmarshal(data, &generateResponse)
	if err != nil {
		return 0, err
	}

	// completion chunk
	if w.stream {
783
784
785
786
787
		c := toCompleteChunk(w.id, generateResponse)
		if w.streamOptions != nil && w.streamOptions.IncludeUsage {
			c.Usage = &Usage{}
		}
		d, err := json.Marshal(c)
788
789
790
791
792
793
794
795
796
797
798
		if err != nil {
			return 0, err
		}

		w.ResponseWriter.Header().Set("Content-Type", "text/event-stream")
		_, err = w.ResponseWriter.Write([]byte(fmt.Sprintf("data: %s\n\n", d)))
		if err != nil {
			return 0, err
		}

		if generateResponse.Done {
799
800
801
802
803
804
805
806
807
808
809
810
811
			if w.streamOptions != nil && w.streamOptions.IncludeUsage {
				u := toUsageGenerate(generateResponse)
				c.Usage = &u
				c.Choices = []CompleteChunkChoice{}
				d, err := json.Marshal(c)
				if err != nil {
					return 0, err
				}
				_, err = w.ResponseWriter.Write([]byte(fmt.Sprintf("data: %s\n\n", d)))
				if err != nil {
					return 0, err
				}
			}
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
			_, err = w.ResponseWriter.Write([]byte("data: [DONE]\n\n"))
			if err != nil {
				return 0, err
			}
		}

		return len(data), nil
	}

	// completion
	w.ResponseWriter.Header().Set("Content-Type", "application/json")
	err = json.NewEncoder(w.ResponseWriter).Encode(toCompletion(w.id, generateResponse))
	if err != nil {
		return 0, err
	}

	return len(data), nil
}

func (w *CompleteWriter) Write(data []byte) (int, error) {
	code := w.ResponseWriter.Status()
	if code != http.StatusOK {
834
		return w.writeError(data)
835
836
837
838
839
	}

	return w.writeResponse(data)
}

840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
func (w *ListWriter) writeResponse(data []byte) (int, error) {
	var listResponse api.ListResponse
	err := json.Unmarshal(data, &listResponse)
	if err != nil {
		return 0, err
	}

	w.ResponseWriter.Header().Set("Content-Type", "application/json")
	err = json.NewEncoder(w.ResponseWriter).Encode(toListCompletion(listResponse))
	if err != nil {
		return 0, err
	}

	return len(data), nil
}

func (w *ListWriter) Write(data []byte) (int, error) {
	code := w.ResponseWriter.Status()
	if code != http.StatusOK {
859
		return w.writeError(data)
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
	}

	return w.writeResponse(data)
}

func (w *RetrieveWriter) writeResponse(data []byte) (int, error) {
	var showResponse api.ShowResponse
	err := json.Unmarshal(data, &showResponse)
	if err != nil {
		return 0, err
	}

	// retrieve completion
	w.ResponseWriter.Header().Set("Content-Type", "application/json")
	err = json.NewEncoder(w.ResponseWriter).Encode(toModel(showResponse, w.model))
	if err != nil {
		return 0, err
	}

	return len(data), nil
}

func (w *RetrieveWriter) Write(data []byte) (int, error) {
	code := w.ResponseWriter.Status()
	if code != http.StatusOK {
885
		return w.writeError(data)
886
887
888
889
890
	}

	return w.writeResponse(data)
}

891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
func (w *EmbedWriter) writeResponse(data []byte) (int, error) {
	var embedResponse api.EmbedResponse
	err := json.Unmarshal(data, &embedResponse)
	if err != nil {
		return 0, err
	}

	w.ResponseWriter.Header().Set("Content-Type", "application/json")
	err = json.NewEncoder(w.ResponseWriter).Encode(toEmbeddingList(w.model, embedResponse))
	if err != nil {
		return 0, err
	}

	return len(data), nil
}

func (w *EmbedWriter) Write(data []byte) (int, error) {
	code := w.ResponseWriter.Status()
	if code != http.StatusOK {
910
		return w.writeError(data)
911
912
913
914
915
	}

	return w.writeResponse(data)
}

916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
func ListMiddleware() gin.HandlerFunc {
	return func(c *gin.Context) {
		w := &ListWriter{
			BaseWriter: BaseWriter{ResponseWriter: c.Writer},
		}

		c.Writer = w

		c.Next()
	}
}

func RetrieveMiddleware() gin.HandlerFunc {
	return func(c *gin.Context) {
		var b bytes.Buffer
		if err := json.NewEncoder(&b).Encode(api.ShowRequest{Name: c.Param("model")}); err != nil {
			c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
			return
		}

		c.Request.Body = io.NopCloser(&b)

		// response writer
		w := &RetrieveWriter{
			BaseWriter: BaseWriter{ResponseWriter: c.Writer},
			model:      c.Param("model"),
		}

		c.Writer = w

		c.Next()
	}
}

950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
func CompletionsMiddleware() gin.HandlerFunc {
	return func(c *gin.Context) {
		var req CompletionRequest
		err := c.ShouldBindJSON(&req)
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, err.Error()))
			return
		}

		var b bytes.Buffer
		genReq, err := fromCompleteRequest(req)
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, err.Error()))
			return
		}

		if err := json.NewEncoder(&b).Encode(genReq); err != nil {
			c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
			return
		}

		c.Request.Body = io.NopCloser(&b)

		w := &CompleteWriter{
974
975
976
977
			BaseWriter:    BaseWriter{ResponseWriter: c.Writer},
			stream:        req.Stream,
			id:            fmt.Sprintf("cmpl-%d", rand.Intn(999)),
			streamOptions: req.StreamOptions,
978
979
		}

980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
		c.Writer = w
		c.Next()
	}
}

func EmbeddingsMiddleware() gin.HandlerFunc {
	return func(c *gin.Context) {
		var req EmbedRequest
		err := c.ShouldBindJSON(&req)
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, err.Error()))
			return
		}

		if req.Input == "" {
			req.Input = []string{""}
		}

		if req.Input == nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, "invalid input"))
			return
		}

		if v, ok := req.Input.([]any); ok && len(v) == 0 {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, "invalid input"))
			return
		}

		var b bytes.Buffer
1009
		if err := json.NewEncoder(&b).Encode(api.EmbedRequest{Model: req.Model, Input: req.Input, Dimensions: req.Dimensions}); err != nil {
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
			c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
			return
		}

		c.Request.Body = io.NopCloser(&b)

		w := &EmbedWriter{
			BaseWriter: BaseWriter{ResponseWriter: c.Writer},
			model:      req.Model,
		}

1021
1022
1023
1024
1025
1026
		c.Writer = w

		c.Next()
	}
}

1027
func ChatMiddleware() gin.HandlerFunc {
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
	return func(c *gin.Context) {
		var req ChatCompletionRequest
		err := c.ShouldBindJSON(&req)
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, err.Error()))
			return
		}

		if len(req.Messages) == 0 {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, "[] is too short - 'messages'"))
			return
		}

		var b bytes.Buffer
1042
1043
1044
1045

		chatReq, err := fromChatRequest(req)
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, NewError(http.StatusBadRequest, err.Error()))
1046
			return
1047
1048
1049
		}

		if err := json.NewEncoder(&b).Encode(chatReq); err != nil {
1050
1051
1052
1053
1054
1055
			c.AbortWithStatusJSON(http.StatusInternalServerError, NewError(http.StatusInternalServerError, err.Error()))
			return
		}

		c.Request.Body = io.NopCloser(&b)

1056
		w := &ChatWriter{
1057
1058
1059
1060
			BaseWriter:    BaseWriter{ResponseWriter: c.Writer},
			stream:        req.Stream,
			id:            fmt.Sprintf("chatcmpl-%d", rand.Intn(999)),
			streamOptions: req.StreamOptions,
1061
1062
1063
1064
1065
1066
1067
		}

		c.Writer = w

		c.Next()
	}
}