routes.go 31 KB
Newer Older
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1
2
3
package server

import (
4
	"context"
Michael Yang's avatar
Michael Yang committed
5
	"encoding/json"
6
	"errors"
7
	"fmt"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
8
	"io"
9
	"io/fs"
10
	"log/slog"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
11
12
	"net"
	"net/http"
13
	"os"
14
	"os/signal"
Michael Yang's avatar
Michael Yang committed
15
	"path/filepath"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
16
	"reflect"
17
	"runtime"
Michael Yang's avatar
Michael Yang committed
18
	"strings"
Michael Yang's avatar
Michael Yang committed
19
	"sync"
20
	"syscall"
21
	"time"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
22

Michael Yang's avatar
Michael Yang committed
23
	"github.com/gin-contrib/cors"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
24
	"github.com/gin-gonic/gin"
25
	"golang.org/x/exp/slices"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
26

Jeffrey Morgan's avatar
Jeffrey Morgan committed
27
	"github.com/jmorganca/ollama/api"
28
	"github.com/jmorganca/ollama/gpu"
29
	"github.com/jmorganca/ollama/llm"
30
	"github.com/jmorganca/ollama/openai"
Michael Yang's avatar
Michael Yang committed
31
	"github.com/jmorganca/ollama/parser"
Michael Yang's avatar
Michael Yang committed
32
	"github.com/jmorganca/ollama/version"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
33
34
)

Michael Yang's avatar
Michael Yang committed
35
36
var mode string = gin.DebugMode

37
38
39
40
type Server struct {
	WorkDir string
}

Michael Yang's avatar
Michael Yang committed
41
42
43
44
45
46
47
48
49
50
51
52
func init() {
	switch mode {
	case gin.DebugMode:
	case gin.ReleaseMode:
	case gin.TestMode:
	default:
		mode = gin.DebugMode
	}

	gin.SetMode(mode)
}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
53
var loaded struct {
Michael Yang's avatar
Michael Yang committed
54
55
	mu sync.Mutex

56
	runner llm.LLM
Michael Yang's avatar
Michael Yang committed
57
58
59

	expireAt    time.Time
	expireTimer *time.Timer
Jeffrey Morgan's avatar
Jeffrey Morgan committed
60

61
62
	*Model
	*api.Options
Michael Yang's avatar
Michael Yang committed
63
64
}

65
66
var defaultSessionDuration = 5 * time.Minute

Bruce MacDonald's avatar
Bruce MacDonald committed
67
// load a model into memory if it is not already loaded, it is up to the caller to lock loaded.mu before calling this function
68
func load(c *gin.Context, model *Model, opts api.Options, sessionDuration time.Duration) error {
Bruce MacDonald's avatar
Bruce MacDonald committed
69
70
	workDir := c.GetString("workDir")

71
72
73
74
75
76
77
	needLoad := loaded.runner == nil || // is there a model loaded?
		loaded.ModelPath != model.ModelPath || // has the base model changed?
		!reflect.DeepEqual(loaded.AdapterPaths, model.AdapterPaths) || // have the adapters changed?
		!reflect.DeepEqual(loaded.Options.Runner, opts.Runner) // have the runner options changed?

	if needLoad {
		if loaded.runner != nil {
78
			slog.Info("changing loaded model")
79
80
81
82
			loaded.runner.Close()
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
Michael Yang's avatar
Michael Yang committed
83
		}
Michael Yang's avatar
Michael Yang committed
84

Michael Yang's avatar
Michael Yang committed
85
		llmRunner, err := llm.New(workDir, model.ModelPath, model.AdapterPaths, model.ProjectorPaths, opts)
Michael Yang's avatar
Michael Yang committed
86
		if err != nil {
87
88
89
			// some older models are not compatible with newer versions of llama.cpp
			// show a generalized compatibility error until there is a better way to
			// check for model compatibility
Bruce MacDonald's avatar
Bruce MacDonald committed
90
			if errors.Is(llm.ErrUnsupportedFormat, err) || strings.Contains(err.Error(), "failed to load model") {
91
92
93
				err = fmt.Errorf("%v: this model may be incompatible with your version of Ollama. If you previously pulled this model, try updating it by running `ollama pull %s`", err, model.ShortName)
			}

94
			return err
Michael Yang's avatar
Michael Yang committed
95
96
		}

97
98
99
		loaded.Model = model
		loaded.runner = llmRunner
		loaded.Options = &opts
Michael Yang's avatar
Michael Yang committed
100
	}
101

Jeffrey Morgan's avatar
Jeffrey Morgan committed
102
	loaded.expireAt = time.Now().Add(sessionDuration)
Bruce MacDonald's avatar
Bruce MacDonald committed
103

Jeffrey Morgan's avatar
Jeffrey Morgan committed
104
105
106
107
	if loaded.expireTimer == nil {
		loaded.expireTimer = time.AfterFunc(sessionDuration, func() {
			loaded.mu.Lock()
			defer loaded.mu.Unlock()
Michael Yang's avatar
Michael Yang committed
108

Jeffrey Morgan's avatar
Jeffrey Morgan committed
109
			if time.Now().Before(loaded.expireAt) {
Michael Yang's avatar
Michael Yang committed
110
111
112
				return
			}

113
114
			if loaded.runner != nil {
				loaded.runner.Close()
Michael Yang's avatar
Michael Yang committed
115
116
			}

117
118
119
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
Michael Yang's avatar
Michael Yang committed
120
		})
Michael Yang's avatar
Michael Yang committed
121
	}
122

Jeffrey Morgan's avatar
Jeffrey Morgan committed
123
	loaded.expireTimer.Reset(sessionDuration)
124
125
126
127
128
129
130
131
132
133
134
135
136
137
	return nil
}

func modelOptions(model *Model, requestOpts map[string]interface{}) (api.Options, error) {
	opts := api.DefaultOptions()
	if err := opts.FromMap(model.Options); err != nil {
		return api.Options{}, err
	}

	if err := opts.FromMap(requestOpts); err != nil {
		return api.Options{}, err
	}

	return opts, nil
Bruce MacDonald's avatar
Bruce MacDonald committed
138
139
}

140
141
142
143
144
145
func isSupportedImageType(image []byte) bool {
	contentType := http.DetectContentType(image)
	allowedTypes := []string{"image/jpeg", "image/jpg", "image/png"}
	return slices.Contains(allowedTypes, contentType)
}

Bruce MacDonald's avatar
Bruce MacDonald committed
146
147
148
149
150
151
func GenerateHandler(c *gin.Context) {
	loaded.mu.Lock()
	defer loaded.mu.Unlock()

	checkpointStart := time.Now()
	var req api.GenerateRequest
Michael Yang's avatar
Michael Yang committed
152
	err := c.ShouldBindJSON(&req)
Patrick Devine's avatar
Patrick Devine committed
153

Michael Yang's avatar
Michael Yang committed
154
155
156
157
158
159
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
Bruce MacDonald's avatar
Bruce MacDonald committed
160
161
162
		return
	}

163
164
165
	// validate the request
	switch {
	case req.Model == "":
166
167
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
		return
168
169
170
	case len(req.Format) > 0 && req.Format != "json":
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "format must be json"})
		return
171
172
173
	case req.Raw && (req.Template != "" || req.System != "" || len(req.Context) > 0):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "raw mode does not support template, system, or context"})
		return
174
175
	}

176
177
178
179
180
181
182
	for _, img := range req.Images {
		if !isSupportedImageType(img) {
			c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "unsupported image format"})
			return
		}
	}

183
	model, err := GetModel(req.Model)
Bruce MacDonald's avatar
Bruce MacDonald committed
184
	if err != nil {
185
		var pErr *fs.PathError
186
		if errors.As(err, &pErr) {
187
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found, try pulling it first", req.Model)})
188
189
190
191
192
193
194
195
196
			return
		}
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	opts, err := modelOptions(model, req.Options)
	if err != nil {
		if errors.Is(err, api.ErrInvalidOpts) {
Bruce MacDonald's avatar
Bruce MacDonald committed
197
			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
198
			return
199
		}
200
201
202
203
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

204
205
206
207
208
209
210
	var sessionDuration time.Duration
	if req.KeepAlive == nil {
		sessionDuration = defaultSessionDuration
	} else {
		sessionDuration = req.KeepAlive.Duration
	}

211
212
	if err := load(c, model, opts, sessionDuration); err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
Bruce MacDonald's avatar
Bruce MacDonald committed
213
214
215
		return
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
216
	// an empty request loads the model
217
218
	// note: for a short while template was used in lieu
	// of `raw` mode so we need to check for it too
Bruce MacDonald's avatar
Bruce MacDonald committed
219
	if req.Prompt == "" && req.Template == "" && req.System == "" {
220
		c.JSON(http.StatusOK, api.GenerateResponse{
221
222
			CreatedAt: time.Now().UTC(),
			Model:     req.Model,
Michael Yang's avatar
Michael Yang committed
223
224
			Done:      true,
		})
Bruce MacDonald's avatar
Bruce MacDonald committed
225
226
227
228
229
		return
	}

	checkpointLoaded := time.Now()

Bruce MacDonald's avatar
Bruce MacDonald committed
230
231
232
233
234
	var prompt string
	switch {
	case req.Raw:
		prompt = req.Prompt
	case req.Prompt != "":
235
236
		if req.Template == "" {
			req.Template = model.Template
Bruce MacDonald's avatar
Bruce MacDonald committed
237
238
		}

239
240
241
242
243
244
245
246
247
		if req.System == "" {
			req.System = model.System
		}

		slog.Debug("generate handler", "prompt", req.Prompt)
		slog.Debug("generate handler", "template", req.Template)
		slog.Debug("generate handler", "system", req.System)

		var sb strings.Builder
Bruce MacDonald's avatar
Bruce MacDonald committed
248
		if req.Context != nil {
249
			prev, err := loaded.runner.Decode(c.Request.Context(), req.Context)
Bruce MacDonald's avatar
Bruce MacDonald committed
250
251
252
253
254
			if err != nil {
				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
				return
			}

255
			sb.WriteString(prev)
256
257
		}

258
259
		// write image tags
		// TODO: limit the number of images to fit in the context similar to the chat endpoint
260
		for i := range req.Images {
261
			req.Prompt += fmt.Sprintf(" [img-%d]", i)
262
263
		}

264
		p, err := Prompt(req.Template, req.System, req.Prompt, "", true)
265
266
267
268
		if err != nil {
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
			return
		}
269
270
271
272

		sb.WriteString(p)

		prompt = sb.String()
Bruce MacDonald's avatar
Bruce MacDonald committed
273
274
	}

Michael Yang's avatar
Michael Yang committed
275
	slog.Debug("generate handler", "prompt", prompt)
276

Bruce MacDonald's avatar
Bruce MacDonald committed
277
	ch := make(chan any)
Bruce MacDonald's avatar
Bruce MacDonald committed
278
	var generated strings.Builder
Bruce MacDonald's avatar
Bruce MacDonald committed
279
280
281
	go func() {
		defer close(ch)

Bruce MacDonald's avatar
Bruce MacDonald committed
282
283
		fn := func(r llm.PredictResult) {
			// Update model expiration
Bruce MacDonald's avatar
Bruce MacDonald committed
284
285
286
			loaded.expireAt = time.Now().Add(sessionDuration)
			loaded.expireTimer.Reset(sessionDuration)

Bruce MacDonald's avatar
Bruce MacDonald committed
287
288
289
290
			// Build up the full response
			if _, err := generated.WriteString(r.Content); err != nil {
				ch <- gin.H{"error": err.Error()}
				return
Bruce MacDonald's avatar
Bruce MacDonald committed
291
292
			}

Bruce MacDonald's avatar
Bruce MacDonald committed
293
			resp := api.GenerateResponse{
294
				Model:     req.Model,
295
				CreatedAt: time.Now().UTC(),
296
297
				Done:      r.Done,
				Response:  r.Content,
Bruce MacDonald's avatar
Bruce MacDonald committed
298
299
300
301
302
303
				Metrics: api.Metrics{
					PromptEvalCount:    r.PromptEvalCount,
					PromptEvalDuration: r.PromptEvalDuration,
					EvalCount:          r.EvalCount,
					EvalDuration:       r.EvalDuration,
				},
Bruce MacDonald's avatar
Bruce MacDonald committed
304
305
			}

306
307
308
309
310
			if r.Done {
				resp.TotalDuration = time.Since(checkpointStart)
				resp.LoadDuration = checkpointLoaded.Sub(checkpointStart)

				if !req.Raw {
311
					p, err := Prompt(req.Template, req.System, req.Prompt, generated.String(), false)
312
					if err != nil {
313
						c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
314
315
						return
					}
316
317
318

					// TODO (jmorganca): encode() should not strip special tokens
					tokens, err := loaded.runner.Encode(c.Request.Context(), p)
319
320
321
322
					if err != nil {
						ch <- gin.H{"error": err.Error()}
						return
					}
323
324

					resp.Context = append(req.Context, tokens...)
Bruce MacDonald's avatar
Bruce MacDonald committed
325
326
327
328
				}
			}

			ch <- resp
Bruce MacDonald's avatar
Bruce MacDonald committed
329
330
		}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
331
		var images []llm.ImageData
Michael Yang's avatar
Michael Yang committed
332
		for i := range req.Images {
Jeffrey Morgan's avatar
Jeffrey Morgan committed
333
334
335
336
			images = append(images, llm.ImageData{
				ID:   i,
				Data: req.Images[i],
			})
Michael Yang's avatar
Michael Yang committed
337
338
		}

Bruce MacDonald's avatar
Bruce MacDonald committed
339
340
		// Start prediction
		predictReq := llm.PredictOpts{
341
342
			Prompt:  prompt,
			Format:  req.Format,
Michael Yang's avatar
Michael Yang committed
343
			Images:  images,
344
			Options: opts,
Bruce MacDonald's avatar
Bruce MacDonald committed
345
346
		}
		if err := loaded.runner.Predict(c.Request.Context(), predictReq, fn); err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
347
348
349
350
351
			ch <- gin.H{"error": err.Error()}
		}
	}()

	if req.Stream != nil && !*req.Stream {
352
353
		// Accumulate responses into the final response
		var final api.GenerateResponse
Bruce MacDonald's avatar
Bruce MacDonald committed
354
		var sb strings.Builder
Bruce MacDonald's avatar
Bruce MacDonald committed
355
		for resp := range ch {
356
357
358
359
360
361
362
363
364
365
366
367
368
369
			switch r := resp.(type) {
			case api.GenerateResponse:
				sb.WriteString(r.Response)
				final = r
			case gin.H:
				if errorMsg, ok := r["error"].(string); ok {
					c.JSON(http.StatusInternalServerError, gin.H{"error": errorMsg})
					return
				} else {
					c.JSON(http.StatusInternalServerError, gin.H{"error": "unexpected error format in response"})
					return
				}
			default:
				c.JSON(http.StatusInternalServerError, gin.H{"error": "unexpected error"})
Bruce MacDonald's avatar
Bruce MacDonald committed
370
371
372
				return
			}
		}
373
374
375

		final.Response = sb.String()
		c.JSON(http.StatusOK, final)
Bruce MacDonald's avatar
Bruce MacDonald committed
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
		return
	}

	streamResponse(c, ch)
}

func EmbeddingHandler(c *gin.Context) {
	loaded.mu.Lock()
	defer loaded.mu.Unlock()

	var req api.EmbeddingRequest
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

	if req.Model == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
		return
	}

402
	model, err := GetModel(req.Model)
Bruce MacDonald's avatar
Bruce MacDonald committed
403
	if err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
404
		var pErr *fs.PathError
405
		if errors.As(err, &pErr) {
Bruce MacDonald's avatar
Bruce MacDonald committed
406
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found, try pulling it first", req.Model)})
407
408
409
410
411
412
413
414
415
			return
		}
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	opts, err := modelOptions(model, req.Options)
	if err != nil {
		if errors.Is(err, api.ErrInvalidOpts) {
Bruce MacDonald's avatar
Bruce MacDonald committed
416
			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
417
			return
Bruce MacDonald's avatar
Bruce MacDonald committed
418
		}
419
420
421
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
422
423
424
425
426
427
428
429

	var sessionDuration time.Duration
	if req.KeepAlive == nil {
		sessionDuration = defaultSessionDuration
	} else {
		sessionDuration = req.KeepAlive.Duration
	}

430
431
	if err := load(c, model, opts, sessionDuration); err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
Bruce MacDonald's avatar
Bruce MacDonald committed
432
433
434
		return
	}

435
	if !loaded.Options.EmbeddingOnly {
Bruce MacDonald's avatar
Bruce MacDonald committed
436
437
438
439
		c.JSON(http.StatusBadRequest, gin.H{"error": "embedding option must be set to true"})
		return
	}

440
	embedding, err := loaded.runner.Embedding(c.Request.Context(), req.Prompt)
Bruce MacDonald's avatar
Bruce MacDonald committed
441
	if err != nil {
442
		slog.Info(fmt.Sprintf("embedding generation failed: %v", err))
Bruce MacDonald's avatar
Bruce MacDonald committed
443
444
445
446
447
448
449
450
451
452
		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate embedding"})
		return
	}

	resp := api.EmbeddingResponse{
		Embedding: embedding,
	}
	c.JSON(http.StatusOK, resp)
}

453
func PullModelHandler(c *gin.Context) {
Michael Yang's avatar
Michael Yang committed
454
	var req api.PullRequest
Michael Yang's avatar
Michael Yang committed
455
456
457
458
459
460
461
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
Michael Yang's avatar
Michael Yang committed
462
463
464
		return
	}

Michael Yang's avatar
Michael Yang committed
465
466
467
468
469
470
471
	var model string
	if req.Model != "" {
		model = req.Model
	} else if req.Name != "" {
		model = req.Name
	} else {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
472
473
474
		return
	}

475
476
477
	ch := make(chan any)
	go func() {
		defer close(ch)
478
479
		fn := func(r api.ProgressResponse) {
			ch <- r
480
		}
481

Michael Yang's avatar
Michael Yang committed
482
		regOpts := &registryOptions{
483
484
485
			Insecure: req.Insecure,
		}

486
487
488
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

Michael Yang's avatar
Michael Yang committed
489
		if err := PullModel(ctx, model, regOpts, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
490
			ch <- gin.H{"error": err.Error()}
491
492
493
		}
	}()

494
495
496
497
498
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

499
500
501
	streamResponse(c, ch)
}

502
func PushModelHandler(c *gin.Context) {
503
	var req api.PushRequest
Michael Yang's avatar
Michael Yang committed
504
505
506
507
508
509
510
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
Michael Yang's avatar
Michael Yang committed
511
512
		return
	}
Michael Yang's avatar
Michael Yang committed
513

Michael Yang's avatar
Michael Yang committed
514
515
516
517
518
519
520
	var model string
	if req.Model != "" {
		model = req.Model
	} else if req.Name != "" {
		model = req.Name
	} else {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
521
522
523
		return
	}

524
525
526
	ch := make(chan any)
	go func() {
		defer close(ch)
527
528
		fn := func(r api.ProgressResponse) {
			ch <- r
529
		}
530

Michael Yang's avatar
Michael Yang committed
531
		regOpts := &registryOptions{
532
533
534
			Insecure: req.Insecure,
		}

Michael Yang's avatar
Michael Yang committed
535
536
537
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

Michael Yang's avatar
Michael Yang committed
538
		if err := PushModel(ctx, model, regOpts, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
539
			ch <- gin.H{"error": err.Error()}
540
541
542
		}
	}()

543
544
545
546
547
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

548
549
550
	streamResponse(c, ch)
}

551
func CreateModelHandler(c *gin.Context) {
552
	var req api.CreateRequest
Michael Yang's avatar
Michael Yang committed
553
554
555
556
557
558
559
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
Michael Yang's avatar
Michael Yang committed
560
		return
561
562
	}

Michael Yang's avatar
Michael Yang committed
563
564
565
566
567
568
569
	var model string
	if req.Model != "" {
		model = req.Model
	} else if req.Name != "" {
		model = req.Name
	} else {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
570
571
572
		return
	}

Michael Yang's avatar
Michael Yang committed
573
	if err := ParseModelPath(model).Validate(); err != nil {
574
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
575
576
577
		return
	}

Michael Yang's avatar
Michael Yang committed
578
579
	if req.Path == "" && req.Modelfile == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "path or modelfile are required"})
Michael Yang's avatar
Michael Yang committed
580
581
		return
	}
Michael Yang's avatar
Michael Yang committed
582
583
584

	var modelfile io.Reader = strings.NewReader(req.Modelfile)
	if req.Path != "" && req.Modelfile == "" {
585
		mf, err := os.Open(req.Path)
Michael Yang's avatar
Michael Yang committed
586
587
588
589
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("error reading modelfile: %s", err)})
			return
		}
590
		defer mf.Close()
Michael Yang's avatar
Michael Yang committed
591

592
		modelfile = mf
Michael Yang's avatar
Michael Yang committed
593
	}
Michael Yang's avatar
Michael Yang committed
594
595
596
597
598
599
600

	commands, err := parser.Parse(modelfile)
	if err != nil {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

Michael Yang's avatar
Michael Yang committed
601
	ch := make(chan any)
Michael Yang's avatar
Michael Yang committed
602
603
	go func() {
		defer close(ch)
604
605
		fn := func(resp api.ProgressResponse) {
			ch <- resp
606
607
		}

608
609
610
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

Michael Yang's avatar
Michael Yang committed
611
		if err := CreateModel(ctx, model, filepath.Dir(req.Path), commands, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
612
			ch <- gin.H{"error": err.Error()}
613
		}
Michael Yang's avatar
Michael Yang committed
614
	}()
Michael Yang's avatar
Michael Yang committed
615

616
617
618
619
620
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

Michael Yang's avatar
Michael Yang committed
621
	streamResponse(c, ch)
Bruce MacDonald's avatar
Bruce MacDonald committed
622
623
}

624
625
func DeleteModelHandler(c *gin.Context) {
	var req api.DeleteRequest
Michael Yang's avatar
Michael Yang committed
626
627
628
629
630
631
632
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
633
634
635
		return
	}

Michael Yang's avatar
Michael Yang committed
636
637
638
639
640
641
642
	var model string
	if req.Model != "" {
		model = req.Model
	} else if req.Name != "" {
		model = req.Name
	} else {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
643
644
645
		return
	}

Michael Yang's avatar
Michael Yang committed
646
	if err := DeleteModel(model); err != nil {
647
		if os.IsNotExist(err) {
Michael Yang's avatar
Michael Yang committed
648
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", model)})
649
		} else {
650
651
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
652
653
		return
	}
Michael Yang's avatar
Michael Yang committed
654
655
656
657
658
659
660
661
662
663
664
665

	manifestsPath, err := GetManifestPath()
	if err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	if err := PruneDirectory(manifestsPath); err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

666
	c.JSON(http.StatusOK, nil)
667
668
}

Patrick Devine's avatar
Patrick Devine committed
669
670
func ShowModelHandler(c *gin.Context) {
	var req api.ShowRequest
Michael Yang's avatar
Michael Yang committed
671
672
673
674
675
676
677
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
Patrick Devine's avatar
Patrick Devine committed
678
679
680
		return
	}

Michael Yang's avatar
Michael Yang committed
681
	if req.Model != "" {
Michael Yang's avatar
Michael Yang committed
682
		// noop
Michael Yang's avatar
Michael Yang committed
683
	} else if req.Name != "" {
Michael Yang's avatar
Michael Yang committed
684
		req.Model = req.Name
Michael Yang's avatar
Michael Yang committed
685
	} else {
686
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
687
688
689
		return
	}

690
	resp, err := GetModelInfo(req)
Patrick Devine's avatar
Patrick Devine committed
691
692
	if err != nil {
		if os.IsNotExist(err) {
Michael Yang's avatar
Michael Yang committed
693
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Model)})
Patrick Devine's avatar
Patrick Devine committed
694
695
696
697
698
699
700
701
702
		} else {
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
		return
	}

	c.JSON(http.StatusOK, resp)
}

703
704
func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
	model, err := GetModel(req.Model)
Patrick Devine's avatar
Patrick Devine committed
705
706
707
708
	if err != nil {
		return nil, err
	}

Patrick Devine's avatar
Patrick Devine committed
709
	modelDetails := api.ModelDetails{
710
		ParentModel:       model.ParentModel,
Patrick Devine's avatar
Patrick Devine committed
711
712
713
714
715
716
717
		Format:            model.Config.ModelFormat,
		Family:            model.Config.ModelFamily,
		Families:          model.Config.ModelFamilies,
		ParameterSize:     model.Config.ModelType,
		QuantizationLevel: model.Config.FileType,
	}

718
719
720
721
722
723
724
725
	if req.System != "" {
		model.System = req.System
	}

	if req.Template != "" {
		model.Template = req.Template
	}

726
727
728
729
730
	msgs := make([]api.Message, 0)
	for _, msg := range model.Messages {
		msgs = append(msgs, api.Message{Role: msg.Role, Content: msg.Content})
	}

Patrick Devine's avatar
Patrick Devine committed
731
732
733
734
	resp := &api.ShowResponse{
		License:  strings.Join(model.License, "\n"),
		System:   model.System,
		Template: model.Template,
Patrick Devine's avatar
Patrick Devine committed
735
		Details:  modelDetails,
736
		Messages: msgs,
Patrick Devine's avatar
Patrick Devine committed
737
738
739
740
741
742
743
744
	}

	var params []string
	cs := 30
	for k, v := range model.Options {
		switch val := v.(type) {
		case []interface{}:
			for _, nv := range val {
Patrick Devine's avatar
Patrick Devine committed
745
				params = append(params, fmt.Sprintf("%-*s %#v", cs, k, nv))
Patrick Devine's avatar
Patrick Devine committed
746
			}
Patrick Devine's avatar
Patrick Devine committed
747
748
		default:
			params = append(params, fmt.Sprintf("%-*s %#v", cs, k, v))
Patrick Devine's avatar
Patrick Devine committed
749
750
751
752
		}
	}
	resp.Parameters = strings.Join(params, "\n")

753
754
755
756
757
758
759
760
761
762
763
764
765
	for k, v := range req.Options {
		if _, ok := req.Options[k]; ok {
			model.Options[k] = v
		}
	}

	mf, err := ShowModelfile(model)
	if err != nil {
		return nil, err
	}

	resp.Modelfile = mf

Patrick Devine's avatar
Patrick Devine committed
766
767
768
	return resp, nil
}

769
func ListModelsHandler(c *gin.Context) {
770
	models := make([]api.ModelResponse, 0)
771
	manifestsPath, err := GetManifestPath()
Patrick Devine's avatar
Patrick Devine committed
772
773
774
775
	if err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
Michael Yang's avatar
Michael Yang committed
776

Patrick Devine's avatar
Patrick Devine committed
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
	modelResponse := func(modelName string) (api.ModelResponse, error) {
		model, err := GetModel(modelName)
		if err != nil {
			return api.ModelResponse{}, err
		}

		modelDetails := api.ModelDetails{
			Format:            model.Config.ModelFormat,
			Family:            model.Config.ModelFamily,
			Families:          model.Config.ModelFamilies,
			ParameterSize:     model.Config.ModelType,
			QuantizationLevel: model.Config.FileType,
		}

		return api.ModelResponse{
Michael Yang's avatar
Michael Yang committed
792
			Model:   model.ShortName,
Patrick Devine's avatar
Patrick Devine committed
793
794
795
796
797
798
799
			Name:    model.ShortName,
			Size:    model.Size,
			Digest:  model.Digest,
			Details: modelDetails,
		}, nil
	}

Michael Yang's avatar
Michael Yang committed
800
	walkFunc := func(path string, info os.FileInfo, _ error) error {
Patrick Devine's avatar
Patrick Devine committed
801
		if !info.IsDir() {
802
803
804
805
			path, tag := filepath.Split(path)
			model := strings.Trim(strings.TrimPrefix(path, manifestsPath), string(os.PathSeparator))
			modelPath := strings.Join([]string{model, tag}, ":")
			canonicalModelPath := strings.ReplaceAll(modelPath, string(os.PathSeparator), "/")
806

807
			resp, err := modelResponse(canonicalModelPath)
Patrick Devine's avatar
Patrick Devine committed
808
			if err != nil {
809
				slog.Info(fmt.Sprintf("skipping file: %s", canonicalModelPath))
Michael Yang's avatar
Michael Yang committed
810
				// nolint: nilerr
811
				return nil
Patrick Devine's avatar
Patrick Devine committed
812
			}
Michael Yang's avatar
Michael Yang committed
813

Patrick Devine's avatar
Patrick Devine committed
814
815
			resp.ModifiedAt = info.ModTime()
			models = append(models, resp)
Patrick Devine's avatar
Patrick Devine committed
816
		}
Michael Yang's avatar
Michael Yang committed
817

Patrick Devine's avatar
Patrick Devine committed
818
		return nil
Michael Yang's avatar
Michael Yang committed
819
820
	}

821
	if err := filepath.Walk(manifestsPath, walkFunc); err != nil {
Patrick Devine's avatar
Patrick Devine committed
822
823
824
825
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

Michael Yang's avatar
Michael Yang committed
826
	c.JSON(http.StatusOK, api.ListResponse{Models: models})
Patrick Devine's avatar
Patrick Devine committed
827
828
}

Patrick Devine's avatar
Patrick Devine committed
829
830
func CopyModelHandler(c *gin.Context) {
	var req api.CopyRequest
Michael Yang's avatar
Michael Yang committed
831
832
833
834
835
836
837
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
Patrick Devine's avatar
Patrick Devine committed
838
839
840
		return
	}

841
842
843
844
845
	if req.Source == "" || req.Destination == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "source add destination are required"})
		return
	}

846
847
848
849
850
	if err := ParseModelPath(req.Destination).Validate(); err != nil {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

Patrick Devine's avatar
Patrick Devine committed
851
852
853
854
855
856
857
858
859
860
	if err := CopyModel(req.Source, req.Destination); err != nil {
		if os.IsNotExist(err) {
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Source)})
		} else {
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
		return
	}
}

Michael Yang's avatar
Michael Yang committed
861
func HeadBlobHandler(c *gin.Context) {
Michael Yang's avatar
Michael Yang committed
862
863
864
865
866
867
868
869
870
871
872
	path, err := GetBlobsPath(c.Param("digest"))
	if err != nil {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

	if _, err := os.Stat(path); err != nil {
		c.AbortWithStatusJSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("blob %q not found", c.Param("digest"))})
		return
	}

Michael Yang's avatar
Michael Yang committed
873
	c.Status(http.StatusOK)
Michael Yang's avatar
Michael Yang committed
874
875
876
}

func CreateBlobHandler(c *gin.Context) {
877
	layer, err := NewLayer(c.Request.Body, "")
Michael Yang's avatar
Michael Yang committed
878
879
880
881
882
	if err != nil {
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

883
884
	if layer.Digest != c.Param("digest") {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("digest mismatch, expected %q, got %q", c.Param("digest"), layer.Digest)})
Michael Yang's avatar
Michael Yang committed
885
886
887
		return
	}

888
	if _, err := layer.Commit(); err != nil {
Michael Yang's avatar
Michael Yang committed
889
890
891
892
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

Michael Yang's avatar
Michael Yang committed
893
	c.Status(http.StatusCreated)
Michael Yang's avatar
Michael Yang committed
894
895
}

Michael Yang's avatar
Michael Yang committed
896
897
898
899
900
901
var defaultAllowOrigins = []string{
	"localhost",
	"127.0.0.1",
	"0.0.0.0",
}

902
903
904
905
906
func NewServer() (*Server, error) {
	workDir, err := os.MkdirTemp("", "ollama")
	if err != nil {
		return nil, err
	}
907

908
909
910
911
	return &Server{
		WorkDir: workDir,
	}, nil
}
912

913
914
915
916
func (s *Server) GenerateRoutes() http.Handler {
	var origins []string
	if o := os.Getenv("OLLAMA_ORIGINS"); o != "" {
		origins = strings.Split(o, ",")
917
918
	}

Michael Yang's avatar
Michael Yang committed
919
920
	config := cors.DefaultConfig()
	config.AllowWildcard = true
921
	config.AllowBrowserExtensions = true
Michael Yang's avatar
Michael Yang committed
922

923
	config.AllowOrigins = origins
Michael Yang's avatar
Michael Yang committed
924
925
926
927
928
929
930
931
	for _, allowOrigin := range defaultAllowOrigins {
		config.AllowOrigins = append(config.AllowOrigins,
			fmt.Sprintf("http://%s", allowOrigin),
			fmt.Sprintf("https://%s", allowOrigin),
			fmt.Sprintf("http://%s:*", allowOrigin),
			fmt.Sprintf("https://%s:*", allowOrigin),
		)
	}
Michael Yang's avatar
Michael Yang committed
932

Bruce MacDonald's avatar
Bruce MacDonald committed
933
	r := gin.Default()
934
935
936
	r.Use(
		cors.New(config),
		func(c *gin.Context) {
937
			c.Set("workDir", s.WorkDir)
938
939
940
			c.Next()
		},
	)
Bruce MacDonald's avatar
Bruce MacDonald committed
941

942
943
	r.POST("/api/pull", PullModelHandler)
	r.POST("/api/generate", GenerateHandler)
Bruce MacDonald's avatar
Bruce MacDonald committed
944
	r.POST("/api/chat", ChatHandler)
Bruce MacDonald's avatar
Bruce MacDonald committed
945
	r.POST("/api/embeddings", EmbeddingHandler)
946
947
	r.POST("/api/create", CreateModelHandler)
	r.POST("/api/push", PushModelHandler)
Patrick Devine's avatar
Patrick Devine committed
948
	r.POST("/api/copy", CopyModelHandler)
949
	r.DELETE("/api/delete", DeleteModelHandler)
Patrick Devine's avatar
Patrick Devine committed
950
	r.POST("/api/show", ShowModelHandler)
Michael Yang's avatar
Michael Yang committed
951
	r.POST("/api/blobs/:digest", CreateBlobHandler)
Michael Yang's avatar
Michael Yang committed
952
	r.HEAD("/api/blobs/:digest", HeadBlobHandler)
Jeffrey Morgan's avatar
Jeffrey Morgan committed
953

954
955
956
	// Compatibility endpoints
	r.POST("/v1/chat/completions", openai.Middleware(), ChatHandler)

Michael Yang's avatar
Michael Yang committed
957
958
959
960
961
962
	for _, method := range []string{http.MethodGet, http.MethodHead} {
		r.Handle(method, "/", func(c *gin.Context) {
			c.String(http.StatusOK, "Ollama is running")
		})

		r.Handle(method, "/api/tags", ListModelsHandler)
Michael Yang's avatar
Michael Yang committed
963
964
965
		r.Handle(method, "/api/version", func(c *gin.Context) {
			c.JSON(http.StatusOK, gin.H{"version": version.Version})
		})
Michael Yang's avatar
Michael Yang committed
966
967
	}

968
969
970
971
	return r
}

func Serve(ln net.Listener) error {
Michael Yang's avatar
Michael Yang committed
972
	level := slog.LevelInfo
973
	if debug := os.Getenv("OLLAMA_DEBUG"); debug != "" {
Michael Yang's avatar
Michael Yang committed
974
		level = slog.LevelDebug
975
	}
Michael Yang's avatar
Michael Yang committed
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991

	handler := slog.NewTextHandler(os.Stderr, &slog.HandlerOptions{
		Level:     level,
		AddSource: true,
		ReplaceAttr: func(_ []string, attr slog.Attr) slog.Attr {
			if attr.Key == slog.SourceKey {
				source := attr.Value.Any().(*slog.Source)
				source.File = filepath.Base(source.File)
			}

			return attr
		},
	})

	slog.SetDefault(slog.New(handler))

992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
	if noprune := os.Getenv("OLLAMA_NOPRUNE"); noprune == "" {
		// clean up unused layers and manifests
		if err := PruneLayers(); err != nil {
			return err
		}

		manifestsPath, err := GetManifestPath()
		if err != nil {
			return err
		}

		if err := PruneDirectory(manifestsPath); err != nil {
			return err
		}
	}

	s, err := NewServer()
	if err != nil {
		return err
	}
	r := s.GenerateRoutes()

1014
	slog.Info(fmt.Sprintf("Listening on %s (version %s)", ln.Addr(), version.Version))
1015
	srvr := &http.Server{
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1016
1017
1018
		Handler: r,
	}

1019
1020
	// listen for a ctrl+c and stop any loaded llm
	signals := make(chan os.Signal, 1)
1021
	signal.Notify(signals, syscall.SIGINT, syscall.SIGTERM)
1022
1023
	go func() {
		<-signals
1024
1025
		if loaded.runner != nil {
			loaded.runner.Close()
1026
		}
1027
		os.RemoveAll(s.WorkDir)
1028
1029
1030
		os.Exit(0)
	}()

1031
1032
1033
1034
	if err := llm.Init(s.WorkDir); err != nil {
		return fmt.Errorf("unable to initialize llm library %w", err)
	}
	if runtime.GOOS == "linux" { // TODO - windows too
1035
		// check compatibility to log warnings
1036
		if _, err := gpu.CheckVRAM(); err != nil {
1037
			slog.Info(err.Error())
1038
1039
1040
		}
	}

1041
	return srvr.Serve(ln)
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1042
}
Michael Yang's avatar
Michael Yang committed
1043

1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
func waitForStream(c *gin.Context, ch chan interface{}) {
	c.Header("Content-Type", "application/json")
	for resp := range ch {
		switch r := resp.(type) {
		case api.ProgressResponse:
			if r.Status == "success" {
				c.JSON(http.StatusOK, r)
				return
			}
		case gin.H:
			if errorMsg, ok := r["error"].(string); ok {
				c.JSON(http.StatusInternalServerError, gin.H{"error": errorMsg})
				return
			} else {
				c.JSON(http.StatusInternalServerError, gin.H{"error": "unexpected error format in progress response"})
				return
			}
		default:
			c.JSON(http.StatusInternalServerError, gin.H{"error": "unexpected progress response"})
			return
		}
	}
	c.JSON(http.StatusInternalServerError, gin.H{"error": "unexpected end of progress response"})
}

Michael Yang's avatar
Michael Yang committed
1069
func streamResponse(c *gin.Context, ch chan any) {
1070
	c.Header("Content-Type", "application/x-ndjson")
Michael Yang's avatar
Michael Yang committed
1071
1072
1073
1074
1075
1076
1077
1078
	c.Stream(func(w io.Writer) bool {
		val, ok := <-ch
		if !ok {
			return false
		}

		bts, err := json.Marshal(val)
		if err != nil {
1079
			slog.Info(fmt.Sprintf("streamResponse: json.Marshal failed with %s", err))
Michael Yang's avatar
Michael Yang committed
1080
1081
1082
			return false
		}

1083
		// Delineate chunks with new-line delimiter
Michael Yang's avatar
Michael Yang committed
1084
1085
		bts = append(bts, '\n')
		if _, err := w.Write(bts); err != nil {
1086
			slog.Info(fmt.Sprintf("streamResponse: w.Write failed with %s", err))
Michael Yang's avatar
Michael Yang committed
1087
1088
1089
1090
1091
1092
			return false
		}

		return true
	})
}
Bruce MacDonald's avatar
Bruce MacDonald committed
1093

1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
// ChatPrompt builds up a prompt from a series of messages for the currently `loaded` model
func chatPrompt(ctx context.Context, messages []api.Message) (string, error) {
	encode := func(s string) ([]int, error) {
		return loaded.runner.Encode(ctx, s)
	}

	prompt, err := ChatPrompt(loaded.Model.Template, loaded.Model.System, messages, loaded.Options.NumCtx, encode)
	if err != nil {
		return "", err
	}

	return prompt, nil
}

Bruce MacDonald's avatar
Bruce MacDonald committed
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
func ChatHandler(c *gin.Context) {
	loaded.mu.Lock()
	defer loaded.mu.Unlock()

	checkpointStart := time.Now()

	var req api.ChatRequest
	err := c.ShouldBindJSON(&req)
	switch {
	case errors.Is(err, io.EOF):
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "missing request body"})
		return
	case err != nil:
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

	// validate the request
	switch {
	case req.Model == "":
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
		return
	case len(req.Format) > 0 && req.Format != "json":
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "format must be json"})
		return
	}

1135
	model, err := GetModel(req.Model)
Bruce MacDonald's avatar
Bruce MacDonald committed
1136
1137
	if err != nil {
		var pErr *fs.PathError
1138
		if errors.As(err, &pErr) {
Bruce MacDonald's avatar
Bruce MacDonald committed
1139
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found, try pulling it first", req.Model)})
1140
1141
1142
1143
1144
1145
1146
1147
1148
			return
		}
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	opts, err := modelOptions(model, req.Options)
	if err != nil {
		if errors.Is(err, api.ErrInvalidOpts) {
Bruce MacDonald's avatar
Bruce MacDonald committed
1149
			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
1150
			return
Bruce MacDonald's avatar
Bruce MacDonald committed
1151
		}
1152
1153
1154
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
1155
1156
1157
1158
1159
1160
1161
1162

	var sessionDuration time.Duration
	if req.KeepAlive == nil {
		sessionDuration = defaultSessionDuration
	} else {
		sessionDuration = req.KeepAlive.Duration
	}

1163
1164
	if err := load(c, model, opts, sessionDuration); err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
Bruce MacDonald's avatar
Bruce MacDonald committed
1165
1166
1167
1168
1169
		return
	}

	checkpointLoaded := time.Now()

1170
	prompt, err := chatPrompt(c.Request.Context(), req.Messages)
Bruce MacDonald's avatar
Bruce MacDonald committed
1171
1172
1173
1174
	if err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}
Michael Yang's avatar
Michael Yang committed
1175

1176
	// an empty request loads the model
1177
	if len(req.Messages) == 0 || prompt == "" {
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
		resp := api.ChatResponse{
			CreatedAt: time.Now().UTC(),
			Model:     req.Model,
			Done:      true,
			Message:   api.Message{Role: "assistant"},
		}
		c.JSON(http.StatusOK, resp)
		return
	}

1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
	// only send images that are in the prompt
	var i int
	var images []llm.ImageData
	for _, m := range req.Messages {
		for _, img := range m.Images {
			if !isSupportedImageType(img) {
				c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "unsupported image format"})
				return
			}

			if strings.Contains(prompt, fmt.Sprintf("[img-%d]", i)) {
				images = append(images, llm.ImageData{Data: img, ID: i})
			}
			i += 1
		}
	}

	slog.Debug("chat handler", "prompt", prompt, "images", len(images))
1206

Bruce MacDonald's avatar
Bruce MacDonald committed
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
	ch := make(chan any)

	go func() {
		defer close(ch)

		fn := func(r llm.PredictResult) {
			// Update model expiration
			loaded.expireAt = time.Now().Add(sessionDuration)
			loaded.expireTimer.Reset(sessionDuration)

			resp := api.ChatResponse{
1218
				Model:     req.Model,
1219
				CreatedAt: time.Now().UTC(),
1220
				Message:   api.Message{Role: "assistant", Content: r.Content},
Bruce MacDonald's avatar
Bruce MacDonald committed
1221
1222
1223
1224
1225
1226
1227
1228
1229
				Done:      r.Done,
				Metrics: api.Metrics{
					PromptEvalCount:    r.PromptEvalCount,
					PromptEvalDuration: r.PromptEvalDuration,
					EvalCount:          r.EvalCount,
					EvalDuration:       r.EvalDuration,
				},
			}

1230
1231
1232
			if r.Done {
				resp.TotalDuration = time.Since(checkpointStart)
				resp.LoadDuration = checkpointLoaded.Sub(checkpointStart)
Bruce MacDonald's avatar
Bruce MacDonald committed
1233
1234
1235
1236
1237
1238
1239
			}

			ch <- resp
		}

		// Start prediction
		predictReq := llm.PredictOpts{
1240
1241
			Prompt:  prompt,
			Format:  req.Format,
Michael Yang's avatar
Michael Yang committed
1242
			Images:  images,
1243
			Options: opts,
Bruce MacDonald's avatar
Bruce MacDonald committed
1244
1245
1246
1247
1248
1249
1250
		}
		if err := loaded.runner.Predict(c.Request.Context(), predictReq, fn); err != nil {
			ch <- gin.H{"error": err.Error()}
		}
	}()

	if req.Stream != nil && !*req.Stream {
1251
1252
		// Accumulate responses into the final response
		var final api.ChatResponse
Bruce MacDonald's avatar
Bruce MacDonald committed
1253
1254
		var sb strings.Builder
		for resp := range ch {
1255
1256
			switch r := resp.(type) {
			case api.ChatResponse:
1257
				sb.WriteString(r.Message.Content)
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
				final = r
			case gin.H:
				if errorMsg, ok := r["error"].(string); ok {
					c.JSON(http.StatusInternalServerError, gin.H{"error": errorMsg})
					return
				} else {
					c.JSON(http.StatusInternalServerError, gin.H{"error": "unexpected error format in response"})
					return
				}
			default:
				c.JSON(http.StatusInternalServerError, gin.H{"error": "unexpected error"})
				return
Bruce MacDonald's avatar
Bruce MacDonald committed
1270
1271
			}
		}
1272

1273
		final.Message = api.Message{Role: "assistant", Content: sb.String()}
1274
		c.JSON(http.StatusOK, final)
Bruce MacDonald's avatar
Bruce MacDonald committed
1275
1276
1277
1278
1279
		return
	}

	streamResponse(c, ch)
}