routes.go 25.3 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"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
10
11
12
	"log"
	"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"
Patrick Devine's avatar
Patrick Devine committed
18
	"strconv"
Michael Yang's avatar
Michael Yang committed
19
	"strings"
Michael Yang's avatar
Michael Yang committed
20
	"sync"
21
	"syscall"
22
	"time"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
23

Michael Yang's avatar
Michael Yang committed
24
	"github.com/gin-contrib/cors"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
25
26
	"github.com/gin-gonic/gin"

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

Michael Yang's avatar
Michael Yang committed
33
34
35
36
37
38
39
40
41
42
43
44
45
46
var mode string = gin.DebugMode

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
47
var loaded struct {
Michael Yang's avatar
Michael Yang committed
48
49
	mu sync.Mutex

50
	runner llm.LLM
Michael Yang's avatar
Michael Yang committed
51
52
53

	expireAt    time.Time
	expireTimer *time.Timer
Jeffrey Morgan's avatar
Jeffrey Morgan committed
54

55
56
	*Model
	*api.Options
Michael Yang's avatar
Michael Yang committed
57
58
}

59
60
var defaultSessionDuration = 5 * time.Minute

Bruce MacDonald's avatar
Bruce MacDonald committed
61
// 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
Bruce MacDonald's avatar
Bruce MacDonald committed
62
63
64
65
66
67
68
69
func load(c *gin.Context, modelName string, reqOpts map[string]interface{}, sessionDuration time.Duration) (*Model, error) {
	model, err := GetModel(modelName)
	if err != nil {
		return nil, err
	}

	workDir := c.GetString("workDir")

70
71
72
	opts := api.DefaultOptions()
	if err := opts.FromMap(model.Options); err != nil {
		log.Printf("could not load model options: %v", err)
Bruce MacDonald's avatar
Bruce MacDonald committed
73
		return nil, err
74
75
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
76
	if err := opts.FromMap(reqOpts); err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
77
		return nil, err
78
79
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
80
81
	ctx := c.Request.Context()

82
	// check if the loaded model is still running in a subprocess, in case something unexpected happened
83
84
	if loaded.runner != nil {
		if err := loaded.runner.Ping(ctx); err != nil {
85
86
			log.Print("loaded llm process not responding, closing now")
			// the subprocess is no longer running, so close it
87
88
89
90
			loaded.runner.Close()
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
91
92
93
		}
	}

94
95
96
97
98
99
100
	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 {
101
			log.Println("changing loaded model")
102
103
104
105
			loaded.runner.Close()
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
Michael Yang's avatar
Michael Yang committed
106
		}
Michael Yang's avatar
Michael Yang committed
107

108
		llmRunner, err := llm.New(workDir, model.ModelPath, model.AdapterPaths, opts)
Michael Yang's avatar
Michael Yang committed
109
		if err != nil {
110
111
112
113
114
115
116
			// 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
			if strings.Contains(err.Error(), "failed to load model") {
				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)
			}

Bruce MacDonald's avatar
Bruce MacDonald committed
117
			return nil, err
Michael Yang's avatar
Michael Yang committed
118
119
		}

120
121
122
		loaded.Model = model
		loaded.runner = llmRunner
		loaded.Options = &opts
Michael Yang's avatar
Michael Yang committed
123
	}
124

Michael Yang's avatar
Michael Yang committed
125
126
127
128
	// update options for the loaded llm
	// TODO(mxyng): this isn't thread safe, but it should be fine for now
	loaded.runner.SetOptions(opts)

Jeffrey Morgan's avatar
Jeffrey Morgan committed
129
	loaded.expireAt = time.Now().Add(sessionDuration)
Bruce MacDonald's avatar
Bruce MacDonald committed
130

Jeffrey Morgan's avatar
Jeffrey Morgan committed
131
132
133
134
	if loaded.expireTimer == nil {
		loaded.expireTimer = time.AfterFunc(sessionDuration, func() {
			loaded.mu.Lock()
			defer loaded.mu.Unlock()
Michael Yang's avatar
Michael Yang committed
135

Jeffrey Morgan's avatar
Jeffrey Morgan committed
136
			if time.Now().Before(loaded.expireAt) {
Michael Yang's avatar
Michael Yang committed
137
138
139
				return
			}

140
141
			if loaded.runner != nil {
				loaded.runner.Close()
Michael Yang's avatar
Michael Yang committed
142
143
			}

144
145
146
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
Michael Yang's avatar
Michael Yang committed
147
		})
Michael Yang's avatar
Michael Yang committed
148
	}
149

Jeffrey Morgan's avatar
Jeffrey Morgan committed
150
	loaded.expireTimer.Reset(sessionDuration)
Bruce MacDonald's avatar
Bruce MacDonald committed
151
	return model, nil
Bruce MacDonald's avatar
Bruce MacDonald committed
152
153
154
155
156
157
158
159
160
}

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
161
162
163
164
165
166
167
	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()})
Bruce MacDonald's avatar
Bruce MacDonald committed
168
169
170
		return
	}

171
172
173
	// validate the request
	switch {
	case req.Model == "":
174
175
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
		return
176
177
178
	case len(req.Format) > 0 && req.Format != "json":
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "format must be json"})
		return
179
180
181
	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
182
183
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
184
185
	sessionDuration := defaultSessionDuration
	model, err := load(c, req.Model, req.Options, sessionDuration)
Bruce MacDonald's avatar
Bruce MacDonald committed
186
	if err != nil {
187
		var pErr *fs.PathError
Bruce MacDonald's avatar
Bruce MacDonald committed
188
189
		switch {
		case errors.As(err, &pErr):
190
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found, try pulling it first", req.Model)})
Bruce MacDonald's avatar
Bruce MacDonald committed
191
192
193
194
		case errors.Is(err, api.ErrInvalidOpts):
			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		default:
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
195
		}
Bruce MacDonald's avatar
Bruce MacDonald committed
196
197
198
		return
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
199
200
201
	// an empty request loads the model
	if req.Prompt == "" && req.Template == "" && req.System == "" {
		c.JSON(http.StatusOK, api.GenerateResponse{CreatedAt: time.Now().UTC(), Model: req.Model, Done: true})
Bruce MacDonald's avatar
Bruce MacDonald committed
202
203
204
205
206
		return
	}

	checkpointLoaded := time.Now()

Bruce MacDonald's avatar
Bruce MacDonald committed
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
	var prompt string
	switch {
	case req.Raw:
		prompt = req.Prompt
	case req.Prompt != "":
		if req.Template != "" {
			// override the default model template
			model.Template = req.Template
		}

		var rebuild strings.Builder
		if req.Context != nil {
			// TODO: context is deprecated, at some point the context logic within this conditional should be removed
			prevCtx, err := loaded.runner.Decode(c.Request.Context(), req.Context)
			if err != nil {
				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
				return
			}

			// Remove leading spaces from prevCtx if present
			prevCtx = strings.TrimPrefix(prevCtx, " ")
			rebuild.WriteString(prevCtx)
		}
		p, err := model.Prompt(PromptVars{
			System: req.System,
			Prompt: req.Prompt,
			First:  len(req.Context) == 0,
		})
235
236
237
238
		if err != nil {
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
			return
		}
Bruce MacDonald's avatar
Bruce MacDonald committed
239
240
		rebuild.WriteString(p)
		prompt = rebuild.String()
Bruce MacDonald's avatar
Bruce MacDonald committed
241
242
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
243
	ch := make(chan any)
Bruce MacDonald's avatar
Bruce MacDonald committed
244
	var generated strings.Builder
Bruce MacDonald's avatar
Bruce MacDonald committed
245
246
247
	go func() {
		defer close(ch)

Bruce MacDonald's avatar
Bruce MacDonald committed
248
249
		fn := func(r llm.PredictResult) {
			// Update model expiration
Bruce MacDonald's avatar
Bruce MacDonald committed
250
251
252
			loaded.expireAt = time.Now().Add(sessionDuration)
			loaded.expireTimer.Reset(sessionDuration)

Bruce MacDonald's avatar
Bruce MacDonald committed
253
254
255
256
			// 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
257
258
			}

Bruce MacDonald's avatar
Bruce MacDonald committed
259
260
261
262
263
264
265
266
267
268
269
270
271
			resp := api.GenerateResponse{
				Model:     r.Model,
				CreatedAt: r.CreatedAt,
				Done:      r.Done,
				Response:  r.Content,
				Metrics: api.Metrics{
					TotalDuration:      r.TotalDuration,
					LoadDuration:       r.LoadDuration,
					PromptEvalCount:    r.PromptEvalCount,
					PromptEvalDuration: r.PromptEvalDuration,
					EvalCount:          r.EvalCount,
					EvalDuration:       r.EvalDuration,
				},
Bruce MacDonald's avatar
Bruce MacDonald committed
272
273
			}

Bruce MacDonald's avatar
Bruce MacDonald committed
274
275
276
277
278
279
280
281
282
283
			if r.Done && !req.Raw {
				embd, err := loaded.runner.Encode(c.Request.Context(), req.Prompt+generated.String())
				if err != nil {
					ch <- gin.H{"error": err.Error()}
					return
				}
				resp.Context = embd
			}

			ch <- resp
Bruce MacDonald's avatar
Bruce MacDonald committed
284
285
		}

Bruce MacDonald's avatar
Bruce MacDonald committed
286
287
288
289
290
291
292
293
294
		// Start prediction
		predictReq := llm.PredictOpts{
			Model:            model.Name,
			Prompt:           prompt,
			Format:           req.Format,
			CheckpointStart:  checkpointStart,
			CheckpointLoaded: checkpointLoaded,
		}
		if err := loaded.runner.Predict(c.Request.Context(), predictReq, fn); err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
295
296
297
298
299
			ch <- gin.H{"error": err.Error()}
		}
	}()

	if req.Stream != nil && !*req.Stream {
Bruce MacDonald's avatar
Bruce MacDonald committed
300
301
302
		// Wait for the channel to close
		var r api.GenerateResponse
		var sb strings.Builder
Bruce MacDonald's avatar
Bruce MacDonald committed
303
		for resp := range ch {
Bruce MacDonald's avatar
Bruce MacDonald committed
304
305
			var ok bool
			if r, ok = resp.(api.GenerateResponse); !ok {
Bruce MacDonald's avatar
Bruce MacDonald committed
306
307
308
				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
				return
			}
Bruce MacDonald's avatar
Bruce MacDonald committed
309
			sb.WriteString(r.Response)
Bruce MacDonald's avatar
Bruce MacDonald committed
310
		}
Bruce MacDonald's avatar
Bruce MacDonald committed
311
312
		r.Response = sb.String()
		c.JSON(http.StatusOK, r)
Bruce MacDonald's avatar
Bruce MacDonald committed
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
		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
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
339
340
	sessionDuration := defaultSessionDuration
	_, err = load(c, req.Model, req.Options, sessionDuration)
Bruce MacDonald's avatar
Bruce MacDonald committed
341
	if err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
342
343
344
345
346
347
348
349
350
		var pErr *fs.PathError
		switch {
		case errors.As(err, &pErr):
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found, try pulling it first", req.Model)})
		case errors.Is(err, api.ErrInvalidOpts):
			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		default:
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
Bruce MacDonald's avatar
Bruce MacDonald committed
351
352
353
		return
	}

354
	if !loaded.Options.EmbeddingOnly {
Bruce MacDonald's avatar
Bruce MacDonald committed
355
356
357
358
		c.JSON(http.StatusBadRequest, gin.H{"error": "embedding option must be set to true"})
		return
	}

359
	embedding, err := loaded.runner.Embedding(c.Request.Context(), req.Prompt)
Bruce MacDonald's avatar
Bruce MacDonald committed
360
361
362
363
364
365
366
367
368
369
370
371
	if err != nil {
		log.Printf("embedding generation failed: %v", err)
		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate embedding"})
		return
	}

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

372
func PullModelHandler(c *gin.Context) {
Michael Yang's avatar
Michael Yang committed
373
	var req api.PullRequest
Michael Yang's avatar
Michael Yang committed
374
375
376
377
378
379
380
	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
381
382
383
		return
	}

384
385
386
387
388
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

389
390
391
	ch := make(chan any)
	go func() {
		defer close(ch)
392
393
		fn := func(r api.ProgressResponse) {
			ch <- r
394
		}
395

396
397
398
399
		regOpts := &RegistryOptions{
			Insecure: req.Insecure,
		}

400
401
402
403
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

		if err := PullModel(ctx, req.Name, regOpts, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
404
			ch <- gin.H{"error": err.Error()}
405
406
407
		}
	}()

408
409
410
411
412
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

413
414
415
	streamResponse(c, ch)
}

416
func PushModelHandler(c *gin.Context) {
417
	var req api.PushRequest
Michael Yang's avatar
Michael Yang committed
418
419
420
421
422
423
424
	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
425
426
		return
	}
Michael Yang's avatar
Michael Yang committed
427

428
429
430
431
432
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

433
434
435
	ch := make(chan any)
	go func() {
		defer close(ch)
436
437
		fn := func(r api.ProgressResponse) {
			ch <- r
438
		}
439

440
441
442
443
		regOpts := &RegistryOptions{
			Insecure: req.Insecure,
		}

Michael Yang's avatar
Michael Yang committed
444
445
446
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

447
		if err := PushModel(ctx, req.Name, regOpts, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
448
			ch <- gin.H{"error": err.Error()}
449
450
451
		}
	}()

452
453
454
455
456
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

457
458
459
	streamResponse(c, ch)
}

460
func CreateModelHandler(c *gin.Context) {
461
	var req api.CreateRequest
Michael Yang's avatar
Michael Yang committed
462
463
464
465
466
467
468
	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
469
		return
470
471
	}

Michael Yang's avatar
Michael Yang committed
472
473
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
474
475
476
		return
	}

477
478
	if err := ParseModelPath(req.Name).Validate(); err != nil {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": err.Error()})
479
480
481
		return
	}

Michael Yang's avatar
Michael Yang committed
482
483
	if req.Path == "" && req.Modelfile == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "path or modelfile are required"})
Michael Yang's avatar
Michael Yang committed
484
485
		return
	}
Michael Yang's avatar
Michael Yang committed
486
487
488

	var modelfile io.Reader = strings.NewReader(req.Modelfile)
	if req.Path != "" && req.Modelfile == "" {
489
		mf, err := os.Open(req.Path)
Michael Yang's avatar
Michael Yang committed
490
491
492
493
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("error reading modelfile: %s", err)})
			return
		}
494
		defer mf.Close()
Michael Yang's avatar
Michael Yang committed
495

496
		modelfile = mf
Michael Yang's avatar
Michael Yang committed
497
	}
Michael Yang's avatar
Michael Yang committed
498
499
500
501
502
503
504

	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
505
	ch := make(chan any)
Michael Yang's avatar
Michael Yang committed
506
507
	go func() {
		defer close(ch)
508
509
		fn := func(resp api.ProgressResponse) {
			ch <- resp
510
511
		}

512
513
514
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

515
		if err := CreateModel(ctx, req.Name, filepath.Dir(req.Path), commands, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
516
			ch <- gin.H{"error": err.Error()}
517
		}
Michael Yang's avatar
Michael Yang committed
518
	}()
Michael Yang's avatar
Michael Yang committed
519

520
521
522
523
524
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

Michael Yang's avatar
Michael Yang committed
525
	streamResponse(c, ch)
Bruce MacDonald's avatar
Bruce MacDonald committed
526
527
}

528
529
func DeleteModelHandler(c *gin.Context) {
	var req api.DeleteRequest
Michael Yang's avatar
Michael Yang committed
530
531
532
533
534
535
536
	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()})
537
538
539
		return
	}

540
541
542
543
544
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

545
546
547
548
	if err := DeleteModel(req.Name); err != nil {
		if os.IsNotExist(err) {
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Name)})
		} else {
549
550
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
551
552
		return
	}
Michael Yang's avatar
Michael Yang committed
553
554
555
556
557
558
559
560
561
562
563
564

	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
	}

565
	c.JSON(http.StatusOK, nil)
566
567
}

Patrick Devine's avatar
Patrick Devine committed
568
569
func ShowModelHandler(c *gin.Context) {
	var req api.ShowRequest
Michael Yang's avatar
Michael Yang committed
570
571
572
573
574
575
576
	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
577
578
579
		return
	}

580
581
582
583
584
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

Patrick Devine's avatar
Patrick Devine committed
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
	resp, err := GetModelInfo(req.Name)
	if err != nil {
		if os.IsNotExist(err) {
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Name)})
		} else {
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
		return
	}

	c.JSON(http.StatusOK, resp)
}

func GetModelInfo(name string) (*api.ShowResponse, error) {
	model, err := GetModel(name)
	if err != nil {
		return nil, err
	}

	resp := &api.ShowResponse{
		License:  strings.Join(model.License, "\n"),
		System:   model.System,
		Template: model.Template,
	}

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

	resp.Modelfile = mf

	var params []string
	cs := 30
	for k, v := range model.Options {
		switch val := v.(type) {
		case string:
			params = append(params, fmt.Sprintf("%-*s %s", cs, k, val))
		case int:
			params = append(params, fmt.Sprintf("%-*s %s", cs, k, strconv.Itoa(val)))
		case float64:
			params = append(params, fmt.Sprintf("%-*s %s", cs, k, strconv.FormatFloat(val, 'f', 0, 64)))
		case bool:
			params = append(params, fmt.Sprintf("%-*s %s", cs, k, strconv.FormatBool(val)))
		case []interface{}:
			for _, nv := range val {
				switch nval := nv.(type) {
				case string:
					params = append(params, fmt.Sprintf("%-*s %s", cs, k, nval))
				case int:
					params = append(params, fmt.Sprintf("%-*s %s", cs, k, strconv.Itoa(nval)))
				case float64:
					params = append(params, fmt.Sprintf("%-*s %s", cs, k, strconv.FormatFloat(nval, 'f', 0, 64)))
				case bool:
					params = append(params, fmt.Sprintf("%-*s %s", cs, k, strconv.FormatBool(nval)))
				}
			}
		}
	}
	resp.Parameters = strings.Join(params, "\n")

	return resp, nil
}

649
func ListModelsHandler(c *gin.Context) {
650
	models := make([]api.ModelResponse, 0)
Patrick Devine's avatar
Patrick Devine committed
651
652
653
654
655
	fp, err := GetManifestPath()
	if err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
Michael Yang's avatar
Michael Yang committed
656
657

	walkFunc := func(path string, info os.FileInfo, _ error) error {
Patrick Devine's avatar
Patrick Devine committed
658
		if !info.IsDir() {
Michael Yang's avatar
Michael Yang committed
659
660
661
			dir, file := filepath.Split(path)
			dir = strings.Trim(strings.TrimPrefix(dir, fp), string(os.PathSeparator))
			tag := strings.Join([]string{dir, file}, ":")
662

663
			mp := ParseModelPath(tag)
Patrick Devine's avatar
Patrick Devine committed
664
			manifest, digest, err := GetManifest(mp)
Patrick Devine's avatar
Patrick Devine committed
665
			if err != nil {
666
667
				log.Printf("skipping file: %s", fp)
				return nil
Patrick Devine's avatar
Patrick Devine committed
668
			}
Michael Yang's avatar
Michael Yang committed
669
670

			models = append(models, api.ModelResponse{
Patrick Devine's avatar
Patrick Devine committed
671
672
				Name:       mp.GetShortTagname(),
				Size:       manifest.GetTotalSize(),
Patrick Devine's avatar
Patrick Devine committed
673
				Digest:     digest,
Michael Yang's avatar
Michael Yang committed
674
675
				ModifiedAt: info.ModTime(),
			})
Patrick Devine's avatar
Patrick Devine committed
676
		}
Michael Yang's avatar
Michael Yang committed
677

Patrick Devine's avatar
Patrick Devine committed
678
		return nil
Michael Yang's avatar
Michael Yang committed
679
680
681
	}

	if err := filepath.Walk(fp, walkFunc); err != nil {
Patrick Devine's avatar
Patrick Devine committed
682
683
684
685
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

Michael Yang's avatar
Michael Yang committed
686
	c.JSON(http.StatusOK, api.ListResponse{Models: models})
Patrick Devine's avatar
Patrick Devine committed
687
688
}

Patrick Devine's avatar
Patrick Devine committed
689
690
func CopyModelHandler(c *gin.Context) {
	var req api.CopyRequest
Michael Yang's avatar
Michael Yang committed
691
692
693
694
695
696
697
	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
698
699
700
		return
	}

701
702
703
704
705
	if req.Source == "" || req.Destination == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "source add destination are required"})
		return
	}

706
707
708
709
710
	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
711
712
713
714
715
716
717
718
719
720
	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
721
func HeadBlobHandler(c *gin.Context) {
Michael Yang's avatar
Michael Yang committed
722
723
724
725
726
727
728
729
730
731
732
	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
733
	c.Status(http.StatusOK)
Michael Yang's avatar
Michael Yang committed
734
735
736
}

func CreateBlobHandler(c *gin.Context) {
737
	layer, err := NewLayer(c.Request.Body, "")
Michael Yang's avatar
Michael Yang committed
738
739
740
741
742
	if err != nil {
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

743
744
	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
745
746
747
		return
	}

748
	if _, err := layer.Commit(); err != nil {
Michael Yang's avatar
Michael Yang committed
749
750
751
752
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

Michael Yang's avatar
Michael Yang committed
753
	c.Status(http.StatusCreated)
Michael Yang's avatar
Michael Yang committed
754
755
}

Michael Yang's avatar
Michael Yang committed
756
757
758
759
760
761
762
var defaultAllowOrigins = []string{
	"localhost",
	"127.0.0.1",
	"0.0.0.0",
}

func Serve(ln net.Listener, allowOrigins []string) error {
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
	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
		}
	}

Michael Yang's avatar
Michael Yang committed
779
780
	config := cors.DefaultConfig()
	config.AllowWildcard = true
Michael Yang's avatar
Michael Yang committed
781
782
783
784
785
786
787
788
789
790

	config.AllowOrigins = allowOrigins
	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
791

792
793
794
795
796
797
	workDir, err := os.MkdirTemp("", "ollama")
	if err != nil {
		return err
	}
	defer os.RemoveAll(workDir)

Bruce MacDonald's avatar
Bruce MacDonald committed
798
	r := gin.Default()
799
800
801
802
803
804
805
	r.Use(
		cors.New(config),
		func(c *gin.Context) {
			c.Set("workDir", workDir)
			c.Next()
		},
	)
Bruce MacDonald's avatar
Bruce MacDonald committed
806

807
808
	r.POST("/api/pull", PullModelHandler)
	r.POST("/api/generate", GenerateHandler)
Bruce MacDonald's avatar
Bruce MacDonald committed
809
	r.POST("/api/chat", ChatHandler)
Bruce MacDonald's avatar
Bruce MacDonald committed
810
	r.POST("/api/embeddings", EmbeddingHandler)
811
812
	r.POST("/api/create", CreateModelHandler)
	r.POST("/api/push", PushModelHandler)
Patrick Devine's avatar
Patrick Devine committed
813
	r.POST("/api/copy", CopyModelHandler)
814
	r.DELETE("/api/delete", DeleteModelHandler)
Patrick Devine's avatar
Patrick Devine committed
815
	r.POST("/api/show", ShowModelHandler)
Michael Yang's avatar
Michael Yang committed
816
	r.POST("/api/blobs/:digest", CreateBlobHandler)
Michael Yang's avatar
Michael Yang committed
817
	r.HEAD("/api/blobs/:digest", HeadBlobHandler)
Jeffrey Morgan's avatar
Jeffrey Morgan committed
818

Michael Yang's avatar
Michael Yang committed
819
820
821
822
823
824
825
826
	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
827
	log.Printf("Listening on %s (version %s)", ln.Addr(), version.Version)
Jeffrey Morgan's avatar
Jeffrey Morgan committed
828
829
830
831
	s := &http.Server{
		Handler: r,
	}

832
833
	// listen for a ctrl+c and stop any loaded llm
	signals := make(chan os.Signal, 1)
834
	signal.Notify(signals, syscall.SIGINT, syscall.SIGTERM)
835
836
	go func() {
		<-signals
837
838
		if loaded.runner != nil {
			loaded.runner.Close()
839
		}
840
		os.RemoveAll(workDir)
841
842
843
		os.Exit(0)
	}()

844
845
846
	if runtime.GOOS == "linux" {
		// check compatibility to log warnings
		if _, err := llm.CheckVRAM(); err != nil {
847
			log.Printf(err.Error())
848
849
850
		}
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
851
852
	return s.Serve(ln)
}
Michael Yang's avatar
Michael Yang committed
853

854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
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
879
func streamResponse(c *gin.Context, ch chan any) {
880
	c.Header("Content-Type", "application/x-ndjson")
Michael Yang's avatar
Michael Yang committed
881
882
883
884
885
886
887
888
	c.Stream(func(w io.Writer) bool {
		val, ok := <-ch
		if !ok {
			return false
		}

		bts, err := json.Marshal(val)
		if err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
889
			log.Printf("streamResponse: json.Marshal failed with %s", err)
Michael Yang's avatar
Michael Yang committed
890
891
892
			return false
		}

893
		// Delineate chunks with new-line delimiter
Michael Yang's avatar
Michael Yang committed
894
895
		bts = append(bts, '\n')
		if _, err := w.Write(bts); err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
896
			log.Printf("streamResponse: w.Write failed with %s", err)
Michael Yang's avatar
Michael Yang committed
897
898
899
900
901
902
			return false
		}

		return true
	})
}
Bruce MacDonald's avatar
Bruce MacDonald committed
903
904
905
906
907
908
909
910
911
912
913
914
915
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
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
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
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024

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
	}

	sessionDuration := defaultSessionDuration
	model, err := load(c, req.Model, req.Options, sessionDuration)
	if err != nil {
		var pErr *fs.PathError
		switch {
		case errors.As(err, &pErr):
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found, try pulling it first", req.Model)})
		case errors.Is(err, api.ErrInvalidOpts):
			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		default:
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
		return
	}

	// an empty request loads the model
	if len(req.Messages) == 0 {
		c.JSON(http.StatusOK, api.ChatResponse{CreatedAt: time.Now().UTC(), Model: req.Model, Done: true})
		return
	}

	checkpointLoaded := time.Now()

	prompt, err := model.ChatPrompt(req.Messages)
	if err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

	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{
				Model:     r.Model,
				CreatedAt: r.CreatedAt,
				Done:      r.Done,
				Metrics: api.Metrics{
					TotalDuration:      r.TotalDuration,
					LoadDuration:       r.LoadDuration,
					PromptEvalCount:    r.PromptEvalCount,
					PromptEvalDuration: r.PromptEvalDuration,
					EvalCount:          r.EvalCount,
					EvalDuration:       r.EvalDuration,
				},
			}

			if !r.Done {
				resp.Message = &api.Message{Role: "assistant", Content: r.Content}
			}

			ch <- resp
		}

		// Start prediction
		predictReq := llm.PredictOpts{
			Model:            model.Name,
			Prompt:           prompt,
			Format:           req.Format,
			CheckpointStart:  checkpointStart,
			CheckpointLoaded: checkpointLoaded,
		}
		if err := loaded.runner.Predict(c.Request.Context(), predictReq, fn); err != nil {
			ch <- gin.H{"error": err.Error()}
		}
	}()

	if req.Stream != nil && !*req.Stream {
		// Wait for the channel to close
		var r api.ChatResponse
		var sb strings.Builder
		for resp := range ch {
			var ok bool
			if r, ok = resp.(api.ChatResponse); !ok {
				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
				return
			}
			if r.Message != nil {
				sb.WriteString(r.Message.Content)
			}
		}
		r.Message = &api.Message{Role: "assistant", Content: sb.String()}
		c.JSON(http.StatusOK, r)
		return
	}

	streamResponse(c, ch)
}