routes.go 21 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
	"crypto/sha256"
Michael Yang's avatar
Michael Yang committed
6
	"encoding/json"
7
	"errors"
8
	"fmt"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
9
	"io"
10
	"io/fs"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
11
12
13
	"log"
	"net"
	"net/http"
14
	"os"
15
	"os/signal"
Michael Yang's avatar
Michael Yang committed
16
	"path/filepath"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
17
	"reflect"
18
	"runtime"
Patrick Devine's avatar
Patrick Devine committed
19
	"strconv"
Michael Yang's avatar
Michael Yang committed
20
	"strings"
Michael Yang's avatar
Michael Yang committed
21
	"sync"
22
	"syscall"
23
	"time"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
24

Michael Yang's avatar
Michael Yang committed
25
	"github.com/gin-contrib/cors"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
26
27
	"github.com/gin-gonic/gin"

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

Michael Yang's avatar
Michael Yang committed
34
35
36
37
38
39
40
41
42
43
44
45
46
47
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
48
var loaded struct {
Michael Yang's avatar
Michael Yang committed
49
50
	mu sync.Mutex

51
	runner llm.LLM
Michael Yang's avatar
Michael Yang committed
52
53
54

	expireAt    time.Time
	expireTimer *time.Timer
Jeffrey Morgan's avatar
Jeffrey Morgan committed
55

56
57
	*Model
	*api.Options
Michael Yang's avatar
Michael Yang committed
58
59
}

60
61
var defaultSessionDuration = 5 * time.Minute

Bruce MacDonald's avatar
Bruce MacDonald committed
62
// 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
63
func load(ctx context.Context, workDir string, model *Model, reqOpts map[string]interface{}, sessionDuration time.Duration) error {
64
65
66
	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
67
		return err
68
69
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
70
71
	if err := opts.FromMap(reqOpts); err != nil {
		return err
72
73
	}

74
	// check if the loaded model is still running in a subprocess, in case something unexpected happened
75
76
	if loaded.runner != nil {
		if err := loaded.runner.Ping(ctx); err != nil {
77
78
			log.Print("loaded llm process not responding, closing now")
			// the subprocess is no longer running, so close it
79
80
81
82
			loaded.runner.Close()
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
83
84
85
		}
	}

86
87
88
89
90
91
92
	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 {
93
			log.Println("changing loaded model")
94
95
96
97
			loaded.runner.Close()
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
Michael Yang's avatar
Michael Yang committed
98
		}
Michael Yang's avatar
Michael Yang committed
99

100
		llmRunner, err := llm.New(workDir, model.ModelPath, model.AdapterPaths, opts)
Michael Yang's avatar
Michael Yang committed
101
		if err != nil {
102
103
104
105
106
107
108
			// 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
109
			return err
Michael Yang's avatar
Michael Yang committed
110
111
		}

112
113
114
		loaded.Model = model
		loaded.runner = llmRunner
		loaded.Options = &opts
Michael Yang's avatar
Michael Yang committed
115
	}
116

Michael Yang's avatar
Michael Yang committed
117
118
119
120
	// 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
121
	loaded.expireAt = time.Now().Add(sessionDuration)
Bruce MacDonald's avatar
Bruce MacDonald committed
122

Jeffrey Morgan's avatar
Jeffrey Morgan committed
123
124
125
126
	if loaded.expireTimer == nil {
		loaded.expireTimer = time.AfterFunc(sessionDuration, func() {
			loaded.mu.Lock()
			defer loaded.mu.Unlock()
Michael Yang's avatar
Michael Yang committed
127

Jeffrey Morgan's avatar
Jeffrey Morgan committed
128
			if time.Now().Before(loaded.expireAt) {
Michael Yang's avatar
Michael Yang committed
129
130
131
				return
			}

132
133
			if loaded.runner != nil {
				loaded.runner.Close()
Michael Yang's avatar
Michael Yang committed
134
135
			}

136
137
138
			loaded.runner = nil
			loaded.Model = nil
			loaded.Options = nil
Michael Yang's avatar
Michael Yang committed
139
		})
Michael Yang's avatar
Michael Yang committed
140
	}
141

Jeffrey Morgan's avatar
Jeffrey Morgan committed
142
	loaded.expireTimer.Reset(sessionDuration)
Bruce MacDonald's avatar
Bruce MacDonald committed
143
144
145
146
147
148
149
150
151
152
	return nil
}

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
153
154
155
156
157
158
159
	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
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
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
176
177
	model, err := GetModel(req.Model)
	if err != nil {
178
179
180
181
182
		var pErr *fs.PathError
		if errors.As(err, &pErr) {
			c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found, try pulling it first", req.Model)})
			return
		}
Bruce MacDonald's avatar
Bruce MacDonald committed
183
184
185
186
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

187
188
189
190
191
	workDir := c.GetString("workDir")

	// TODO: set this duration from the request if specified
	sessionDuration := defaultSessionDuration
	if err := load(c.Request.Context(), workDir, model, req.Options, sessionDuration); err != nil {
192
193
194
195
		if errors.Is(err, api.ErrInvalidOpts) {
			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
			return
		}
Bruce MacDonald's avatar
Bruce MacDonald committed
196
197
198
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
Michael Yang's avatar
Michael Yang committed
199

Michael Yang's avatar
Michael Yang committed
200
201
	checkpointLoaded := time.Now()

202
203
204
205
206
207
208
	prompt := req.Prompt
	if !req.Raw {
		prompt, err = model.Prompt(req)
		if err != nil {
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
			return
		}
Michael Yang's avatar
Michael Yang committed
209
	}
Jeffrey Morgan's avatar
Jeffrey Morgan committed
210

Michael Yang's avatar
Michael Yang committed
211
212
213
	ch := make(chan any)
	go func() {
		defer close(ch)
Michael Yang's avatar
Michael Yang committed
214
215
216
217
218
219
		// an empty request loads the model
		if req.Prompt == "" && req.Template == "" && req.System == "" {
			ch <- api.GenerateResponse{CreatedAt: time.Now().UTC(), Model: req.Model, Done: true}
			return
		}

Michael Yang's avatar
Michael Yang committed
220
		fn := func(r api.GenerateResponse) {
Jeffrey Morgan's avatar
Jeffrey Morgan committed
221
222
			loaded.expireAt = time.Now().Add(sessionDuration)
			loaded.expireTimer.Reset(sessionDuration)
Michael Yang's avatar
Michael Yang committed
223

Michael Yang's avatar
Michael Yang committed
224
225
226
			r.Model = req.Model
			r.CreatedAt = time.Now().UTC()
			if r.Done {
Michael Yang's avatar
Michael Yang committed
227
228
				r.TotalDuration = time.Since(checkpointStart)
				r.LoadDuration = checkpointLoaded.Sub(checkpointStart)
Michael Yang's avatar
Michael Yang committed
229
230
			}

231
232
233
234
235
			if req.Raw {
				// in raw mode the client must manage history on their own
				r.Context = nil
			}

Michael Yang's avatar
Michael Yang committed
236
			ch <- r
Michael Yang's avatar
Michael Yang committed
237
238
		}

239
		if err := loaded.runner.Predict(c.Request.Context(), req.Context, prompt, req.Format, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
240
			ch <- gin.H{"error": err.Error()}
Michael Yang's avatar
Michael Yang committed
241
		}
Michael Yang's avatar
Michael Yang committed
242
	}()
Michael Yang's avatar
Michael Yang committed
243

244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
	if req.Stream != nil && !*req.Stream {
		var response api.GenerateResponse
		generated := ""
		for resp := range ch {
			if r, ok := resp.(api.GenerateResponse); ok {
				generated += r.Response
				response = r
			} else {
				c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
				return
			}
		}
		response.Response = generated
		c.JSON(http.StatusOK, response)
		return
	}

Michael Yang's avatar
Michael Yang committed
261
	streamResponse(c, ch)
Michael Yang's avatar
Michael Yang committed
262
}
Michael Yang's avatar
Michael Yang committed
263

Bruce MacDonald's avatar
Bruce MacDonald committed
264
265
266
267
268
func EmbeddingHandler(c *gin.Context) {
	loaded.mu.Lock()
	defer loaded.mu.Unlock()

	var req api.EmbeddingRequest
Michael Yang's avatar
Michael Yang committed
269
270
271
272
273
274
275
	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
276
277
278
		return
	}

279
280
281
282
283
	if req.Model == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "model is required"})
		return
	}

Bruce MacDonald's avatar
Bruce MacDonald committed
284
285
286
287
288
	model, err := GetModel(req.Model)
	if err != nil {
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}
289
290
291

	workDir := c.GetString("workDir")
	if err := load(c.Request.Context(), workDir, model, req.Options, 5*time.Minute); err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
292
293
294
295
		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
		return
	}

296
	if !loaded.Options.EmbeddingOnly {
Bruce MacDonald's avatar
Bruce MacDonald committed
297
298
299
300
		c.JSON(http.StatusBadRequest, gin.H{"error": "embedding option must be set to true"})
		return
	}

301
	embedding, err := loaded.runner.Embedding(c.Request.Context(), req.Prompt)
Bruce MacDonald's avatar
Bruce MacDonald committed
302
303
304
305
306
307
308
309
310
311
312
313
	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)
}

314
func PullModelHandler(c *gin.Context) {
Michael Yang's avatar
Michael Yang committed
315
	var req api.PullRequest
Michael Yang's avatar
Michael Yang committed
316
317
318
319
320
321
322
	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
323
324
325
		return
	}

326
327
328
329
330
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

331
332
333
	ch := make(chan any)
	go func() {
		defer close(ch)
334
335
		fn := func(r api.ProgressResponse) {
			ch <- r
336
		}
337

338
339
340
341
		regOpts := &RegistryOptions{
			Insecure: req.Insecure,
		}

342
343
344
345
		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
346
			ch <- gin.H{"error": err.Error()}
347
348
349
		}
	}()

350
351
352
353
354
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

355
356
357
	streamResponse(c, ch)
}

358
func PushModelHandler(c *gin.Context) {
359
	var req api.PushRequest
Michael Yang's avatar
Michael Yang committed
360
361
362
363
364
365
366
	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
367
368
		return
	}
Michael Yang's avatar
Michael Yang committed
369

370
371
372
373
374
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

375
376
377
	ch := make(chan any)
	go func() {
		defer close(ch)
378
379
		fn := func(r api.ProgressResponse) {
			ch <- r
380
		}
381

382
383
384
385
		regOpts := &RegistryOptions{
			Insecure: req.Insecure,
		}

Michael Yang's avatar
Michael Yang committed
386
387
388
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

389
		if err := PushModel(ctx, req.Name, regOpts, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
390
			ch <- gin.H{"error": err.Error()}
391
392
393
		}
	}()

394
395
396
397
398
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

399
400
401
	streamResponse(c, ch)
}

402
func CreateModelHandler(c *gin.Context) {
403
	var req api.CreateRequest
Michael Yang's avatar
Michael Yang committed
404
405
406
407
408
409
410
	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
411
		return
412
413
	}

Michael Yang's avatar
Michael Yang committed
414
415
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
416
417
418
		return
	}

Michael Yang's avatar
Michael Yang committed
419
420
	if req.Path == "" && req.Modelfile == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "path or modelfile are required"})
Michael Yang's avatar
Michael Yang committed
421
422
		return
	}
Michael Yang's avatar
Michael Yang committed
423
424
425
426
427
428
429
430
431
432
433
434

	var modelfile io.Reader = strings.NewReader(req.Modelfile)
	if req.Path != "" && req.Modelfile == "" {
		bin, err := os.Open(req.Path)
		if err != nil {
			c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("error reading modelfile: %s", err)})
			return
		}
		defer bin.Close()

		modelfile = bin
	}
Michael Yang's avatar
Michael Yang committed
435
436
437
438
439
440
441

	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
442
	ch := make(chan any)
Michael Yang's avatar
Michael Yang committed
443
444
	go func() {
		defer close(ch)
445
446
		fn := func(resp api.ProgressResponse) {
			ch <- resp
447
448
		}

449
450
451
		ctx, cancel := context.WithCancel(c.Request.Context())
		defer cancel()

Michael Yang's avatar
Michael Yang committed
452
		if err := CreateModel(ctx, req.Name, commands, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
453
			ch <- gin.H{"error": err.Error()}
454
		}
Michael Yang's avatar
Michael Yang committed
455
	}()
Michael Yang's avatar
Michael Yang committed
456

457
458
459
460
461
	if req.Stream != nil && !*req.Stream {
		waitForStream(c, ch)
		return
	}

Michael Yang's avatar
Michael Yang committed
462
	streamResponse(c, ch)
Bruce MacDonald's avatar
Bruce MacDonald committed
463
464
}

465
466
func DeleteModelHandler(c *gin.Context) {
	var req api.DeleteRequest
Michael Yang's avatar
Michael Yang committed
467
468
469
470
471
472
473
	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()})
474
475
476
		return
	}

477
478
479
480
481
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

482
483
484
485
	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 {
486
487
			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		}
488
489
		return
	}
Michael Yang's avatar
Michael Yang committed
490
491
492
493
494
495
496
497
498
499
500
501

	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
	}

502
	c.JSON(http.StatusOK, nil)
503
504
}

Patrick Devine's avatar
Patrick Devine committed
505
506
func ShowModelHandler(c *gin.Context) {
	var req api.ShowRequest
Michael Yang's avatar
Michael Yang committed
507
508
509
510
511
512
513
	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
514
515
516
		return
	}

517
518
519
520
521
	if req.Name == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "name is required"})
		return
	}

Patrick Devine's avatar
Patrick Devine committed
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
	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
}

586
func ListModelsHandler(c *gin.Context) {
587
	models := make([]api.ModelResponse, 0)
Patrick Devine's avatar
Patrick Devine committed
588
589
590
591
592
	fp, err := GetManifestPath()
	if err != nil {
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
Michael Yang's avatar
Michael Yang committed
593
594

	walkFunc := func(path string, info os.FileInfo, _ error) error {
Patrick Devine's avatar
Patrick Devine committed
595
		if !info.IsDir() {
Michael Yang's avatar
Michael Yang committed
596
597
598
			dir, file := filepath.Split(path)
			dir = strings.Trim(strings.TrimPrefix(dir, fp), string(os.PathSeparator))
			tag := strings.Join([]string{dir, file}, ":")
599

600
			mp := ParseModelPath(tag)
Patrick Devine's avatar
Patrick Devine committed
601
			manifest, digest, err := GetManifest(mp)
Patrick Devine's avatar
Patrick Devine committed
602
			if err != nil {
603
604
				log.Printf("skipping file: %s", fp)
				return nil
Patrick Devine's avatar
Patrick Devine committed
605
			}
Michael Yang's avatar
Michael Yang committed
606
607

			models = append(models, api.ModelResponse{
Patrick Devine's avatar
Patrick Devine committed
608
609
				Name:       mp.GetShortTagname(),
				Size:       manifest.GetTotalSize(),
Patrick Devine's avatar
Patrick Devine committed
610
				Digest:     digest,
Michael Yang's avatar
Michael Yang committed
611
612
				ModifiedAt: info.ModTime(),
			})
Patrick Devine's avatar
Patrick Devine committed
613
		}
Michael Yang's avatar
Michael Yang committed
614

Patrick Devine's avatar
Patrick Devine committed
615
		return nil
Michael Yang's avatar
Michael Yang committed
616
617
618
	}

	if err := filepath.Walk(fp, walkFunc); err != nil {
Patrick Devine's avatar
Patrick Devine committed
619
620
621
622
		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

Michael Yang's avatar
Michael Yang committed
623
	c.JSON(http.StatusOK, api.ListResponse{Models: models})
Patrick Devine's avatar
Patrick Devine committed
624
625
}

Patrick Devine's avatar
Patrick Devine committed
626
627
func CopyModelHandler(c *gin.Context) {
	var req api.CopyRequest
Michael Yang's avatar
Michael Yang committed
628
629
630
631
632
633
634
	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
635
636
637
		return
	}

638
639
640
641
642
	if req.Source == "" || req.Destination == "" {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "source add destination are required"})
		return
	}

Patrick Devine's avatar
Patrick Devine committed
643
644
645
646
647
648
649
650
651
652
	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
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
func GetBlobHandler(c *gin.Context) {
	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
	}

	c.JSON(http.StatusOK, api.CreateBlobResponse{Path: path})
}

func CreateBlobHandler(c *gin.Context) {
	hash := sha256.New()
	temp, err := os.CreateTemp("", c.Param("digest"))
	if err != nil {
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}
	defer temp.Close()
	defer os.Remove(temp.Name())

	if _, err := io.Copy(temp, io.TeeReader(c.Request.Body, hash)); err != nil {
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	if fmt.Sprintf("sha256:%x", hash.Sum(nil)) != c.Param("digest") {
		c.AbortWithStatusJSON(http.StatusBadRequest, gin.H{"error": "digest does not match body"})
		return
	}

	if err := temp.Close(); err != nil {
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	targetPath, err := GetBlobsPath(c.Param("digest"))
	if err != nil {
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	if err := os.Rename(temp.Name(), targetPath); err != nil {
		c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
		return
	}

	c.JSON(http.StatusOK, api.CreateBlobResponse{Path: targetPath})
}

Michael Yang's avatar
Michael Yang committed
707
708
709
710
711
712
713
var defaultAllowOrigins = []string{
	"localhost",
	"127.0.0.1",
	"0.0.0.0",
}

func Serve(ln net.Listener, allowOrigins []string) error {
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
	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
730
731
	config := cors.DefaultConfig()
	config.AllowWildcard = true
Michael Yang's avatar
Michael Yang committed
732
733
734
735
736
737
738
739
740
741

	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
742

743
744
745
746
747
748
	workDir, err := os.MkdirTemp("", "ollama")
	if err != nil {
		return err
	}
	defer os.RemoveAll(workDir)

Bruce MacDonald's avatar
Bruce MacDonald committed
749
	r := gin.Default()
750
751
752
753
754
755
756
	r.Use(
		cors.New(config),
		func(c *gin.Context) {
			c.Set("workDir", workDir)
			c.Next()
		},
	)
Bruce MacDonald's avatar
Bruce MacDonald committed
757

758
759
	r.POST("/api/pull", PullModelHandler)
	r.POST("/api/generate", GenerateHandler)
Bruce MacDonald's avatar
Bruce MacDonald committed
760
	r.POST("/api/embeddings", EmbeddingHandler)
761
762
	r.POST("/api/create", CreateModelHandler)
	r.POST("/api/push", PushModelHandler)
Patrick Devine's avatar
Patrick Devine committed
763
	r.POST("/api/copy", CopyModelHandler)
764
	r.DELETE("/api/delete", DeleteModelHandler)
Patrick Devine's avatar
Patrick Devine committed
765
	r.POST("/api/show", ShowModelHandler)
Michael Yang's avatar
Michael Yang committed
766
	r.POST("/api/blobs/:digest", CreateBlobHandler)
Jeffrey Morgan's avatar
Jeffrey Morgan committed
767

Michael Yang's avatar
Michael Yang committed
768
769
770
771
772
773
	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
774
		r.Handle(method, "/api/blobs/:digest/path", GetBlobHandler)
Michael Yang's avatar
Michael Yang committed
775
776
	}

Michael Yang's avatar
Michael Yang committed
777
	log.Printf("Listening on %s (version %s)", ln.Addr(), version.Version)
Jeffrey Morgan's avatar
Jeffrey Morgan committed
778
779
780
781
	s := &http.Server{
		Handler: r,
	}

782
783
	// listen for a ctrl+c and stop any loaded llm
	signals := make(chan os.Signal, 1)
784
	signal.Notify(signals, syscall.SIGINT, syscall.SIGTERM)
785
786
	go func() {
		<-signals
787
788
		if loaded.runner != nil {
			loaded.runner.Close()
789
		}
790
		os.RemoveAll(workDir)
791
792
793
		os.Exit(0)
	}()

794
795
796
	if runtime.GOOS == "linux" {
		// check compatibility to log warnings
		if _, err := llm.CheckVRAM(); err != nil {
797
			log.Printf("Warning: GPU support may not be enabled, check you have installed GPU drivers: %v", err)
798
799
800
		}
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
801
802
	return s.Serve(ln)
}
Michael Yang's avatar
Michael Yang committed
803

804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
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
829
func streamResponse(c *gin.Context, ch chan any) {
830
	c.Header("Content-Type", "application/x-ndjson")
Michael Yang's avatar
Michael Yang committed
831
832
833
834
835
836
837
838
	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
839
			log.Printf("streamResponse: json.Marshal failed with %s", err)
Michael Yang's avatar
Michael Yang committed
840
841
842
			return false
		}

843
		// Delineate chunks with new-line delimiter
Michael Yang's avatar
Michael Yang committed
844
845
		bts = append(bts, '\n')
		if _, err := w.Write(bts); err != nil {
Bruce MacDonald's avatar
Bruce MacDonald committed
846
			log.Printf("streamResponse: w.Write failed with %s", err)
Michael Yang's avatar
Michael Yang committed
847
848
849
850
851
852
			return false
		}

		return true
	})
}