cmd.go 33.1 KB
Newer Older
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1
2
3
package cmd

import (
4
	"archive/zip"
Michael Yang's avatar
Michael Yang committed
5
	"bufio"
Michael Yang's avatar
Michael Yang committed
6
	"bytes"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
7
	"context"
8
9
	"crypto/ed25519"
	"crypto/rand"
Michael Yang's avatar
Michael Yang committed
10
	"crypto/sha256"
11
	"encoding/pem"
Michael Yang's avatar
Michael Yang committed
12
	"errors"
Bruce MacDonald's avatar
Bruce MacDonald committed
13
	"fmt"
Michael Yang's avatar
Michael Yang committed
14
	"io"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
15
	"log"
16
	"math"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
17
	"net"
18
	"net/http"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
19
	"os"
20
	"os/signal"
21
	"path/filepath"
22
	"regexp"
23
	"runtime"
24
	"slices"
Michael Yang's avatar
Michael Yang committed
25
	"strconv"
Michael Yang's avatar
Michael Yang committed
26
	"strings"
27
	"sync/atomic"
28
	"syscall"
Michael Yang's avatar
Michael Yang committed
29
	"time"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
30

31
	"github.com/containerd/console"
32
	"github.com/mattn/go-runewidth"
Patrick Devine's avatar
Patrick Devine committed
33
	"github.com/olekukonko/tablewriter"
Michael Yang's avatar
Michael Yang committed
34
	"github.com/spf13/cobra"
35
	"golang.org/x/crypto/ssh"
36
	"golang.org/x/term"
Michael Yang's avatar
Michael Yang committed
37

38
	"github.com/ollama/ollama/api"
39
	"github.com/ollama/ollama/auth"
40
	"github.com/ollama/ollama/envconfig"
41
	"github.com/ollama/ollama/format"
42
	"github.com/ollama/ollama/parser"
43
44
	"github.com/ollama/ollama/progress"
	"github.com/ollama/ollama/server"
45
46
	"github.com/ollama/ollama/types/errtypes"
	"github.com/ollama/ollama/types/model"
47
	"github.com/ollama/ollama/version"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
48
49
)

50
func CreateHandler(cmd *cobra.Command, args []string) error {
51
	filename, _ := cmd.Flags().GetString("file")
52
53
54
55
56
	filename, err := filepath.Abs(filename)
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
57
	client, err := api.ClientFromEnvironment()
58
59
60
	if err != nil {
		return err
	}
61

Michael Yang's avatar
Michael Yang committed
62
63
64
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

Michael Yang's avatar
Michael Yang committed
65
	f, err := os.Open(filename)
Michael Yang's avatar
Michael Yang committed
66
67
68
	if err != nil {
		return err
	}
Michael Yang's avatar
Michael Yang committed
69
	defer f.Close()
Michael Yang's avatar
Michael Yang committed
70

71
	modelfile, err := parser.ParseFile(f)
Michael Yang's avatar
Michael Yang committed
72
73
74
75
76
77
78
79
80
	if err != nil {
		return err
	}

	home, err := os.UserHomeDir()
	if err != nil {
		return err
	}

81
82
	status := "transferring model data"
	spinner := progress.NewSpinner(status)
83
	p.Add(status, spinner)
84
	defer p.Stop()
85

Michael Yang's avatar
Michael Yang committed
86
87
	for i := range modelfile.Commands {
		switch modelfile.Commands[i].Name {
Michael Yang's avatar
Michael Yang committed
88
		case "model", "adapter":
Michael Yang's avatar
Michael Yang committed
89
			path := modelfile.Commands[i].Args
Michael Yang's avatar
Michael Yang committed
90
91
92
93
94
95
			if path == "~" {
				path = home
			} else if strings.HasPrefix(path, "~/") {
				path = filepath.Join(home, path[2:])
			}

96
97
98
99
			if !filepath.IsAbs(path) {
				path = filepath.Join(filepath.Dir(filename), path)
			}

100
			fi, err := os.Stat(path)
Michael Yang's avatar
Michael Yang committed
101
			if errors.Is(err, os.ErrNotExist) && modelfile.Commands[i].Name == "model" {
Michael Yang's avatar
Michael Yang committed
102
				continue
Michael Yang's avatar
Michael Yang committed
103
104
105
106
			} else if err != nil {
				return err
			}

107
			if fi.IsDir() {
Michael Yang's avatar
Michael Yang committed
108
109
110
				// this is likely a safetensors or pytorch directory
				// TODO make this work w/ adapters
				tempfile, err := tempZipFiles(path)
111
112
113
				if err != nil {
					return err
				}
Michael Yang's avatar
Michael Yang committed
114
				defer os.RemoveAll(tempfile)
115

Michael Yang's avatar
Michael Yang committed
116
				path = tempfile
Michael Yang's avatar
Michael Yang committed
117
118
			}

119
			digest, err := createBlob(cmd, client, path, spinner)
120
			if err != nil {
Michael Yang's avatar
Michael Yang committed
121
122
123
				return err
			}

Michael Yang's avatar
Michael Yang committed
124
			modelfile.Commands[i].Args = "@" + digest
Michael Yang's avatar
Michael Yang committed
125
126
		}
	}
Michael Yang's avatar
Michael Yang committed
127

Michael Yang's avatar
Michael Yang committed
128
	bars := make(map[string]*progress.Bar)
129
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
130
131
132
133
134
		if resp.Digest != "" {
			spinner.Stop()

			bar, ok := bars[resp.Digest]
			if !ok {
135
				bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
136
137
138
139
140
141
142
143
144
145
146
147
148
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
			spinner.Stop()

			status = resp.Status
			spinner = progress.NewSpinner(status)
			p.Add(status, spinner)
		}

149
150
151
		return nil
	}

152
	quantize, _ := cmd.Flags().GetString("quantize")
Michael Yang's avatar
Michael Yang committed
153

154
	request := api.CreateRequest{Name: args[0], Modelfile: modelfile.String(), Quantize: quantize}
Michael Yang's avatar
Michael Yang committed
155
	if err := client.Create(cmd.Context(), &request, fn); err != nil {
156
157
158
159
160
161
		return err
	}

	return nil
}

Michael Yang's avatar
Michael Yang committed
162
163
164
165
166
167
168
func tempZipFiles(path string) (string, error) {
	tempfile, err := os.CreateTemp("", "ollama-tf")
	if err != nil {
		return "", err
	}
	defer tempfile.Close()

Michael Yang's avatar
Michael Yang committed
169
170
	detectContentType := func(path string) (string, error) {
		f, err := os.Open(path)
Michael Yang's avatar
Michael Yang committed
171
172
173
		if err != nil {
			return "", err
		}
Michael Yang's avatar
Michael Yang committed
174
		defer f.Close()
Michael Yang's avatar
Michael Yang committed
175

Michael Yang's avatar
Michael Yang committed
176
177
		var b bytes.Buffer
		b.Grow(512)
Michael Yang's avatar
Michael Yang committed
178

Michael Yang's avatar
Michael Yang committed
179
180
181
182
183
184
		if _, err := io.CopyN(&b, f, 512); err != nil && !errors.Is(err, io.EOF) {
			return "", err
		}

		contentType, _, _ := strings.Cut(http.DetectContentType(b.Bytes()), ";")
		return contentType, nil
Michael Yang's avatar
Michael Yang committed
185
186
	}

Michael Yang's avatar
Michael Yang committed
187
188
189
190
191
192
193
194
195
196
197
	glob := func(pattern, contentType string) ([]string, error) {
		matches, err := filepath.Glob(pattern)
		if err != nil {
			return nil, err
		}

		for _, safetensor := range matches {
			if ct, err := detectContentType(safetensor); err != nil {
				return nil, err
			} else if ct != contentType {
				return nil, fmt.Errorf("invalid content type: expected %s for %s", ct, safetensor)
Michael Yang's avatar
Michael Yang committed
198
			}
Michael Yang's avatar
Michael Yang committed
199
200
201
202
203
204
205
206
207
208
		}

		return matches, nil
	}

	var files []string
	if st, _ := glob(filepath.Join(path, "model*.safetensors"), "application/octet-stream"); len(st) > 0 {
		// safetensors files might be unresolved git lfs references; skip if they are
		// covers model-x-of-y.safetensors, model.fp32-x-of-y.safetensors, model.safetensors
		files = append(files, st...)
209
210
211
212
213
214
	} else if st, _ := glob(filepath.Join(path, "adapters.safetensors"), "application/octet-stream"); len(st) > 0 {
		// covers adapters.safetensors
		files = append(files, st...)
	} else if st, _ := glob(filepath.Join(path, "adapter_model.safetensors"), "application/octet-stream"); len(st) > 0 {
		// covers adapter_model.safetensors
		files = append(files, st...)
Michael Yang's avatar
Michael Yang committed
215
216
217
218
	} else if pt, _ := glob(filepath.Join(path, "pytorch_model*.bin"), "application/zip"); len(pt) > 0 {
		// pytorch files might also be unresolved git lfs references; skip if they are
		// covers pytorch_model-x-of-y.bin, pytorch_model.fp32-x-of-y.bin, pytorch_model.bin
		files = append(files, pt...)
Patrick Devine's avatar
Patrick Devine committed
219
	} else if pt, _ := glob(filepath.Join(path, "consolidated*.pth"), "application/zip"); len(pt) > 0 {
Michael Yang's avatar
Michael Yang committed
220
221
222
223
224
225
226
227
228
229
230
231
232
233
		// pytorch files might also be unresolved git lfs references; skip if they are
		// covers consolidated.x.pth, consolidated.pth
		files = append(files, pt...)
	} else {
		return "", errors.New("no safetensors or torch files found")
	}

	// add configuration files, json files are detected as text/plain
	js, err := glob(filepath.Join(path, "*.json"), "text/plain")
	if err != nil {
		return "", err
	}
	files = append(files, js...)

Michael Yang's avatar
Michael Yang committed
234
235
236
237
238
239
240
241
	// bert models require a nested config.json
	// TODO(mxyng): merge this with the glob above
	js, err = glob(filepath.Join(path, "**/*.json"), "text/plain")
	if err != nil {
		return "", err
	}
	files = append(files, js...)

Michael Yang's avatar
Michael Yang committed
242
243
244
245
246
247
248
249
250
	if tks, _ := glob(filepath.Join(path, "tokenizer.model"), "application/octet-stream"); len(tks) > 0 {
		// add tokenizer.model if it exists, tokenizer.json is automatically picked up by the previous glob
		// tokenizer.model might be a unresolved git lfs reference; error if it is
		files = append(files, tks...)
	} else if tks, _ := glob(filepath.Join(path, "**/tokenizer.model"), "text/plain"); len(tks) > 0 {
		// some times tokenizer.model is in a subdirectory (e.g. meta-llama/Meta-Llama-3-8B)
		files = append(files, tks...)
	}

251
252
253
	zipfile := zip.NewWriter(tempfile)
	defer zipfile.Close()

Michael Yang's avatar
Michael Yang committed
254
255
256
	for _, file := range files {
		f, err := os.Open(file)
		if err != nil {
Michael Yang's avatar
Michael Yang committed
257
258
			return "", err
		}
Michael Yang's avatar
Michael Yang committed
259
		defer f.Close()
Michael Yang's avatar
Michael Yang committed
260
261
262
263
264
265

		fi, err := f.Stat()
		if err != nil {
			return "", err
		}

Michael Yang's avatar
Michael Yang committed
266
		zfi, err := zip.FileInfoHeader(fi)
Michael Yang's avatar
Michael Yang committed
267
268
269
270
		if err != nil {
			return "", err
		}

Michael Yang's avatar
Michael Yang committed
271
272
273
274
275
		zfi.Name, err = filepath.Rel(path, file)
		if err != nil {
			return "", err
		}

Michael Yang's avatar
Michael Yang committed
276
		zf, err := zipfile.CreateHeader(zfi)
Michael Yang's avatar
Michael Yang committed
277
278
279
280
		if err != nil {
			return "", err
		}

Michael Yang's avatar
Michael Yang committed
281
		if _, err := io.Copy(zf, f); err != nil {
Michael Yang's avatar
Michael Yang committed
282
283
284
285
286
287
288
			return "", err
		}
	}

	return tempfile.Name(), nil
}

289
func createBlob(cmd *cobra.Command, client *api.Client, path string, spinner *progress.Spinner) (string, error) {
290
291
292
293
294
295
	bin, err := os.Open(path)
	if err != nil {
		return "", err
	}
	defer bin.Close()

296
297
298
299
300
301
302
	// Get file info to retrieve the size
	fileInfo, err := bin.Stat()
	if err != nil {
		return "", err
	}
	fileSize := fileInfo.Size()

303
304
305
306
	hash := sha256.New()
	if _, err := io.Copy(hash, bin); err != nil {
		return "", err
	}
307
308
309
310

	if _, err := bin.Seek(0, io.SeekStart); err != nil {
		return "", err
	}
311

312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
	var pw progressWriter
	status := "transferring model data 0%"
	spinner.SetMessage(status)

	done := make(chan struct{})
	defer close(done)

	go func() {
		ticker := time.NewTicker(60 * time.Millisecond)
		defer ticker.Stop()
		for {
			select {
			case <-ticker.C:
				spinner.SetMessage(fmt.Sprintf("transferring model data %d%%", int(100*pw.n.Load()/fileSize)))
			case <-done:
				spinner.SetMessage("transferring model data 100%")
				return
			}
		}
	}()

333
	digest := fmt.Sprintf("sha256:%x", hash.Sum(nil))
334
	if err = client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
335
336
337
338
339
		return "", err
	}
	return digest, nil
}

340
341
342
343
344
345
346
347
348
type progressWriter struct {
	n atomic.Int64
}

func (w *progressWriter) Write(p []byte) (n int, err error) {
	w.n.Add(int64(len(p)))
	return len(p), nil
}

349
func RunHandler(cmd *cobra.Command, args []string) error {
350
351
352
	interactive := true

	opts := runOptions{
353
354
355
		Model:    args[0],
		WordWrap: os.Getenv("TERM") == "xterm-256color",
		Options:  map[string]interface{}{},
356
357
358
359
360
361
362
363
	}

	format, err := cmd.Flags().GetString("format")
	if err != nil {
		return err
	}
	opts.Format = format

364
365
366
367
368
369
370
371
372
373
374
375
	keepAlive, err := cmd.Flags().GetString("keepalive")
	if err != nil {
		return err
	}
	if keepAlive != "" {
		d, err := time.ParseDuration(keepAlive)
		if err != nil {
			return err
		}
		opts.KeepAlive = &api.Duration{Duration: d}
	}

376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
	prompts := args[1:]
	// prepend stdin to the prompt if provided
	if !term.IsTerminal(int(os.Stdin.Fd())) {
		in, err := io.ReadAll(os.Stdin)
		if err != nil {
			return err
		}

		prompts = append([]string{string(in)}, prompts...)
		opts.WordWrap = false
		interactive = false
	}
	opts.Prompt = strings.Join(prompts, " ")
	if len(prompts) > 0 {
		interactive = false
	}

	nowrap, err := cmd.Flags().GetBool("nowordwrap")
	if err != nil {
		return err
	}
	opts.WordWrap = !nowrap

399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
	// Fill out the rest of the options based on information about the
	// model.
	client, err := api.ClientFromEnvironment()
	if err != nil {
		return err
	}

	name := args[0]
	info, err := func() (*api.ShowResponse, error) {
		showReq := &api.ShowRequest{Name: name}
		info, err := client.Show(cmd.Context(), showReq)
		var se api.StatusError
		if errors.As(err, &se) && se.StatusCode == http.StatusNotFound {
			if err := PullHandler(cmd, []string{name}); err != nil {
				return nil, err
			}
			return client.Show(cmd.Context(), &api.ShowRequest{Name: name})
		}
		return info, err
	}()
	if err != nil {
		return err
421
422
	}

423
424
425
426
	opts.MultiModal = slices.Contains(info.Details.Families, "clip")
	opts.ParentModel = info.Details.ParentModel

	if interactive {
Michael Yang's avatar
Michael Yang committed
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
		if err := loadModel(cmd, &opts); err != nil {
			return err
		}

		for _, msg := range info.Messages {
			switch msg.Role {
			case "user":
				fmt.Printf(">>> %s\n", msg.Content)
			case "assistant":
				state := &displayResponseState{}
				displayResponse(msg.Content, opts.WordWrap, state)
				fmt.Println()
				fmt.Println()
			}
		}

443
444
445
		return generateInteractive(cmd, opts)
	}
	return generate(cmd, opts)
Bruce MacDonald's avatar
Bruce MacDonald committed
446
447
}

448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
func errFromUnknownKey(unknownKeyErr error) error {
	// find SSH public key in the error message
	sshKeyPattern := `ssh-\w+ [^\s"]+`
	re := regexp.MustCompile(sshKeyPattern)
	matches := re.FindStringSubmatch(unknownKeyErr.Error())

	if len(matches) > 0 {
		serverPubKey := matches[0]

		localPubKey, err := auth.GetPublicKey()
		if err != nil {
			return unknownKeyErr
		}

		if runtime.GOOS == "linux" && serverPubKey != localPubKey {
			// try the ollama service public key
			svcPubKey, err := os.ReadFile("/usr/share/ollama/.ollama/id_ed25519.pub")
			if err != nil {
				return unknownKeyErr
			}
			localPubKey = strings.TrimSpace(string(svcPubKey))
		}

		// check if the returned public key matches the local public key, this prevents adding a remote key to the user's account
		if serverPubKey != localPubKey {
			return unknownKeyErr
		}

		var msg strings.Builder
		msg.WriteString(unknownKeyErr.Error())
		msg.WriteString("\n\nYour ollama key is:\n")
		msg.WriteString(localPubKey)
		msg.WriteString("\nAdd your key at:\n")
		msg.WriteString("https://ollama.com/settings/keys")

		return errors.New(msg.String())
	}

	return unknownKeyErr
}

489
func PushHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
490
	client, err := api.ClientFromEnvironment()
491
492
493
	if err != nil {
		return err
	}
494

495
496
497
498
499
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
500
501
502
503
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

	bars := make(map[string]*progress.Bar)
504
505
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
506

507
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
508
		if resp.Digest != "" {
509
510
511
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
512
513
514

			bar, ok := bars[resp.Digest]
			if !ok {
515
				bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
516
517
518
519
520
521
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
522
523
524
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
525
526
527
528
529
530

			status = resp.Status
			spinner = progress.NewSpinner(status)
			p.Add(status, spinner)
		}

531
532
533
		return nil
	}

Michael Yang's avatar
Michael Yang committed
534
	request := api.PushRequest{Name: args[0], Insecure: insecure}
Michael Yang's avatar
Michael Yang committed
535
	if err := client.Push(cmd.Context(), &request, fn); err != nil {
536
537
538
539
540
541
542
543
544
545
546
547
548
549
		if spinner != nil {
			spinner.Stop()
		}
		if strings.Contains(err.Error(), "access denied") {
			return errors.New("you are not authorized to push to this namespace, create the model under a namespace you own")
		}
		host := model.ParseName(args[0]).Host
		isOllamaHost := strings.HasSuffix(host, ".ollama.ai") || strings.HasSuffix(host, ".ollama.com")
		if strings.Contains(err.Error(), errtypes.UnknownOllamaKeyErrMsg) && isOllamaHost {
			// the user has not added their ollama key to ollama.com
			// re-throw an error with a more user-friendly message
			return errFromUnknownKey(err)
		}

Michael Yang's avatar
Michael Yang committed
550
551
552
		return err
	}

553
	spinner.Stop()
Michael Yang's avatar
Michael Yang committed
554
	return nil
555
556
}

557
func ListHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
558
	client, err := api.ClientFromEnvironment()
559
560
561
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
562

Michael Yang's avatar
Michael Yang committed
563
	models, err := client.List(cmd.Context())
Patrick Devine's avatar
Patrick Devine committed
564
565
566
567
568
569
570
	if err != nil {
		return err
	}

	var data [][]string

	for _, m := range models.Models {
Michael Yang's avatar
Michael Yang committed
571
		if len(args) == 0 || strings.HasPrefix(m.Name, args[0]) {
572
			data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), format.HumanTime(m.ModifiedAt, "Never")})
Michael Yang's avatar
Michael Yang committed
573
		}
Patrick Devine's avatar
Patrick Devine committed
574
575
576
	}

	table := tablewriter.NewWriter(os.Stdout)
Patrick Devine's avatar
Patrick Devine committed
577
	table.SetHeader([]string{"NAME", "ID", "SIZE", "MODIFIED"})
Patrick Devine's avatar
Patrick Devine committed
578
579
580
581
582
	table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
	table.SetAlignment(tablewriter.ALIGN_LEFT)
	table.SetHeaderLine(false)
	table.SetBorder(false)
	table.SetNoWhiteSpace(true)
Michael Yang's avatar
Michael Yang committed
583
	table.SetTablePadding("    ")
Patrick Devine's avatar
Patrick Devine committed
584
585
586
587
588
589
	table.AppendBulk(data)
	table.Render()

	return nil
}

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
func ListRunningHandler(cmd *cobra.Command, args []string) error {
	client, err := api.ClientFromEnvironment()
	if err != nil {
		return err
	}

	models, err := client.ListRunning(cmd.Context())
	if err != nil {
		return err
	}

	var data [][]string

	for _, m := range models.Models {
		if len(args) == 0 || strings.HasPrefix(m.Name, args[0]) {
			var procStr string
			switch {
			case m.SizeVRAM == 0:
				procStr = "100% CPU"
			case m.SizeVRAM == m.Size:
				procStr = "100% GPU"
			case m.SizeVRAM > m.Size || m.Size == 0:
				procStr = "Unknown"
			default:
				sizeCPU := m.Size - m.SizeVRAM
				cpuPercent := math.Round(float64(sizeCPU) / float64(m.Size) * 100)
				procStr = fmt.Sprintf("%d%%/%d%% CPU/GPU", int(cpuPercent), int(100-cpuPercent))
			}
			data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), procStr, format.HumanTime(m.ExpiresAt, "Never")})
		}
	}

	table := tablewriter.NewWriter(os.Stdout)
	table.SetHeader([]string{"NAME", "ID", "SIZE", "PROCESSOR", "UNTIL"})
	table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
	table.SetAlignment(tablewriter.ALIGN_LEFT)
	table.SetHeaderLine(false)
	table.SetBorder(false)
	table.SetNoWhiteSpace(true)
Michael Yang's avatar
Michael Yang committed
629
	table.SetTablePadding("    ")
630
631
632
633
634
635
	table.AppendBulk(data)
	table.Render()

	return nil
}

636
func DeleteHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
637
	client, err := api.ClientFromEnvironment()
638
639
640
	if err != nil {
		return err
	}
641

642
643
	for _, name := range args {
		req := api.DeleteRequest{Name: name}
Michael Yang's avatar
Michael Yang committed
644
		if err := client.Delete(cmd.Context(), &req); err != nil {
645
646
647
			return err
		}
		fmt.Printf("deleted '%s'\n", name)
648
649
650
651
	}
	return nil
}

Patrick Devine's avatar
Patrick Devine committed
652
func ShowHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
653
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
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
	if err != nil {
		return err
	}

	license, errLicense := cmd.Flags().GetBool("license")
	modelfile, errModelfile := cmd.Flags().GetBool("modelfile")
	parameters, errParams := cmd.Flags().GetBool("parameters")
	system, errSystem := cmd.Flags().GetBool("system")
	template, errTemplate := cmd.Flags().GetBool("template")

	for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate} {
		if boolErr != nil {
			return errors.New("error retrieving flags")
		}
	}

	flagsSet := 0
	showType := ""

	if license {
		flagsSet++
		showType = "license"
	}

	if modelfile {
		flagsSet++
		showType = "modelfile"
	}

	if parameters {
		flagsSet++
		showType = "parameters"
	}

	if system {
		flagsSet++
		showType = "system"
	}

	if template {
		flagsSet++
		showType = "template"
	}

	if flagsSet > 1 {
699
		return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
700
701
	}

702
703
704
705
706
	req := api.ShowRequest{Name: args[0]}
	resp, err := client.Show(cmd.Context(), &req)
	if err != nil {
		return err
	}
707

708
	if flagsSet == 1 {
709
710
711
712
713
714
715
716
717
718
719
720
721
722
		switch showType {
		case "license":
			fmt.Println(resp.License)
		case "modelfile":
			fmt.Println(resp.Modelfile)
		case "parameters":
			fmt.Println(resp.Parameters)
		case "system":
			fmt.Println(resp.System)
		case "template":
			fmt.Println(resp.Template)
		}

		return nil
Patrick Devine's avatar
Patrick Devine committed
723
724
	}

Michael Yang's avatar
Michael Yang committed
725
	return showInfo(resp, os.Stdout)
726
727
}

Michael Yang's avatar
Michael Yang committed
728
729
730
731
732
733
734
735
func showInfo(resp *api.ShowResponse, w io.Writer) error {
	tableRender := func(header string, rows func() [][]string) {
		fmt.Fprintln(w, " ", header)
		table := tablewriter.NewWriter(w)
		table.SetAlignment(tablewriter.ALIGN_LEFT)
		table.SetBorder(false)
		table.SetNoWhiteSpace(true)
		table.SetTablePadding("    ")
736

Michael Yang's avatar
Michael Yang committed
737
738
739
		switch header {
		case "Template", "System", "License":
			table.SetColWidth(100)
740
741
		}

Michael Yang's avatar
Michael Yang committed
742
743
744
		table.AppendBulk(rows())
		table.Render()
		fmt.Fprintln(w)
Patrick Devine's avatar
Patrick Devine committed
745
746
	}

Michael Yang's avatar
Michael Yang committed
747
748
749
750
751
752
753
754
755
756
757
758
759
760
	tableRender("Model", func() (rows [][]string) {
		if resp.ModelInfo != nil {
			arch := resp.ModelInfo["general.architecture"].(string)
			rows = append(rows, []string{"", "architecture", arch})
			rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ModelInfo["general.parameter_count"].(float64)))})
			rows = append(rows, []string{"", "context length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)].(float64), 'f', -1, 64)})
			rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.embedding_length", arch)].(float64), 'f', -1, 64)})
		} else {
			rows = append(rows, []string{"", "architecture", resp.Details.Family})
			rows = append(rows, []string{"", "parameters", resp.Details.ParameterSize})
		}
		rows = append(rows, []string{"", "quantization", resp.Details.QuantizationLevel})
		return
	})
761

Michael Yang's avatar
Michael Yang committed
762
763
764
765
766
767
768
769
770
	if resp.ProjectorInfo != nil {
		tableRender("Projector", func() (rows [][]string) {
			arch := resp.ProjectorInfo["general.architecture"].(string)
			rows = append(rows, []string{"", "architecture", arch})
			rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ProjectorInfo["general.parameter_count"].(float64)))})
			rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.embedding_length", arch)].(float64), 'f', -1, 64)})
			rows = append(rows, []string{"", "dimensions", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.projection_dim", arch)].(float64), 'f', -1, 64)})
			return
		})
771
772
	}

Michael Yang's avatar
Michael Yang committed
773
774
775
776
777
778
779
780
781
782
	if resp.Parameters != "" {
		tableRender("Parameters", func() (rows [][]string) {
			scanner := bufio.NewScanner(strings.NewReader(resp.Parameters))
			for scanner.Scan() {
				if text := scanner.Text(); text != "" {
					rows = append(rows, append([]string{""}, strings.Fields(text)...))
				}
			}
			return
		})
783
784
	}

Michael Yang's avatar
Michael Yang committed
785
786
787
788
789
	head := func(s string, n int) (rows [][]string) {
		scanner := bufio.NewScanner(strings.NewReader(s))
		for scanner.Scan() && (len(rows) < n || n < 0) {
			if text := scanner.Text(); text != "" {
				rows = append(rows, []string{"", strings.TrimSpace(text)})
790
791
			}
		}
Michael Yang's avatar
Michael Yang committed
792
		return
793
794
	}

Michael Yang's avatar
Michael Yang committed
795
796
797
798
799
	if resp.System != "" {
		tableRender("System", func() [][]string {
			return head(resp.System, 2)
		})
	}
800

Michael Yang's avatar
Michael Yang committed
801
802
803
804
	if resp.License != "" {
		tableRender("License", func() [][]string {
			return head(resp.License, 2)
		})
805
	}
Michael Yang's avatar
Michael Yang committed
806
807

	return nil
808
809
}

Patrick Devine's avatar
Patrick Devine committed
810
func CopyHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
811
	client, err := api.ClientFromEnvironment()
812
813
814
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
815
816

	req := api.CopyRequest{Source: args[0], Destination: args[1]}
Michael Yang's avatar
Michael Yang committed
817
	if err := client.Copy(cmd.Context(), &req); err != nil {
Patrick Devine's avatar
Patrick Devine committed
818
819
820
821
822
823
		return err
	}
	fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
	return nil
}

824
func PullHandler(cmd *cobra.Command, args []string) error {
825
826
827
828
829
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
830
	client, err := api.ClientFromEnvironment()
831
832
833
	if err != nil {
		return err
	}
834

Michael Yang's avatar
Michael Yang committed
835
836
837
838
839
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

	bars := make(map[string]*progress.Bar)

840
841
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
842

843
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
844
		if resp.Digest != "" {
845
846
847
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
848
849
850

			bar, ok := bars[resp.Digest]
			if !ok {
851
				bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
852
853
854
855
856
857
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
858
859
860
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
861
862
863
864
865
866

			status = resp.Status
			spinner = progress.NewSpinner(status)
			p.Add(status, spinner)
		}

867
868
		return nil
	}
869

Michael Yang's avatar
Michael Yang committed
870
	request := api.PullRequest{Name: args[0], Insecure: insecure}
Michael Yang's avatar
Michael Yang committed
871
	if err := client.Pull(cmd.Context(), &request, fn); err != nil {
Michael Yang's avatar
Michael Yang committed
872
873
874
875
		return err
	}

	return nil
Michael Yang's avatar
Michael Yang committed
876
877
}

878
879
type generateContextKey string

880
type runOptions struct {
881
882
883
884
885
886
887
888
889
890
	Model       string
	ParentModel string
	Prompt      string
	Messages    []api.Message
	WordWrap    bool
	Format      string
	System      string
	Images      []api.ImageData
	Options     map[string]interface{}
	MultiModal  bool
891
	KeepAlive   *api.Duration
892
893
}

894
895
896
897
898
899
900
901
902
type displayResponseState struct {
	lineLength int
	wordBuffer string
}

func displayResponse(content string, wordWrap bool, state *displayResponseState) {
	termWidth, _, _ := term.GetSize(int(os.Stdout.Fd()))
	if wordWrap && termWidth >= 10 {
		for _, ch := range content {
Josh Yan's avatar
Josh Yan committed
903
904
			if state.lineLength+1 > termWidth-5 {
				if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
905
906
907
908
909
910
911
					fmt.Printf("%s%c", state.wordBuffer, ch)
					state.wordBuffer = ""
					state.lineLength = 0
					continue
				}

				// backtrack the length of the last word and clear to the end of the line
912
913
				a := runewidth.StringWidth(state.wordBuffer)
				if a > 0 {
914
					fmt.Printf("\x1b[%dD", a)
915
916
				}
				fmt.Printf("\x1b[K\n")
917
				fmt.Printf("%s%c", state.wordBuffer, ch)
918
919
920
				chWidth := runewidth.RuneWidth(ch)

				state.lineLength = runewidth.StringWidth(state.wordBuffer) + chWidth
921
922
			} else {
				fmt.Print(string(ch))
923
924
925
926
				state.lineLength += runewidth.RuneWidth(ch)
				if runewidth.RuneWidth(ch) >= 2 {
					state.wordBuffer = ""
					continue
Josh Yan's avatar
Josh Yan committed
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

				switch ch {
				case ' ':
					state.wordBuffer = ""
				case '\n':
					state.lineLength = 0
				default:
					state.wordBuffer += string(ch)
				}
			}
		}
	} else {
		fmt.Printf("%s%s", state.wordBuffer, content)
		if len(state.wordBuffer) > 0 {
			state.wordBuffer = ""
		}
	}
}

func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
	client, err := api.ClientFromEnvironment()
	if err != nil {
		return nil, err
	}

	p := progress.NewProgress(os.Stderr)
	defer p.StopAndClear()

	spinner := progress.NewSpinner("")
	p.Add("", spinner)

	cancelCtx, cancel := context.WithCancel(cmd.Context())
	defer cancel()

	sigChan := make(chan os.Signal, 1)
	signal.Notify(sigChan, syscall.SIGINT)

	go func() {
		<-sigChan
		cancel()
	}()

	var state *displayResponseState = &displayResponseState{}
	var latest api.ChatResponse
	var fullResponse strings.Builder
	var role string

	fn := func(response api.ChatResponse) error {
		p.StopAndClear()

		latest = response

		role = response.Message.Role
		content := response.Message.Content
		fullResponse.WriteString(content)

		displayResponse(content, opts.WordWrap, state)

		return nil
	}

	req := &api.ChatRequest{
		Model:    opts.Model,
		Messages: opts.Messages,
		Format:   opts.Format,
		Options:  opts.Options,
	}

996
997
998
999
	if opts.KeepAlive != nil {
		req.KeepAlive = opts.KeepAlive
	}

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
	if err := client.Chat(cancelCtx, req, fn); err != nil {
		if errors.Is(err, context.Canceled) {
			return nil, nil
		}
		return nil, err
	}

	if len(opts.Messages) > 0 {
		fmt.Println()
		fmt.Println()
	}

	verbose, err := cmd.Flags().GetBool("verbose")
	if err != nil {
		return nil, err
	}

	if verbose {
		latest.Summary()
	}

	return &api.Message{Role: role, Content: fullResponse.String()}, nil
}

func generate(cmd *cobra.Command, opts runOptions) error {
Michael Yang's avatar
Michael Yang committed
1025
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
1026
	if err != nil {
1027
		return err
Patrick Devine's avatar
Patrick Devine committed
1028
	}
Michael Yang's avatar
Michael Yang committed
1029

Michael Yang's avatar
Michael Yang committed
1030
	p := progress.NewProgress(os.Stderr)
1031
	defer p.StopAndClear()
1032

Michael Yang's avatar
Michael Yang committed
1033
1034
1035
	spinner := progress.NewSpinner("")
	p.Add("", spinner)

1036
1037
1038
1039
1040
1041
1042
	var latest api.GenerateResponse

	generateContext, ok := cmd.Context().Value(generateContextKey("context")).([]int)
	if !ok {
		generateContext = []int{}
	}

Michael Yang's avatar
Michael Yang committed
1043
	ctx, cancel := context.WithCancel(cmd.Context())
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
	defer cancel()

	sigChan := make(chan os.Signal, 1)
	signal.Notify(sigChan, syscall.SIGINT)

	go func() {
		<-sigChan
		cancel()
	}()

1054
	var state *displayResponseState = &displayResponseState{}
1055

1056
	fn := func(response api.GenerateResponse) error {
Michael Yang's avatar
Michael Yang committed
1057
		p.StopAndClear()
1058

Patrick Devine's avatar
Patrick Devine committed
1059
		latest = response
1060
		content := response.Response
1061

1062
		displayResponse(content, opts.WordWrap, state)
1063

Patrick Devine's avatar
Patrick Devine committed
1064
1065
		return nil
	}
1066

1067
1068
1069
1070
1071
1072
1073
	if opts.MultiModal {
		opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
		if err != nil {
			return err
		}
	}

Michael Yang's avatar
Michael Yang committed
1074
	request := api.GenerateRequest{
1075
1076
1077
1078
1079
1080
1081
1082
		Model:     opts.Model,
		Prompt:    opts.Prompt,
		Context:   generateContext,
		Images:    opts.Images,
		Format:    opts.Format,
		System:    opts.System,
		Options:   opts.Options,
		KeepAlive: opts.KeepAlive,
Michael Yang's avatar
Michael Yang committed
1083
1084
1085
	}

	if err := client.Generate(ctx, &request, fn); err != nil {
1086
		if errors.Is(err, context.Canceled) {
1087
			return nil
1088
		}
1089
		return err
Patrick Devine's avatar
Patrick Devine committed
1090
	}
1091

1092
	if opts.Prompt != "" {
Michael Yang's avatar
Michael Yang committed
1093
1094
		fmt.Println()
		fmt.Println()
Patrick Devine's avatar
Patrick Devine committed
1095
	}
1096

1097
1098
1099
1100
	if !latest.Done {
		return nil
	}

Patrick Devine's avatar
Patrick Devine committed
1101
1102
	verbose, err := cmd.Flags().GetBool("verbose")
	if err != nil {
1103
		return err
Patrick Devine's avatar
Patrick Devine committed
1104
	}
Michael Yang's avatar
Michael Yang committed
1105

Patrick Devine's avatar
Patrick Devine committed
1106
1107
	if verbose {
		latest.Summary()
Michael Yang's avatar
Michael Yang committed
1108
	}
Michael Yang's avatar
Michael Yang committed
1109

Patrick Devine's avatar
Patrick Devine committed
1110
1111
1112
	ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
	cmd.SetContext(ctx)

1113
	return nil
Michael Yang's avatar
Michael Yang committed
1114
1115
}

1116
func RunServer(_ *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1117
	if err := initializeKeypair(); err != nil {
1118
1119
1120
		return err
	}

Michael Yang's avatar
host  
Michael Yang committed
1121
	ln, err := net.Listen("tcp", envconfig.Host().Host)
1122
1123
1124
	if err != nil {
		return err
	}
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1125

1126
1127
1128
1129
1130
1131
	err = server.Serve(ln)
	if errors.Is(err, http.ErrServerClosed) {
		return nil
	}

	return err
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1132
1133
}

1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
func initializeKeypair() error {
	home, err := os.UserHomeDir()
	if err != nil {
		return err
	}

	privKeyPath := filepath.Join(home, ".ollama", "id_ed25519")
	pubKeyPath := filepath.Join(home, ".ollama", "id_ed25519.pub")

	_, err = os.Stat(privKeyPath)
	if os.IsNotExist(err) {
		fmt.Printf("Couldn't find '%s'. Generating new private key.\n", privKeyPath)
Michael Yang's avatar
Michael Yang committed
1146
		cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
1147
1148
1149
1150
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1151
		privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
1152
1153
1154
1155
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1156
		if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
1157
1158
1159
			return fmt.Errorf("could not create directory %w", err)
		}

Michael Yang's avatar
Michael Yang committed
1160
		if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
1161
1162
1163
			return err
		}

Michael Yang's avatar
Michael Yang committed
1164
		sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
1165
1166
1167
1168
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1169
		publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
1170

Michael Yang's avatar
Michael Yang committed
1171
		if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
1172
1173
1174
			return err
		}

Michael Yang's avatar
Michael Yang committed
1175
		fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
1176
1177
1178
1179
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1180
func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1181
	client, err := api.ClientFromEnvironment()
1182
1183
1184
	if err != nil {
		return err
	}
Michael Yang's avatar
Michael Yang committed
1185
	if err := client.Heartbeat(cmd.Context()); err != nil {
1186
		if !strings.Contains(err.Error(), " refused") {
Bruce MacDonald's avatar
Bruce MacDonald committed
1187
1188
			return err
		}
1189
		if err := startApp(cmd.Context(), client); err != nil {
Michael Yang's avatar
lint  
Michael Yang committed
1190
			return errors.New("could not connect to ollama app, is it running?")
1191
1192
1193
1194
1195
		}
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1196
1197
1198
1199
1200
1201
1202
1203
func versionHandler(cmd *cobra.Command, _ []string) {
	client, err := api.ClientFromEnvironment()
	if err != nil {
		return
	}

	serverVersion, err := client.Version(cmd.Context())
	if err != nil {
Michael Yang's avatar
Michael Yang committed
1204
1205
1206
1207
1208
		fmt.Println("Warning: could not connect to a running Ollama instance")
	}

	if serverVersion != "" {
		fmt.Printf("ollama version is %s\n", serverVersion)
Michael Yang's avatar
Michael Yang committed
1209
1210
	}

1211
	if serverVersion != version.Version {
Michael Yang's avatar
Michael Yang committed
1212
		fmt.Printf("Warning: client version is %s\n", version.Version)
1213
	}
Michael Yang's avatar
Michael Yang committed
1214
1215
}

1216
func appendEnvDocs(cmd *cobra.Command, envs []envconfig.EnvVar) {
1217
1218
1219
1220
1221
	if len(envs) == 0 {
		return
	}

	envUsage := `
1222
1223
Environment Variables:
`
1224
	for _, e := range envs {
1225
		envUsage += fmt.Sprintf("      %-24s   %s\n", e.Name, e.Description)
1226
1227
1228
	}

	cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
1229
1230
}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1231
1232
func NewCLI() *cobra.Command {
	log.SetFlags(log.LstdFlags | log.Lshortfile)
Michael Yang's avatar
Michael Yang committed
1233
	cobra.EnableCommandSorting = false
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1234

1235
	if runtime.GOOS == "windows" {
1236
		console.ConsoleFromFile(os.Stdin) //nolint:errcheck
1237
1238
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1239
	rootCmd := &cobra.Command{
1240
1241
1242
1243
		Use:           "ollama",
		Short:         "Large language model runner",
		SilenceUsage:  true,
		SilenceErrors: true,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1244
1245
1246
		CompletionOptions: cobra.CompletionOptions{
			DisableDefaultCmd: true,
		},
Michael Yang's avatar
Michael Yang committed
1247
1248
1249
1250
1251
1252
1253
1254
		Run: func(cmd *cobra.Command, args []string) {
			if version, _ := cmd.Flags().GetBool("version"); version {
				versionHandler(cmd, args)
				return
			}

			cmd.Print(cmd.UsageString())
		},
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1255
1256
	}

Michael Yang's avatar
Michael Yang committed
1257
	rootCmd.Flags().BoolP("version", "v", false, "Show version information")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1258

1259
	createCmd := &cobra.Command{
1260
1261
		Use:     "create MODEL",
		Short:   "Create a model from a Modelfile",
Michael Yang's avatar
Michael Yang committed
1262
		Args:    cobra.ExactArgs(1),
1263
1264
		PreRunE: checkServerHeartbeat,
		RunE:    CreateHandler,
1265
1266
	}

1267
	createCmd.Flags().StringP("file", "f", "Modelfile", "Name of the Modelfile")
1268
	createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_0)")
1269

Patrick Devine's avatar
Patrick Devine committed
1270
1271
1272
	showCmd := &cobra.Command{
		Use:     "show MODEL",
		Short:   "Show information for a model",
Michael Yang's avatar
Michael Yang committed
1273
		Args:    cobra.ExactArgs(1),
Patrick Devine's avatar
Patrick Devine committed
1274
1275
1276
1277
1278
1279
1280
1281
		PreRunE: checkServerHeartbeat,
		RunE:    ShowHandler,
	}

	showCmd.Flags().Bool("license", false, "Show license of a model")
	showCmd.Flags().Bool("modelfile", false, "Show Modelfile of a model")
	showCmd.Flags().Bool("parameters", false, "Show parameters of a model")
	showCmd.Flags().Bool("template", false, "Show template of a model")
1282
	showCmd.Flags().Bool("system", false, "Show system message of a model")
Patrick Devine's avatar
Patrick Devine committed
1283

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1284
	runCmd := &cobra.Command{
1285
1286
1287
1288
1289
		Use:     "run MODEL [PROMPT]",
		Short:   "Run a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    RunHandler,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1290
1291
	}

1292
	runCmd.Flags().String("keepalive", "", "Duration to keep a model loaded (e.g. 5m)")
1293
	runCmd.Flags().Bool("verbose", false, "Show timings for response")
1294
	runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
1295
	runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1296
	runCmd.Flags().String("format", "", "Response format (e.g. json)")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1297
1298
1299
1300
	serveCmd := &cobra.Command{
		Use:     "serve",
		Aliases: []string{"start"},
		Short:   "Start ollama",
Michael Yang's avatar
Michael Yang committed
1301
		Args:    cobra.ExactArgs(0),
Michael Yang's avatar
Michael Yang committed
1302
		RunE:    RunServer,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1303
1304
	}

1305
	pullCmd := &cobra.Command{
1306
1307
		Use:     "pull MODEL",
		Short:   "Pull a model from a registry",
Michael Yang's avatar
Michael Yang committed
1308
		Args:    cobra.ExactArgs(1),
1309
1310
		PreRunE: checkServerHeartbeat,
		RunE:    PullHandler,
1311
1312
	}

1313
1314
	pullCmd.Flags().Bool("insecure", false, "Use an insecure registry")

1315
	pushCmd := &cobra.Command{
1316
1317
		Use:     "push MODEL",
		Short:   "Push a model to a registry",
Michael Yang's avatar
Michael Yang committed
1318
		Args:    cobra.ExactArgs(1),
1319
1320
		PreRunE: checkServerHeartbeat,
		RunE:    PushHandler,
1321
1322
	}

1323
1324
	pushCmd.Flags().Bool("insecure", false, "Use an insecure registry")

Patrick Devine's avatar
Patrick Devine committed
1325
	listCmd := &cobra.Command{
1326
		Use:     "list",
Patrick Devine's avatar
Patrick Devine committed
1327
		Aliases: []string{"ls"},
1328
		Short:   "List models",
1329
		PreRunE: checkServerHeartbeat,
1330
		RunE:    ListHandler,
1331
	}
1332
1333
1334
1335
1336
1337
1338
1339

	psCmd := &cobra.Command{
		Use:     "ps",
		Short:   "List running models",
		PreRunE: checkServerHeartbeat,
		RunE:    ListRunningHandler,
	}

Patrick Devine's avatar
Patrick Devine committed
1340
	copyCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1341
		Use:     "cp SOURCE DESTINATION",
1342
		Short:   "Copy a model",
Michael Yang's avatar
Michael Yang committed
1343
		Args:    cobra.ExactArgs(2),
1344
1345
		PreRunE: checkServerHeartbeat,
		RunE:    CopyHandler,
Patrick Devine's avatar
Patrick Devine committed
1346
1347
	}

1348
	deleteCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1349
		Use:     "rm MODEL [MODEL...]",
1350
1351
1352
1353
		Short:   "Remove a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    DeleteHandler,
Patrick Devine's avatar
Patrick Devine committed
1354
1355
	}

1356
1357
1358
	envVars := envconfig.AsMap()

	envs := []envconfig.EnvVar{envVars["OLLAMA_HOST"]}
1359

1360
1361
1362
1363
1364
1365
1366
	for _, cmd := range []*cobra.Command{
		createCmd,
		showCmd,
		runCmd,
		pullCmd,
		pushCmd,
		listCmd,
1367
		psCmd,
1368
1369
		copyCmd,
		deleteCmd,
1370
		serveCmd,
1371
	} {
1372
1373
		switch cmd {
		case runCmd:
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
			appendEnvDocs(cmd, []envconfig.EnvVar{envVars["OLLAMA_HOST"], envVars["OLLAMA_NOHISTORY"]})
		case serveCmd:
			appendEnvDocs(cmd, []envconfig.EnvVar{
				envVars["OLLAMA_DEBUG"],
				envVars["OLLAMA_HOST"],
				envVars["OLLAMA_KEEP_ALIVE"],
				envVars["OLLAMA_MAX_LOADED_MODELS"],
				envVars["OLLAMA_MAX_QUEUE"],
				envVars["OLLAMA_MODELS"],
				envVars["OLLAMA_NUM_PARALLEL"],
				envVars["OLLAMA_NOPRUNE"],
				envVars["OLLAMA_ORIGINS"],
1386
				envVars["OLLAMA_SCHED_SPREAD"],
1387
				envVars["OLLAMA_TMPDIR"],
1388
1389
				envVars["OLLAMA_FLASH_ATTENTION"],
				envVars["OLLAMA_LLM_LIBRARY"],
1390
				envVars["OLLAMA_GPU_OVERHEAD"],
1391
				envVars["OLLAMA_LOAD_TIMEOUT"],
1392
			})
1393
1394
1395
		default:
			appendEnvDocs(cmd, envs)
		}
1396
1397
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1398
1399
	rootCmd.AddCommand(
		serveCmd,
1400
		createCmd,
Patrick Devine's avatar
Patrick Devine committed
1401
		showCmd,
1402
		runCmd,
1403
1404
		pullCmd,
		pushCmd,
Patrick Devine's avatar
Patrick Devine committed
1405
		listCmd,
1406
		psCmd,
Patrick Devine's avatar
Patrick Devine committed
1407
		copyCmd,
1408
		deleteCmd,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1409
1410
1411
1412
	)

	return rootCmd
}