cmd.go 44.6 KB
Newer Older
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1
2
3
package cmd

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

29
	"github.com/containerd/console"
30
	"github.com/mattn/go-runewidth"
Patrick Devine's avatar
Patrick Devine committed
31
	"github.com/olekukonko/tablewriter"
Michael Yang's avatar
Michael Yang committed
32
	"github.com/spf13/cobra"
33
	"golang.org/x/crypto/ssh"
34
	"golang.org/x/sync/errgroup"
35
	"golang.org/x/term"
Michael Yang's avatar
Michael Yang committed
36

37
	"github.com/ollama/ollama/api"
38
	"github.com/ollama/ollama/envconfig"
39
	"github.com/ollama/ollama/format"
40
	"github.com/ollama/ollama/parser"
41
	"github.com/ollama/ollama/progress"
42
	"github.com/ollama/ollama/readline"
Jesse Gross's avatar
Jesse Gross committed
43
	"github.com/ollama/ollama/runner"
44
	"github.com/ollama/ollama/server"
45
	"github.com/ollama/ollama/types/model"
46
	"github.com/ollama/ollama/types/syncmap"
47
	"github.com/ollama/ollama/version"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
48
49
)

50
const ConnectInstructions = "To sign in, navigate to:\n    %s\n\n"
51

52
53
54
55
56
57
58
59
60
// ensureThinkingSupport emits a warning if the model does not advertise thinking support
func ensureThinkingSupport(ctx context.Context, client *api.Client, name string) {
	if name == "" {
		return
	}
	resp, err := client.Show(ctx, &api.ShowRequest{Model: name})
	if err != nil {
		return
	}
61
62
	if slices.Contains(resp.Capabilities, model.CapabilityThinking) {
		return
63
64
65
66
	}
	fmt.Fprintf(os.Stderr, "warning: model %q does not support thinking output\n", name)
}

67
var errModelfileNotFound = errors.New("specified Modelfile wasn't found")
68
69

func getModelfileName(cmd *cobra.Command) (string, error) {
70
	filename, _ := cmd.Flags().GetString("file")
71
72
73
74
75
76

	if filename == "" {
		filename = "Modelfile"
	}

	absName, err := filepath.Abs(filename)
77
	if err != nil {
78
		return "", err
79
80
	}

81
	_, err = os.Stat(absName)
82
	if err != nil {
83
		return "", err
84
	}
85

86
87
88
89
	return absName, nil
}

func CreateHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
90
91
92
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

93
94
95
96
97
98
99
100
101
102
	var reader io.Reader

	filename, err := getModelfileName(cmd)
	if os.IsNotExist(err) {
		if filename == "" {
			reader = strings.NewReader("FROM .\n")
		} else {
			return errModelfileNotFound
		}
	} else if err != nil {
Michael Yang's avatar
Michael Yang committed
103
		return err
104
105
106
107
108
109
110
111
	} else {
		f, err := os.Open(filename)
		if err != nil {
			return err
		}

		reader = f
		defer f.Close()
Michael Yang's avatar
Michael Yang committed
112
113
	}

114
	modelfile, err := parser.ParseFile(reader)
Michael Yang's avatar
Michael Yang committed
115
116
117
118
	if err != nil {
		return err
	}

119
120
121
122
	status := "gathering model components"
	spinner := progress.NewSpinner(status)
	p.Add(status, spinner)

123
	req, err := modelfile.CreateRequest(filepath.Dir(filename))
Michael Yang's avatar
Michael Yang committed
124
125
126
	if err != nil {
		return err
	}
127
	spinner.Stop()
Michael Yang's avatar
Michael Yang committed
128

129
	req.Model = args[0]
130
131
132
133
	quantize, _ := cmd.Flags().GetString("quantize")
	if quantize != "" {
		req.Quantize = quantize
	}
134

135
136
137
138
139
	client, err := api.ClientFromEnvironment()
	if err != nil {
		return err
	}

140
141
142
143
144
145
	var g errgroup.Group
	g.SetLimit(max(runtime.GOMAXPROCS(0)-1, 1))

	files := syncmap.NewSyncMap[string, string]()
	for f, digest := range req.Files {
		g.Go(func() error {
146
			if _, err := createBlob(cmd, client, f, digest, p); err != nil {
Michael Yang's avatar
Michael Yang committed
147
148
				return err
			}
149
150
151
152
153
154
155

			// TODO: this is incorrect since the file might be in a subdirectory
			//       instead this should take the path relative to the model directory
			//       but the current implementation does not allow this
			files.Store(filepath.Base(f), digest)
			return nil
		})
156
	}
Michael Yang's avatar
Michael Yang committed
157

158
159
160
	adapters := syncmap.NewSyncMap[string, string]()
	for f, digest := range req.Adapters {
		g.Go(func() error {
161
			if _, err := createBlob(cmd, client, f, digest, p); err != nil {
Michael Yang's avatar
Michael Yang committed
162
163
				return err
			}
164
165
166
167
168

			// TODO: same here
			adapters.Store(filepath.Base(f), digest)
			return nil
		})
Michael Yang's avatar
Michael Yang committed
169
	}
Michael Yang's avatar
Michael Yang committed
170

171
172
173
174
175
176
177
	if err := g.Wait(); err != nil {
		return err
	}

	req.Files = files.Items()
	req.Adapters = adapters.Items()

Michael Yang's avatar
Michael Yang committed
178
	bars := make(map[string]*progress.Bar)
179
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
180
181
182
		if resp.Digest != "" {
			bar, ok := bars[resp.Digest]
			if !ok {
183
184
185
186
187
				msg := resp.Status
				if msg == "" {
					msg = fmt.Sprintf("pulling %s...", resp.Digest[7:19])
				}
				bar = progress.NewBar(msg, resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
188
189
190
191
192
193
194
195
196
197
198
199
200
				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)
		}

201
202
203
		return nil
	}

204
	if err := client.Create(cmd.Context(), req, fn); err != nil {
205
206
207
		if strings.Contains(err.Error(), "path or Modelfile are required") {
			return fmt.Errorf("the ollama server must be updated to use `ollama create` with this client")
		}
208
209
210
211
212
213
		return err
	}

	return nil
}

214
215
func createBlob(cmd *cobra.Command, client *api.Client, path string, digest string, p *progress.Progress) (string, error) {
	realPath, err := filepath.EvalSymlinks(path)
Michael Yang's avatar
Michael Yang committed
216
217
218
219
	if err != nil {
		return "", err
	}

220
	bin, err := os.Open(realPath)
221
222
223
224
225
	if err != nil {
		return "", err
	}
	defer bin.Close()

226
227
228
229
230
231
232
233
	// Get file info to retrieve the size
	fileInfo, err := bin.Stat()
	if err != nil {
		return "", err
	}
	fileSize := fileInfo.Size()

	var pw progressWriter
234
235
236
237
	status := fmt.Sprintf("copying file %s 0%%", digest)
	spinner := progress.NewSpinner(status)
	p.Add(status, spinner)
	defer spinner.Stop()
238
239
240
241
242
243
244
245
246
247

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

	go func() {
		ticker := time.NewTicker(60 * time.Millisecond)
		defer ticker.Stop()
		for {
			select {
			case <-ticker.C:
248
				spinner.SetMessage(fmt.Sprintf("copying file %s %d%%", digest, int(100*pw.n.Load()/fileSize)))
249
			case <-done:
250
				spinner.SetMessage(fmt.Sprintf("copying file %s 100%%", digest))
251
252
253
254
255
				return
			}
		}
	}()

256
	if err := client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
257
258
259
260
261
		return "", err
	}
	return digest, nil
}

262
263
264
265
266
267
268
269
270
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
}

Patrick Devine's avatar
Patrick Devine committed
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
func loadOrUnloadModel(cmd *cobra.Command, opts *runOptions) error {
	p := progress.NewProgress(os.Stderr)
	defer p.StopAndClear()

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

	client, err := api.ClientFromEnvironment()
	if err != nil {
		return err
	}

	req := &api.GenerateRequest{
		Model:     opts.Model,
		KeepAlive: opts.KeepAlive,
286
287
288

		// pass Think here so we fail before getting to the chat prompt if the model doesn't support it
		Think: opts.Think,
Patrick Devine's avatar
Patrick Devine committed
289
290
	}

291
292
293
294
295
296
297
298
299
300
301
	return client.Generate(cmd.Context(), req, func(r api.GenerateResponse) error {
		if r.RemoteModel != "" && opts.ShowConnect {
			p.StopAndClear()
			if strings.HasPrefix(r.RemoteHost, "https://ollama.com") {
				fmt.Fprintf(os.Stderr, "Connecting to '%s' on 'ollama.com' ⚡\n", r.RemoteModel)
			} else {
				fmt.Fprintf(os.Stderr, "Connecting to '%s' on '%s'\n", r.RemoteModel, r.RemoteHost)
			}
		}
		return nil
	})
Patrick Devine's avatar
Patrick Devine committed
302
303
304
305
306
307
308
309
310
311
312
}

func StopHandler(cmd *cobra.Command, args []string) error {
	opts := &runOptions{
		Model:     args[0],
		KeepAlive: &api.Duration{Duration: 0},
	}
	if err := loadOrUnloadModel(cmd, opts); err != nil {
		if strings.Contains(err.Error(), "not found") {
			return fmt.Errorf("couldn't find model \"%s\" to stop", args[0])
		}
313
		return err
Patrick Devine's avatar
Patrick Devine committed
314
315
316
317
	}
	return nil
}

318
func RunHandler(cmd *cobra.Command, args []string) error {
319
320
321
	interactive := true

	opts := runOptions{
322
323
324
325
		Model:       args[0],
		WordWrap:    os.Getenv("TERM") == "xterm-256color",
		Options:     map[string]any{},
		ShowConnect: true,
326
327
328
329
330
331
332
333
	}

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

334
335
	thinkFlag := cmd.Flags().Lookup("think")
	if thinkFlag.Changed {
Michael Yang's avatar
Michael Yang committed
336
		thinkStr, err := cmd.Flags().GetString("think")
337
338
339
		if err != nil {
			return err
		}
Michael Yang's avatar
Michael Yang committed
340
341
342
343
344
345
346
347
348
349
350
351
352

		// Handle different values for --think
		switch thinkStr {
		case "", "true":
			// --think or --think=true
			opts.Think = &api.ThinkValue{Value: true}
		case "false":
			opts.Think = &api.ThinkValue{Value: false}
		case "high", "medium", "low":
			opts.Think = &api.ThinkValue{Value: thinkStr}
		default:
			return fmt.Errorf("invalid value for --think: %q (must be true, false, high, medium, or low)", thinkStr)
		}
353
354
355
356
357
358
359
360
361
	} else {
		opts.Think = nil
	}
	hidethinking, err := cmd.Flags().GetBool("hidethinking")
	if err != nil {
		return err
	}
	opts.HideThinking = hidethinking

362
363
364
365
366
367
368
369
370
371
372
373
	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}
	}

374
375
376
377
378
379
380
381
382
	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...)
383
		opts.ShowConnect = false
384
385
386
387
388
389
390
		opts.WordWrap = false
		interactive = false
	}
	opts.Prompt = strings.Join(prompts, " ")
	if len(prompts) > 0 {
		interactive = false
	}
391
392
393
394
	// Be quiet if we're redirecting to a pipe or file
	if !term.IsTerminal(int(os.Stdout.Fd())) {
		interactive = false
	}
395
396
397
398
399
400
401

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

402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
	// 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
424
425
	}

426
427
428
429
430
	opts.Think, err = inferThinkingOption(&info.Capabilities, &opts, thinkFlag.Changed)
	if err != nil {
		return err
	}

431
432
433
434
435
	opts.MultiModal = slices.Contains(info.Capabilities, model.CapabilityVision)

	// TODO: remove the projector info and vision info checks below,
	// these are left in for backwards compatibility with older servers
	// that don't have the capabilities field in the model info
436
437
438
439
440
441
442
443
444
445
	if len(info.ProjectorInfo) != 0 {
		opts.MultiModal = true
	}
	for k := range info.ModelInfo {
		if strings.Contains(k, ".vision.") {
			opts.MultiModal = true
			break
		}
	}

446
447
448
	opts.ParentModel = info.Details.ParentModel

	if interactive {
Patrick Devine's avatar
Patrick Devine committed
449
		if err := loadOrUnloadModel(cmd, &opts); err != nil {
450
451
			var sErr api.AuthorizationError
			if errors.As(err, &sErr) && sErr.StatusCode == http.StatusUnauthorized {
452
453
454
455
				fmt.Printf("You need to be signed in to Ollama to run Cloud models.\n\n")

				if sErr.SigninURL != "" {
					fmt.Printf(ConnectInstructions, sErr.SigninURL)
456
457
458
				}
				return nil
			}
Michael Yang's avatar
Michael Yang committed
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
			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()
			}
		}

474
475
476
		return generateInteractive(cmd, opts)
	}
	return generate(cmd, opts)
Bruce MacDonald's avatar
Bruce MacDonald committed
477
478
}

479
480
481
482
483
484
485
486
func SigninHandler(cmd *cobra.Command, args []string) error {
	client, err := api.ClientFromEnvironment()
	if err != nil {
		return err
	}

	user, err := client.Whoami(cmd.Context())
	if err != nil {
487
488
489
490
491
492
493
494
495
496
		var aErr api.AuthorizationError
		if errors.As(err, &aErr) && aErr.StatusCode == http.StatusUnauthorized {
			fmt.Println("You need to be signed in to Ollama to run Cloud models.")
			fmt.Println()

			if aErr.SigninURL != "" {
				fmt.Printf(ConnectInstructions, aErr.SigninURL)
			}
			return nil
		}
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
		return err
	}

	if user != nil && user.Name != "" {
		fmt.Printf("You are already signed in as user '%s'\n", user.Name)
		fmt.Println()
		return nil
	}

	return nil
}

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

515
	err = client.Signout(cmd.Context())
516
	if err != nil {
517
518
519
520
521
522
523
524
		var aErr api.AuthorizationError
		if errors.As(err, &aErr) && aErr.StatusCode == http.StatusUnauthorized {
			fmt.Println("You are not signed in to ollama.com")
			fmt.Println()
			return nil
		} else {
			return err
		}
525
	}
526

527
528
529
530
531
	fmt.Println("You have signed out of ollama.com")
	fmt.Println()
	return nil
}

532
func PushHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
533
	client, err := api.ClientFromEnvironment()
534
535
536
	if err != nil {
		return err
	}
537

538
539
540
541
542
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
	n := model.ParseName(args[0])
	if strings.HasSuffix(n.Host, ".ollama.ai") || strings.HasSuffix(n.Host, ".ollama.com") {
		_, err := client.Whoami(cmd.Context())
		if err != nil {
			var aErr api.AuthorizationError
			if errors.As(err, &aErr) && aErr.StatusCode == http.StatusUnauthorized {
				fmt.Println("You need to be signed in to push models to ollama.com.")
				fmt.Println()

				if aErr.SigninURL != "" {
					fmt.Printf(ConnectInstructions, aErr.SigninURL)
				}
				return nil
			}

			return err
		}
	}

Michael Yang's avatar
Michael Yang committed
562
563
564
565
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

	bars := make(map[string]*progress.Bar)
566
567
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
568

569
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
570
		if resp.Digest != "" {
571
572
573
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
574
575
576

			bar, ok := bars[resp.Digest]
			if !ok {
577
				bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
578
579
580
581
582
583
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
584
585
586
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
587
588
589
590
591
592

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

593
594
595
		return nil
	}

Michael Yang's avatar
Michael Yang committed
596
	request := api.PushRequest{Name: args[0], Insecure: insecure}
597

Michael Yang's avatar
Michael Yang committed
598
	if err := client.Push(cmd.Context(), &request, fn); err != nil {
599
600
601
		if spinner != nil {
			spinner.Stop()
		}
602
603
		errStr := strings.ToLower(err.Error())
		if strings.Contains(errStr, "access denied") || strings.Contains(errStr, "unauthorized") {
604
605
			return errors.New("you are not authorized to push to this namespace, create the model under a namespace you own")
		}
Michael Yang's avatar
Michael Yang committed
606
607
608
		return err
	}

609
	p.Stop()
610
	spinner.Stop()
611
612
613
614
615
616
617
618

	destination := n.String()
	if strings.HasSuffix(n.Host, ".ollama.ai") || strings.HasSuffix(n.Host, ".ollama.com") {
		destination = "https://ollama.com/" + strings.TrimSuffix(n.DisplayShortest(), ":latest")
	}
	fmt.Printf("\nYou can find your model at:\n\n")
	fmt.Printf("\t%s\n", destination)

Michael Yang's avatar
Michael Yang committed
619
	return nil
620
621
}

622
func ListHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
623
	client, err := api.ClientFromEnvironment()
624
625
626
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
627

Michael Yang's avatar
Michael Yang committed
628
	models, err := client.List(cmd.Context())
Patrick Devine's avatar
Patrick Devine committed
629
630
631
632
633
634
635
	if err != nil {
		return err
	}

	var data [][]string

	for _, m := range models.Models {
636
		if len(args) == 0 || strings.HasPrefix(strings.ToLower(m.Name), strings.ToLower(args[0])) {
637
638
639
640
641
642
643
644
			var size string
			if m.RemoteModel != "" {
				size = "-"
			} else {
				size = format.HumanBytes(m.Size)
			}

			data = append(data, []string{m.Name, m.Digest[:12], size, format.HumanTime(m.ModifiedAt, "Never")})
Michael Yang's avatar
Michael Yang committed
645
		}
Patrick Devine's avatar
Patrick Devine committed
646
647
648
	}

	table := tablewriter.NewWriter(os.Stdout)
Patrick Devine's avatar
Patrick Devine committed
649
	table.SetHeader([]string{"NAME", "ID", "SIZE", "MODIFIED"})
Patrick Devine's avatar
Patrick Devine committed
650
651
652
653
654
	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
655
	table.SetTablePadding("    ")
Patrick Devine's avatar
Patrick Devine committed
656
657
658
659
660
661
	table.AppendBulk(data)
	table.Render()

	return nil
}

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
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))
			}
Patrick Devine's avatar
Patrick Devine committed
690
691
692
693
694
695
696
697

			var until string
			delta := time.Since(m.ExpiresAt)
			if delta > 0 {
				until = "Stopping..."
			} else {
				until = format.HumanTime(m.ExpiresAt, "Never")
			}
698
699
			ctxStr := strconv.Itoa(m.ContextLength)
			data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), procStr, ctxStr, until})
700
701
702
703
		}
	}

	table := tablewriter.NewWriter(os.Stdout)
704
	table.SetHeader([]string{"NAME", "ID", "SIZE", "PROCESSOR", "CONTEXT", "UNTIL"})
705
706
707
708
709
	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
710
	table.SetTablePadding("    ")
711
712
713
714
715
716
	table.AppendBulk(data)
	table.Render()

	return nil
}

717
func DeleteHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
718
	client, err := api.ClientFromEnvironment()
719
720
721
	if err != nil {
		return err
	}
722

723
724
725
726
727
728
	// Unload the model if it's running before deletion
	opts := &runOptions{
		Model:     args[0],
		KeepAlive: &api.Duration{Duration: 0},
	}
	if err := loadOrUnloadModel(cmd, opts); err != nil {
729
730
		if !strings.Contains(strings.ToLower(err.Error()), "not found") {
			fmt.Fprintf(os.Stderr, "Warning: unable to stop model '%s'\n", args[0])
731
732
733
		}
	}

734
735
	for _, name := range args {
		req := api.DeleteRequest{Name: name}
Michael Yang's avatar
Michael Yang committed
736
		if err := client.Delete(cmd.Context(), &req); err != nil {
737
738
739
			return err
		}
		fmt.Printf("deleted '%s'\n", name)
740
741
742
743
	}
	return nil
}

Patrick Devine's avatar
Patrick Devine committed
744
func ShowHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
745
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
746
747
748
749
750
751
752
753
754
	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")
755
	verbose, errVerbose := cmd.Flags().GetBool("verbose")
Patrick Devine's avatar
Patrick Devine committed
756

757
	for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate, errVerbose} {
Patrick Devine's avatar
Patrick Devine committed
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
		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 {
792
		return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
793
794
	}

795
	req := api.ShowRequest{Name: args[0], Verbose: verbose}
796
797
798
799
	resp, err := client.Show(cmd.Context(), &req)
	if err != nil {
		return err
	}
800

801
	if flagsSet == 1 {
802
803
804
805
806
807
808
809
		switch showType {
		case "license":
			fmt.Println(resp.License)
		case "modelfile":
			fmt.Println(resp.Modelfile)
		case "parameters":
			fmt.Println(resp.Parameters)
		case "system":
810
			fmt.Print(resp.System)
811
		case "template":
812
			fmt.Print(resp.Template)
813
814
815
		}

		return nil
Patrick Devine's avatar
Patrick Devine committed
816
817
	}

818
	return showInfo(resp, verbose, os.Stdout)
819
820
}

821
func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
Michael Yang's avatar
Michael Yang committed
822
823
824
825
826
827
828
	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("    ")
829

Michael Yang's avatar
Michael Yang committed
830
831
832
		switch header {
		case "Template", "System", "License":
			table.SetColWidth(100)
833
834
		}

Michael Yang's avatar
Michael Yang committed
835
836
837
		table.AppendBulk(rows())
		table.Render()
		fmt.Fprintln(w)
Patrick Devine's avatar
Patrick Devine committed
838
839
	}

Michael Yang's avatar
Michael Yang committed
840
	tableRender("Model", func() (rows [][]string) {
841
842
843
844
845
		if resp.RemoteHost != "" {
			rows = append(rows, []string{"", "Remote model", resp.RemoteModel})
			rows = append(rows, []string{"", "Remote URL", resp.RemoteHost})
		}

Michael Yang's avatar
Michael Yang committed
846
847
848
		if resp.ModelInfo != nil {
			arch := resp.ModelInfo["general.architecture"].(string)
			rows = append(rows, []string{"", "architecture", arch})
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870

			var paramStr string
			if resp.Details.ParameterSize != "" {
				paramStr = resp.Details.ParameterSize
			} else if v, ok := resp.ModelInfo["general.parameter_count"]; ok {
				if f, ok := v.(float64); ok {
					paramStr = format.HumanNumber(uint64(f))
				}
			}
			rows = append(rows, []string{"", "parameters", paramStr})

			if v, ok := resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)]; ok {
				if f, ok := v.(float64); ok {
					rows = append(rows, []string{"", "context length", strconv.FormatFloat(f, 'f', -1, 64)})
				}
			}

			if v, ok := resp.ModelInfo[fmt.Sprintf("%s.embedding_length", arch)]; ok {
				if f, ok := v.(float64); ok {
					rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(f, 'f', -1, 64)})
				}
			}
Michael Yang's avatar
Michael Yang committed
871
872
873
874
875
876
877
		} 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
	})
878

879
880
881
882
883
884
885
886
887
	if len(resp.Capabilities) > 0 {
		tableRender("Capabilities", func() (rows [][]string) {
			for _, capability := range resp.Capabilities {
				rows = append(rows, []string{"", capability.String()})
			}
			return
		})
	}

Michael Yang's avatar
Michael Yang committed
888
889
890
891
892
893
894
895
896
	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
		})
897
898
	}

Michael Yang's avatar
Michael Yang committed
899
900
901
902
903
904
905
906
907
908
	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
		})
909
910
	}

911
912
913
914
915
916
917
918
919
920
921
	if resp.ModelInfo != nil && verbose {
		tableRender("Metadata", func() (rows [][]string) {
			keys := make([]string, 0, len(resp.ModelInfo))
			for k := range resp.ModelInfo {
				keys = append(keys, k)
			}
			sort.Strings(keys)

			for _, k := range keys {
				var v string
				switch vData := resp.ModelInfo[k].(type) {
922
923
				case bool:
					v = fmt.Sprintf("%t", vData)
924
925
926
927
928
				case string:
					v = vData
				case float64:
					v = fmt.Sprintf("%g", vData)
				case []any:
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
					targetWidth := 10 // Small width where we are displaying the data in a column

					var itemsToShow int
					totalWidth := 1 // Start with 1 for opening bracket

					// Find how many we can fit
					for i := range vData {
						itemStr := fmt.Sprintf("%v", vData[i])
						width := runewidth.StringWidth(itemStr)

						// Add separator width (", ") for all items except the first
						if i > 0 {
							width += 2
						}

						// Check if adding this item would exceed our width limit
						if totalWidth+width > targetWidth && i > 0 {
							break
						}

						totalWidth += width
						itemsToShow++
					}

					// Format the output
					if itemsToShow < len(vData) {
						v = fmt.Sprintf("%v", vData[:itemsToShow])
						v = strings.TrimSuffix(v, "]")
						v += fmt.Sprintf(" ...+%d more]", len(vData)-itemsToShow)
					} else {
						v = fmt.Sprintf("%v", vData)
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
					}
				default:
					v = fmt.Sprintf("%T", vData)
				}
				rows = append(rows, []string{"", k, v})
			}
			return
		})
	}

	if len(resp.Tensors) > 0 && verbose {
		tableRender("Tensors", func() (rows [][]string) {
			for _, t := range resp.Tensors {
				rows = append(rows, []string{"", t.Name, t.Type, fmt.Sprint(t.Shape)})
			}
			return
		})
	}

Michael Yang's avatar
Michael Yang committed
979
980
	head := func(s string, n int) (rows [][]string) {
		scanner := bufio.NewScanner(strings.NewReader(s))
981
982
983
984
985
		count := 0
		for scanner.Scan() {
			text := strings.TrimSpace(scanner.Text())
			if text == "" {
				continue
986
			}
987
988
989
990
991
992
993
			count++
			if n < 0 || count <= n {
				rows = append(rows, []string{"", text})
			}
		}
		if n >= 0 && count > n {
			rows = append(rows, []string{"", "..."})
994
		}
Michael Yang's avatar
Michael Yang committed
995
		return
996
997
	}

Michael Yang's avatar
Michael Yang committed
998
999
1000
1001
1002
	if resp.System != "" {
		tableRender("System", func() [][]string {
			return head(resp.System, 2)
		})
	}
1003

Michael Yang's avatar
Michael Yang committed
1004
1005
1006
1007
	if resp.License != "" {
		tableRender("License", func() [][]string {
			return head(resp.License, 2)
		})
1008
	}
Michael Yang's avatar
Michael Yang committed
1009
1010

	return nil
1011
1012
}

Patrick Devine's avatar
Patrick Devine committed
1013
func CopyHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
1014
	client, err := api.ClientFromEnvironment()
1015
1016
1017
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
1018
1019

	req := api.CopyRequest{Source: args[0], Destination: args[1]}
Michael Yang's avatar
Michael Yang committed
1020
	if err := client.Copy(cmd.Context(), &req); err != nil {
Patrick Devine's avatar
Patrick Devine committed
1021
1022
1023
1024
1025
1026
		return err
	}
	fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
	return nil
}

1027
func PullHandler(cmd *cobra.Command, args []string) error {
1028
1029
1030
1031
1032
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
1033
	client, err := api.ClientFromEnvironment()
1034
1035
1036
	if err != nil {
		return err
	}
1037

Michael Yang's avatar
Michael Yang committed
1038
1039
1040
1041
1042
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

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

1043
1044
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
1045

1046
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
1047
		if resp.Digest != "" {
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
			if resp.Completed == 0 {
				// This is the initial status update for the
				// layer, which the server sends before
				// beginning the download, for clients to
				// compute total size and prepare for
				// downloads, if needed.
				//
				// Skipping this here to avoid showing a 0%
				// progress bar, which *should* clue the user
				// into the fact that many things are being
				// downloaded and that the current active
				// download is not that last. However, in rare
				// cases it seems to be triggering to some, and
				// it isn't worth explaining, so just ignore
				// and regress to the old UI that keeps giving
				// you the "But wait, there is more!" after
				// each "100% done" bar, which is "better."
				return nil
			}

1068
1069
1070
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
1071
1072
1073

			bar, ok := bars[resp.Digest]
			if !ok {
1074
1075
1076
1077
1078
1079
				name, isDigest := strings.CutPrefix(resp.Digest, "sha256:")
				name = strings.TrimSpace(name)
				if isDigest {
					name = name[:min(12, len(name))]
				}
				bar = progress.NewBar(fmt.Sprintf("pulling %s:", name), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
1080
1081
1082
1083
1084
1085
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
1086
1087
1088
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
1089
1090
1091
1092
1093
1094

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

1095
1096
		return nil
	}
1097

Michael Yang's avatar
Michael Yang committed
1098
	request := api.PullRequest{Name: args[0], Insecure: insecure}
1099
	return client.Pull(cmd.Context(), &request, fn)
Michael Yang's avatar
Michael Yang committed
1100
1101
}

1102
1103
type generateContextKey string

1104
type runOptions struct {
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
	Model        string
	ParentModel  string
	Prompt       string
	Messages     []api.Message
	WordWrap     bool
	Format       string
	System       string
	Images       []api.ImageData
	Options      map[string]any
	MultiModal   bool
	KeepAlive    *api.Duration
Michael Yang's avatar
Michael Yang committed
1116
	Think        *api.ThinkValue
1117
	HideThinking bool
1118
	ShowConnect  bool
1119
1120
}

1121
1122
1123
1124
1125
1126
1127
1128
1129
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
1130
1131
			if state.lineLength+1 > termWidth-5 {
				if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
1132
1133
1134
1135
1136
1137
1138
					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
1139
1140
				a := runewidth.StringWidth(state.wordBuffer)
				if a > 0 {
1141
					fmt.Printf("\x1b[%dD", a)
1142
1143
				}
				fmt.Printf("\x1b[K\n")
1144
				fmt.Printf("%s%c", state.wordBuffer, ch)
1145
1146
1147
				chWidth := runewidth.RuneWidth(ch)

				state.lineLength = runewidth.StringWidth(state.wordBuffer) + chWidth
1148
1149
			} else {
				fmt.Print(string(ch))
1150
1151
1152
1153
				state.lineLength += runewidth.RuneWidth(ch)
				if runewidth.RuneWidth(ch) >= 2 {
					state.wordBuffer = ""
					continue
Josh Yan's avatar
Josh Yan committed
1154
				}
1155
1156

				switch ch {
Michael Yang's avatar
Michael Yang committed
1157
				case ' ', '\t':
1158
					state.wordBuffer = ""
Michael Yang's avatar
Michael Yang committed
1159
				case '\n', '\r':
1160
					state.lineLength = 0
Michael Yang's avatar
Michael Yang committed
1161
					state.wordBuffer = ""
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
				default:
					state.wordBuffer += string(ch)
				}
			}
		}
	} else {
		fmt.Printf("%s%s", state.wordBuffer, content)
		if len(state.wordBuffer) > 0 {
			state.wordBuffer = ""
		}
	}
}

1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
func thinkingOutputOpeningText(plainText bool) string {
	text := "Thinking...\n"

	if plainText {
		return text
	}

	return readline.ColorGrey + readline.ColorBold + text + readline.ColorDefault + readline.ColorGrey
}

func thinkingOutputClosingText(plainText bool) string {
	text := "...done thinking.\n\n"

	if plainText {
		return text
	}

	return readline.ColorGrey + readline.ColorBold + text + readline.ColorDefault
}

1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
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{}
Michael Yang's avatar
Michael Yang committed
1219
	var thinkingContent strings.Builder
1220
1221
	var latest api.ChatResponse
	var fullResponse strings.Builder
1222
1223
	var thinkTagOpened bool = false
	var thinkTagClosed bool = false
1224

1225
1226
	role := "assistant"

1227
	fn := func(response api.ChatResponse) error {
1228
1229
1230
		if response.Message.Content != "" || !opts.HideThinking {
			p.StopAndClear()
		}
1231
1232
1233
1234

		latest = response

		role = response.Message.Role
1235
1236
1237
1238
		if response.Message.Thinking != "" && !opts.HideThinking {
			if !thinkTagOpened {
				fmt.Print(thinkingOutputOpeningText(false))
				thinkTagOpened = true
Michael Yang's avatar
Michael Yang committed
1239
				thinkTagClosed = false
1240
			}
Michael Yang's avatar
Michael Yang committed
1241
			thinkingContent.WriteString(response.Message.Thinking)
1242
1243
1244
			displayResponse(response.Message.Thinking, opts.WordWrap, state)
		}

1245
		content := response.Message.Content
Michael Yang's avatar
Michael Yang committed
1246
1247
1248
1249
		if thinkTagOpened && !thinkTagClosed && (content != "" || len(response.Message.ToolCalls) > 0) {
			if !strings.HasSuffix(thinkingContent.String(), "\n") {
				fmt.Println()
			}
1250
			fmt.Print(thinkingOutputClosingText(false))
Michael Yang's avatar
Michael Yang committed
1251
			thinkTagOpened = false
1252
			thinkTagClosed = true
Michael Yang's avatar
Michael Yang committed
1253
			state = &displayResponseState{}
1254
1255
1256
1257
1258
		}
		// purposefully not putting thinking blocks in the response, which would
		// only be needed if we later added tool calling to the cli (they get
		// filtered out anyway since current models don't expect them unless you're
		// about to finish some tool calls)
1259
1260
		fullResponse.WriteString(content)

Michael Yang's avatar
Michael Yang committed
1261
1262
1263
1264
1265
1266
1267
		if response.Message.ToolCalls != nil {
			toolCalls := response.Message.ToolCalls
			if len(toolCalls) > 0 {
				fmt.Print(renderToolCalls(toolCalls, false))
			}
		}

1268
1269
1270
1271
1272
		displayResponse(content, opts.WordWrap, state)

		return nil
	}

1273
1274
1275
1276
	if opts.Format == "json" {
		opts.Format = `"` + opts.Format + `"`
	}

1277
1278
1279
	req := &api.ChatRequest{
		Model:    opts.Model,
		Messages: opts.Messages,
1280
		Format:   json.RawMessage(opts.Format),
1281
		Options:  opts.Options,
1282
		Think:    opts.Think,
1283
1284
	}

1285
1286
1287
1288
	if opts.KeepAlive != nil {
		req.KeepAlive = opts.KeepAlive
	}

1289
1290
1291
1292
	if err := client.Chat(cancelCtx, req, fn); err != nil {
		if errors.Is(err, context.Canceled) {
			return nil, nil
		}
1293
1294
1295
1296
1297
1298
1299
1300

		// this error should ideally be wrapped properly by the client
		if strings.Contains(err.Error(), "upstream error") {
			p.StopAndClear()
			fmt.Println("An error occurred while processing your message. Please try again.")
			fmt.Println()
			return nil, nil
		}
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
		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
1322
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
1323
	if err != nil {
1324
		return err
Patrick Devine's avatar
Patrick Devine committed
1325
	}
Michael Yang's avatar
Michael Yang committed
1326

Michael Yang's avatar
Michael Yang committed
1327
	p := progress.NewProgress(os.Stderr)
1328
	defer p.StopAndClear()
1329

Michael Yang's avatar
Michael Yang committed
1330
1331
1332
	spinner := progress.NewSpinner("")
	p.Add("", spinner)

1333
1334
1335
1336
1337
1338
1339
	var latest api.GenerateResponse

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

Michael Yang's avatar
Michael Yang committed
1340
	ctx, cancel := context.WithCancel(cmd.Context())
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
	defer cancel()

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

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

1351
	var state *displayResponseState = &displayResponseState{}
Michael Yang's avatar
Michael Yang committed
1352
	var thinkingContent strings.Builder
1353
1354
	var thinkTagOpened bool = false
	var thinkTagClosed bool = false
1355

1356
	plainText := !term.IsTerminal(int(os.Stdout.Fd()))
1357

1358
	fn := func(response api.GenerateResponse) error {
Patrick Devine's avatar
Patrick Devine committed
1359
		latest = response
1360
		content := response.Response
1361

1362
1363
1364
1365
1366
1367
1368
1369
		if response.Response != "" || !opts.HideThinking {
			p.StopAndClear()
		}

		if response.Thinking != "" && !opts.HideThinking {
			if !thinkTagOpened {
				fmt.Print(thinkingOutputOpeningText(plainText))
				thinkTagOpened = true
Michael Yang's avatar
Michael Yang committed
1370
				thinkTagClosed = false
1371
			}
Michael Yang's avatar
Michael Yang committed
1372
			thinkingContent.WriteString(response.Thinking)
1373
1374
1375
			displayResponse(response.Thinking, opts.WordWrap, state)
		}

Michael Yang's avatar
Michael Yang committed
1376
1377
1378
1379
		if thinkTagOpened && !thinkTagClosed && (content != "" || len(response.ToolCalls) > 0) {
			if !strings.HasSuffix(thinkingContent.String(), "\n") {
				fmt.Println()
			}
1380
			fmt.Print(thinkingOutputClosingText(plainText))
Michael Yang's avatar
Michael Yang committed
1381
			thinkTagOpened = false
1382
			thinkTagClosed = true
Michael Yang's avatar
Michael Yang committed
1383
			state = &displayResponseState{}
1384
1385
		}

1386
		displayResponse(content, opts.WordWrap, state)
1387

Michael Yang's avatar
Michael Yang committed
1388
1389
1390
1391
1392
1393
1394
		if response.ToolCalls != nil {
			toolCalls := response.ToolCalls
			if len(toolCalls) > 0 {
				fmt.Print(renderToolCalls(toolCalls, plainText))
			}
		}

Patrick Devine's avatar
Patrick Devine committed
1395
1396
		return nil
	}
1397

1398
1399
1400
1401
1402
1403
1404
	if opts.MultiModal {
		opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
		if err != nil {
			return err
		}
	}

1405
1406
1407
1408
	if opts.Format == "json" {
		opts.Format = `"` + opts.Format + `"`
	}

Michael Yang's avatar
Michael Yang committed
1409
	request := api.GenerateRequest{
1410
1411
1412
1413
		Model:     opts.Model,
		Prompt:    opts.Prompt,
		Context:   generateContext,
		Images:    opts.Images,
1414
		Format:    json.RawMessage(opts.Format),
1415
1416
1417
		System:    opts.System,
		Options:   opts.Options,
		KeepAlive: opts.KeepAlive,
1418
		Think:     opts.Think,
Michael Yang's avatar
Michael Yang committed
1419
1420
1421
	}

	if err := client.Generate(ctx, &request, fn); err != nil {
1422
		if errors.Is(err, context.Canceled) {
1423
			return nil
1424
		}
1425
		return err
Patrick Devine's avatar
Patrick Devine committed
1426
	}
1427

1428
	if opts.Prompt != "" {
Michael Yang's avatar
Michael Yang committed
1429
1430
		fmt.Println()
		fmt.Println()
Patrick Devine's avatar
Patrick Devine committed
1431
	}
1432

1433
1434
1435
1436
	if !latest.Done {
		return nil
	}

Patrick Devine's avatar
Patrick Devine committed
1437
1438
	verbose, err := cmd.Flags().GetBool("verbose")
	if err != nil {
1439
		return err
Patrick Devine's avatar
Patrick Devine committed
1440
	}
Michael Yang's avatar
Michael Yang committed
1441

Patrick Devine's avatar
Patrick Devine committed
1442
1443
	if verbose {
		latest.Summary()
Michael Yang's avatar
Michael Yang committed
1444
	}
Michael Yang's avatar
Michael Yang committed
1445

Patrick Devine's avatar
Patrick Devine committed
1446
1447
1448
	ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
	cmd.SetContext(ctx)

1449
	return nil
Michael Yang's avatar
Michael Yang committed
1450
1451
}

1452
func RunServer(_ *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1453
	if err := initializeKeypair(); err != nil {
1454
1455
1456
		return err
	}

Michael Yang's avatar
host  
Michael Yang committed
1457
	ln, err := net.Listen("tcp", envconfig.Host().Host)
1458
1459
1460
	if err != nil {
		return err
	}
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1461

1462
1463
1464
1465
1466
1467
	err = server.Serve(ln)
	if errors.Is(err, http.ErrServerClosed) {
		return nil
	}

	return err
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1468
1469
}

1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
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
1482
		cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
1483
1484
1485
1486
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1487
		privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
1488
1489
1490
1491
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1492
		if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
1493
1494
1495
			return fmt.Errorf("could not create directory %w", err)
		}

Michael Yang's avatar
Michael Yang committed
1496
		if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
1497
1498
1499
			return err
		}

Michael Yang's avatar
Michael Yang committed
1500
		sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
1501
1502
1503
1504
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1505
		publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
1506

Michael Yang's avatar
Michael Yang committed
1507
		if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
1508
1509
1510
			return err
		}

Michael Yang's avatar
Michael Yang committed
1511
		fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
1512
1513
1514
1515
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1516
func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1517
	client, err := api.ClientFromEnvironment()
1518
1519
1520
	if err != nil {
		return err
	}
Michael Yang's avatar
Michael Yang committed
1521
	if err := client.Heartbeat(cmd.Context()); err != nil {
1522
		if !(strings.Contains(err.Error(), " refused") || strings.Contains(err.Error(), "could not connect")) {
Bruce MacDonald's avatar
Bruce MacDonald committed
1523
1524
			return err
		}
1525
		if err := startApp(cmd.Context(), client); err != nil {
1526
			return fmt.Errorf("ollama server not responding - %w", err)
1527
1528
1529
1530
1531
		}
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1532
1533
1534
1535
1536
1537
1538
1539
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
1540
1541
1542
1543
1544
		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
1545
1546
	}

1547
	if serverVersion != version.Version {
Michael Yang's avatar
Michael Yang committed
1548
		fmt.Printf("Warning: client version is %s\n", version.Version)
1549
	}
Michael Yang's avatar
Michael Yang committed
1550
1551
}

1552
func appendEnvDocs(cmd *cobra.Command, envs []envconfig.EnvVar) {
1553
1554
1555
1556
1557
	if len(envs) == 0 {
		return
	}

	envUsage := `
1558
1559
Environment Variables:
`
1560
	for _, e := range envs {
1561
		envUsage += fmt.Sprintf("      %-24s   %s\n", e.Name, e.Description)
1562
1563
1564
	}

	cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
1565
1566
}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1567
1568
func NewCLI() *cobra.Command {
	log.SetFlags(log.LstdFlags | log.Lshortfile)
Michael Yang's avatar
Michael Yang committed
1569
	cobra.EnableCommandSorting = false
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1570

1571
	if runtime.GOOS == "windows" && term.IsTerminal(int(os.Stdout.Fd())) {
1572
		console.ConsoleFromFile(os.Stdin) //nolint:errcheck
1573
1574
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1575
	rootCmd := &cobra.Command{
1576
1577
1578
1579
		Use:           "ollama",
		Short:         "Large language model runner",
		SilenceUsage:  true,
		SilenceErrors: true,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1580
1581
1582
		CompletionOptions: cobra.CompletionOptions{
			DisableDefaultCmd: true,
		},
Michael Yang's avatar
Michael Yang committed
1583
1584
1585
1586
1587
1588
1589
1590
		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
1591
1592
	}

Michael Yang's avatar
Michael Yang committed
1593
	rootCmd.Flags().BoolP("version", "v", false, "Show version information")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1594

1595
	createCmd := &cobra.Command{
1596
		Use:     "create MODEL",
1597
		Short:   "Create a model",
Michael Yang's avatar
Michael Yang committed
1598
		Args:    cobra.ExactArgs(1),
1599
1600
		PreRunE: checkServerHeartbeat,
		RunE:    CreateHandler,
1601
1602
	}

1603
	createCmd.Flags().StringP("file", "f", "", "Name of the Modelfile (default \"Modelfile\")")
1604
	createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_K_M)")
1605

Patrick Devine's avatar
Patrick Devine committed
1606
1607
1608
	showCmd := &cobra.Command{
		Use:     "show MODEL",
		Short:   "Show information for a model",
Michael Yang's avatar
Michael Yang committed
1609
		Args:    cobra.ExactArgs(1),
Patrick Devine's avatar
Patrick Devine committed
1610
1611
1612
1613
1614
1615
1616
1617
		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")
1618
	showCmd.Flags().Bool("system", false, "Show system message of a model")
1619
	showCmd.Flags().BoolP("verbose", "v", false, "Show detailed model information")
Patrick Devine's avatar
Patrick Devine committed
1620

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1621
	runCmd := &cobra.Command{
1622
1623
1624
1625
1626
		Use:     "run MODEL [PROMPT]",
		Short:   "Run a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    RunHandler,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1627
1628
	}

1629
	runCmd.Flags().String("keepalive", "", "Duration to keep a model loaded (e.g. 5m)")
1630
	runCmd.Flags().Bool("verbose", false, "Show timings for response")
1631
	runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
1632
	runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1633
	runCmd.Flags().String("format", "", "Response format (e.g. json)")
Michael Yang's avatar
Michael Yang committed
1634
1635
	runCmd.Flags().String("think", "", "Enable thinking mode: true/false or high/medium/low for supported models")
	runCmd.Flags().Lookup("think").NoOptDefVal = "true"
1636
	runCmd.Flags().Bool("hidethinking", false, "Hide thinking output (if provided)")
Patrick Devine's avatar
Patrick Devine committed
1637
1638
1639
1640
1641
1642
1643
1644
1645

	stopCmd := &cobra.Command{
		Use:     "stop MODEL",
		Short:   "Stop a running model",
		Args:    cobra.ExactArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    StopHandler,
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1646
1647
1648
1649
	serveCmd := &cobra.Command{
		Use:     "serve",
		Aliases: []string{"start"},
		Short:   "Start ollama",
Michael Yang's avatar
Michael Yang committed
1650
		Args:    cobra.ExactArgs(0),
Michael Yang's avatar
Michael Yang committed
1651
		RunE:    RunServer,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1652
1653
	}

1654
	pullCmd := &cobra.Command{
1655
1656
		Use:     "pull MODEL",
		Short:   "Pull a model from a registry",
Michael Yang's avatar
Michael Yang committed
1657
		Args:    cobra.ExactArgs(1),
1658
1659
		PreRunE: checkServerHeartbeat,
		RunE:    PullHandler,
1660
1661
	}

1662
1663
	pullCmd.Flags().Bool("insecure", false, "Use an insecure registry")

1664
	pushCmd := &cobra.Command{
1665
1666
		Use:     "push MODEL",
		Short:   "Push a model to a registry",
Michael Yang's avatar
Michael Yang committed
1667
		Args:    cobra.ExactArgs(1),
1668
1669
		PreRunE: checkServerHeartbeat,
		RunE:    PushHandler,
1670
1671
	}

1672
1673
	pushCmd.Flags().Bool("insecure", false, "Use an insecure registry")

1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
	signinCmd := &cobra.Command{
		Use:     "signin",
		Short:   "Sign in to ollama.com",
		Args:    cobra.ExactArgs(0),
		PreRunE: checkServerHeartbeat,
		RunE:    SigninHandler,
	}

	signoutCmd := &cobra.Command{
		Use:     "signout",
		Short:   "Sign out from ollama.com",
		Args:    cobra.ExactArgs(0),
		PreRunE: checkServerHeartbeat,
		RunE:    SignoutHandler,
	}

Patrick Devine's avatar
Patrick Devine committed
1690
	listCmd := &cobra.Command{
1691
		Use:     "list",
Patrick Devine's avatar
Patrick Devine committed
1692
		Aliases: []string{"ls"},
1693
		Short:   "List models",
1694
		PreRunE: checkServerHeartbeat,
1695
		RunE:    ListHandler,
1696
	}
1697
1698
1699
1700
1701
1702
1703

	psCmd := &cobra.Command{
		Use:     "ps",
		Short:   "List running models",
		PreRunE: checkServerHeartbeat,
		RunE:    ListRunningHandler,
	}
Patrick Devine's avatar
Patrick Devine committed
1704
	copyCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1705
		Use:     "cp SOURCE DESTINATION",
1706
		Short:   "Copy a model",
Michael Yang's avatar
Michael Yang committed
1707
		Args:    cobra.ExactArgs(2),
1708
1709
		PreRunE: checkServerHeartbeat,
		RunE:    CopyHandler,
Patrick Devine's avatar
Patrick Devine committed
1710
1711
	}

1712
	deleteCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1713
		Use:     "rm MODEL [MODEL...]",
1714
1715
1716
1717
		Short:   "Remove a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    DeleteHandler,
Patrick Devine's avatar
Patrick Devine committed
1718
1719
	}

1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
	runnerCmd := &cobra.Command{
		Use:    "runner",
		Hidden: true,
		RunE: func(cmd *cobra.Command, args []string) error {
			return runner.Execute(os.Args[1:])
		},
		FParseErrWhitelist: cobra.FParseErrWhitelist{UnknownFlags: true},
	}
	runnerCmd.SetHelpFunc(func(cmd *cobra.Command, args []string) {
		_ = runner.Execute(args[1:])
	})

1732
1733
1734
	envVars := envconfig.AsMap()

	envs := []envconfig.EnvVar{envVars["OLLAMA_HOST"]}
1735

1736
1737
1738
1739
	for _, cmd := range []*cobra.Command{
		createCmd,
		showCmd,
		runCmd,
Patrick Devine's avatar
Patrick Devine committed
1740
		stopCmd,
1741
1742
1743
		pullCmd,
		pushCmd,
		listCmd,
1744
		psCmd,
1745
1746
		copyCmd,
		deleteCmd,
1747
		serveCmd,
1748
	} {
1749
1750
		switch cmd {
		case runCmd:
1751
1752
1753
1754
1755
			appendEnvDocs(cmd, []envconfig.EnvVar{envVars["OLLAMA_HOST"], envVars["OLLAMA_NOHISTORY"]})
		case serveCmd:
			appendEnvDocs(cmd, []envconfig.EnvVar{
				envVars["OLLAMA_DEBUG"],
				envVars["OLLAMA_HOST"],
1756
				envVars["OLLAMA_CONTEXT_LENGTH"],
1757
1758
1759
1760
1761
1762
1763
				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"],
1764
				envVars["OLLAMA_SCHED_SPREAD"],
1765
				envVars["OLLAMA_FLASH_ATTENTION"],
1766
				envVars["OLLAMA_KV_CACHE_TYPE"],
1767
				envVars["OLLAMA_LLM_LIBRARY"],
1768
				envVars["OLLAMA_GPU_OVERHEAD"],
1769
				envVars["OLLAMA_LOAD_TIMEOUT"],
1770
			})
1771
1772
1773
		default:
			appendEnvDocs(cmd, envs)
		}
1774
1775
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1776
1777
	rootCmd.AddCommand(
		serveCmd,
1778
		createCmd,
Patrick Devine's avatar
Patrick Devine committed
1779
		showCmd,
1780
		runCmd,
Patrick Devine's avatar
Patrick Devine committed
1781
		stopCmd,
1782
1783
		pullCmd,
		pushCmd,
1784
1785
		signinCmd,
		signoutCmd,
Patrick Devine's avatar
Patrick Devine committed
1786
		listCmd,
1787
		psCmd,
Patrick Devine's avatar
Patrick Devine committed
1788
		copyCmd,
1789
		deleteCmd,
1790
		runnerCmd,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1791
1792
1793
1794
	)

	return rootCmd
}
1795
1796
1797
1798
1799
1800
1801
1802
1803

// If the user has explicitly set thinking options, either through the CLI or
// through the `/set think` or `set nothink` interactive options, then we
// respect them. Otherwise, we check model capabilities to see if the model
// supports thinking. If the model does support thinking, we enable it.
// Otherwise, we unset the thinking option (which is different than setting it
// to false).
//
// If capabilities are not provided, we fetch them from the server.
Michael Yang's avatar
Michael Yang committed
1804
func inferThinkingOption(caps *[]model.Capability, runOpts *runOptions, explicitlySetByUser bool) (*api.ThinkValue, error) {
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
	if explicitlySetByUser {
		return runOpts.Think, nil
	}

	if caps == nil {
		client, err := api.ClientFromEnvironment()
		if err != nil {
			return nil, err
		}
		ret, err := client.Show(context.Background(), &api.ShowRequest{
			Model: runOpts.Model,
		})
		if err != nil {
			return nil, err
		}
		caps = &ret.Capabilities
	}

	thinkingSupported := false
	for _, cap := range *caps {
		if cap == model.CapabilityThinking {
			thinkingSupported = true
		}
	}

	if thinkingSupported {
Michael Yang's avatar
Michael Yang committed
1831
		return &api.ThinkValue{Value: true}, nil
1832
1833
1834
1835
	}

	return nil, nil
}
Michael Yang's avatar
Michael Yang committed
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861

func renderToolCalls(toolCalls []api.ToolCall, plainText bool) string {
	out := ""
	formatExplanation := ""
	formatValues := ""
	if !plainText {
		formatExplanation = readline.ColorGrey + readline.ColorBold
		formatValues = readline.ColorDefault
		out += formatExplanation
	}
	for i, toolCall := range toolCalls {
		argsAsJSON, err := json.Marshal(toolCall.Function.Arguments)
		if err != nil {
			return ""
		}
		if i > 0 {
			out += "\n"
		}
		// all tool calls are unexpected since we don't currently support registering any in the CLI
		out += fmt.Sprintf("  Model called a non-existent function '%s()' with arguments: %s", formatValues+toolCall.Function.Name+formatExplanation, formatValues+string(argsAsJSON)+formatExplanation)
	}
	if !plainText {
		out += readline.ColorDefault
	}
	return out
}