cmd.go 38.7 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
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
// 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
	}
	for _, cap := range resp.Capabilities {
		if cap == model.CapabilityThinking {
			return
		}
	}
	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
302
	}

	return client.Generate(cmd.Context(), req, func(api.GenerateResponse) error { return nil })
}

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])
		}
303
		return err
Patrick Devine's avatar
Patrick Devine committed
304
305
306
307
	}
	return nil
}

308
func RunHandler(cmd *cobra.Command, args []string) error {
309
310
311
	interactive := true

	opts := runOptions{
312
313
		Model:    args[0],
		WordWrap: os.Getenv("TERM") == "xterm-256color",
314
		Options:  map[string]any{},
315
316
317
318
319
320
321
322
	}

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

323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
	thinkFlag := cmd.Flags().Lookup("think")
	if thinkFlag.Changed {
		think, err := cmd.Flags().GetBool("think")
		if err != nil {
			return err
		}
		opts.Think = &think
	} else {
		opts.Think = nil
	}
	hidethinking, err := cmd.Flags().GetBool("hidethinking")
	if err != nil {
		return err
	}
	opts.HideThinking = hidethinking

339
340
341
342
343
344
345
346
347
348
349
350
	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}
	}

351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
	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
	}
367
368
369
370
	// Be quiet if we're redirecting to a pipe or file
	if !term.IsTerminal(int(os.Stdout.Fd())) {
		interactive = false
	}
371
372
373
374
375
376
377

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

378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
	// 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
400
401
	}

402
403
404
405
406
	opts.Think, err = inferThinkingOption(&info.Capabilities, &opts, thinkFlag.Changed)
	if err != nil {
		return err
	}

407
408
409
410
411
	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
412
413
414
415
416
417
418
419
420
421
	if len(info.ProjectorInfo) != 0 {
		opts.MultiModal = true
	}
	for k := range info.ModelInfo {
		if strings.Contains(k, ".vision.") {
			opts.MultiModal = true
			break
		}
	}

422
423
424
	opts.ParentModel = info.Details.ParentModel

	if interactive {
Patrick Devine's avatar
Patrick Devine committed
425
		if err := loadOrUnloadModel(cmd, &opts); err != nil {
Michael Yang's avatar
Michael Yang committed
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
			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()
			}
		}

441
442
443
		return generateInteractive(cmd, opts)
	}
	return generate(cmd, opts)
Bruce MacDonald's avatar
Bruce MacDonald committed
444
445
}

446
func PushHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
447
	client, err := api.ClientFromEnvironment()
448
449
450
	if err != nil {
		return err
	}
451

452
453
454
455
456
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
457
458
459
460
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

	bars := make(map[string]*progress.Bar)
461
462
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
463

464
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
465
		if resp.Digest != "" {
466
467
468
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
469
470
471

			bar, ok := bars[resp.Digest]
			if !ok {
472
				bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
473
474
475
476
477
478
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
479
480
481
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
482
483
484
485
486
487

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

488
489
490
		return nil
	}

Michael Yang's avatar
Michael Yang committed
491
	request := api.PushRequest{Name: args[0], Insecure: insecure}
492
493

	n := model.ParseName(args[0])
Michael Yang's avatar
Michael Yang committed
494
	if err := client.Push(cmd.Context(), &request, fn); err != nil {
495
496
497
498
499
500
		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")
		}
Michael Yang's avatar
Michael Yang committed
501
502
503
		return err
	}

504
	p.Stop()
505
	spinner.Stop()
506
507
508
509
510
511
512
513

	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
514
	return nil
515
516
}

517
func ListHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
518
	client, err := api.ClientFromEnvironment()
519
520
521
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
522

Michael Yang's avatar
Michael Yang committed
523
	models, err := client.List(cmd.Context())
Patrick Devine's avatar
Patrick Devine committed
524
525
526
527
528
529
530
	if err != nil {
		return err
	}

	var data [][]string

	for _, m := range models.Models {
531
		if len(args) == 0 || strings.HasPrefix(strings.ToLower(m.Name), strings.ToLower(args[0])) {
532
			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
533
		}
Patrick Devine's avatar
Patrick Devine committed
534
535
536
	}

	table := tablewriter.NewWriter(os.Stdout)
Patrick Devine's avatar
Patrick Devine committed
537
	table.SetHeader([]string{"NAME", "ID", "SIZE", "MODIFIED"})
Patrick Devine's avatar
Patrick Devine committed
538
539
540
541
542
	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
543
	table.SetTablePadding("    ")
Patrick Devine's avatar
Patrick Devine committed
544
545
546
547
548
549
	table.AppendBulk(data)
	table.Render()

	return nil
}

550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
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
578
579
580
581
582
583
584
585

			var until string
			delta := time.Since(m.ExpiresAt)
			if delta > 0 {
				until = "Stopping..."
			} else {
				until = format.HumanTime(m.ExpiresAt, "Never")
			}
586
587
			ctxStr := strconv.Itoa(m.ContextLength)
			data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), procStr, ctxStr, until})
588
589
590
591
		}
	}

	table := tablewriter.NewWriter(os.Stdout)
592
	table.SetHeader([]string{"NAME", "ID", "SIZE", "PROCESSOR", "CONTEXT", "UNTIL"})
593
594
595
596
597
	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
598
	table.SetTablePadding("    ")
599
600
601
602
603
604
	table.AppendBulk(data)
	table.Render()

	return nil
}

605
func DeleteHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
606
	client, err := api.ClientFromEnvironment()
607
608
609
	if err != nil {
		return err
	}
610

611
612
613
614
615
616
617
618
619
620
621
	// 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 {
		if !strings.Contains(err.Error(), "not found") {
			return fmt.Errorf("unable to stop existing running model \"%s\": %s", args[0], err)
		}
	}

622
623
	for _, name := range args {
		req := api.DeleteRequest{Name: name}
Michael Yang's avatar
Michael Yang committed
624
		if err := client.Delete(cmd.Context(), &req); err != nil {
625
626
627
			return err
		}
		fmt.Printf("deleted '%s'\n", name)
628
629
630
631
	}
	return nil
}

Patrick Devine's avatar
Patrick Devine committed
632
func ShowHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
633
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
634
635
636
637
638
639
640
641
642
	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")
643
	verbose, errVerbose := cmd.Flags().GetBool("verbose")
Patrick Devine's avatar
Patrick Devine committed
644

645
	for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate, errVerbose} {
Patrick Devine's avatar
Patrick Devine committed
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
		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 {
680
		return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
681
682
	}

683
	req := api.ShowRequest{Name: args[0], Verbose: verbose}
684
685
686
687
	resp, err := client.Show(cmd.Context(), &req)
	if err != nil {
		return err
	}
688

689
	if flagsSet == 1 {
690
691
692
693
694
695
696
697
		switch showType {
		case "license":
			fmt.Println(resp.License)
		case "modelfile":
			fmt.Println(resp.Modelfile)
		case "parameters":
			fmt.Println(resp.Parameters)
		case "system":
698
			fmt.Print(resp.System)
699
		case "template":
700
			fmt.Print(resp.Template)
701
702
703
		}

		return nil
Patrick Devine's avatar
Patrick Devine committed
704
705
	}

706
	return showInfo(resp, verbose, os.Stdout)
707
708
}

709
func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
Michael Yang's avatar
Michael Yang committed
710
711
712
713
714
715
716
	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("    ")
717

Michael Yang's avatar
Michael Yang committed
718
719
720
		switch header {
		case "Template", "System", "License":
			table.SetColWidth(100)
721
722
		}

Michael Yang's avatar
Michael Yang committed
723
724
725
		table.AppendBulk(rows())
		table.Render()
		fmt.Fprintln(w)
Patrick Devine's avatar
Patrick Devine committed
726
727
	}

Michael Yang's avatar
Michael Yang committed
728
729
730
731
732
733
734
735
736
737
738
739
740
741
	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
	})
742

743
744
745
746
747
748
749
750
751
	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
752
753
754
755
756
757
758
759
760
	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
		})
761
762
	}

Michael Yang's avatar
Michael Yang committed
763
764
765
766
767
768
769
770
771
772
	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
		})
773
774
	}

775
776
777
778
779
780
781
782
783
784
785
	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) {
786
787
				case bool:
					v = fmt.Sprintf("%t", vData)
788
789
790
791
792
				case string:
					v = vData
				case float64:
					v = fmt.Sprintf("%g", vData)
				case []any:
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
					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)
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
					}
				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
843
844
	head := func(s string, n int) (rows [][]string) {
		scanner := bufio.NewScanner(strings.NewReader(s))
845
846
847
848
849
		count := 0
		for scanner.Scan() {
			text := strings.TrimSpace(scanner.Text())
			if text == "" {
				continue
850
			}
851
852
853
854
855
856
857
			count++
			if n < 0 || count <= n {
				rows = append(rows, []string{"", text})
			}
		}
		if n >= 0 && count > n {
			rows = append(rows, []string{"", "..."})
858
		}
Michael Yang's avatar
Michael Yang committed
859
		return
860
861
	}

Michael Yang's avatar
Michael Yang committed
862
863
864
865
866
	if resp.System != "" {
		tableRender("System", func() [][]string {
			return head(resp.System, 2)
		})
	}
867

Michael Yang's avatar
Michael Yang committed
868
869
870
871
	if resp.License != "" {
		tableRender("License", func() [][]string {
			return head(resp.License, 2)
		})
872
	}
Michael Yang's avatar
Michael Yang committed
873
874

	return nil
875
876
}

Patrick Devine's avatar
Patrick Devine committed
877
func CopyHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
878
	client, err := api.ClientFromEnvironment()
879
880
881
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
882
883

	req := api.CopyRequest{Source: args[0], Destination: args[1]}
Michael Yang's avatar
Michael Yang committed
884
	if err := client.Copy(cmd.Context(), &req); err != nil {
Patrick Devine's avatar
Patrick Devine committed
885
886
887
888
889
890
		return err
	}
	fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
	return nil
}

891
func PullHandler(cmd *cobra.Command, args []string) error {
892
893
894
895
896
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
897
	client, err := api.ClientFromEnvironment()
898
899
900
	if err != nil {
		return err
	}
901

Michael Yang's avatar
Michael Yang committed
902
903
904
905
906
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

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

907
908
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
909

910
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
911
		if resp.Digest != "" {
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
			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
			}

932
933
934
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
935
936
937

			bar, ok := bars[resp.Digest]
			if !ok {
938
939
940
941
942
943
				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
944
945
946
947
948
949
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
950
951
952
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
953
954
955
956
957
958

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

959
960
		return nil
	}
961

Michael Yang's avatar
Michael Yang committed
962
	request := api.PullRequest{Name: args[0], Insecure: insecure}
963
	return client.Pull(cmd.Context(), &request, fn)
Michael Yang's avatar
Michael Yang committed
964
965
}

966
967
type generateContextKey string

968
type runOptions struct {
969
970
971
972
973
974
975
976
977
978
979
980
981
	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
	Think        *bool
	HideThinking bool
982
983
}

984
985
986
987
988
989
990
991
992
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
993
994
			if state.lineLength+1 > termWidth-5 {
				if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
995
996
997
998
999
1000
1001
					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
1002
1003
				a := runewidth.StringWidth(state.wordBuffer)
				if a > 0 {
1004
					fmt.Printf("\x1b[%dD", a)
1005
1006
				}
				fmt.Printf("\x1b[K\n")
1007
				fmt.Printf("%s%c", state.wordBuffer, ch)
1008
1009
1010
				chWidth := runewidth.RuneWidth(ch)

				state.lineLength = runewidth.StringWidth(state.wordBuffer) + chWidth
1011
1012
			} else {
				fmt.Print(string(ch))
1013
1014
1015
1016
				state.lineLength += runewidth.RuneWidth(ch)
				if runewidth.RuneWidth(ch) >= 2 {
					state.wordBuffer = ""
					continue
Josh Yan's avatar
Josh Yan committed
1017
				}
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036

				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 = ""
		}
	}
}

1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
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
}

1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
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
1084
1085
	var thinkTagOpened bool = false
	var thinkTagClosed bool = false
1086
1087

	fn := func(response api.ChatResponse) error {
1088
1089
1090
		if response.Message.Content != "" || !opts.HideThinking {
			p.StopAndClear()
		}
1091
1092
1093
1094

		latest = response

		role = response.Message.Role
1095
1096
1097
1098
1099
1100
1101
1102
		if response.Message.Thinking != "" && !opts.HideThinking {
			if !thinkTagOpened {
				fmt.Print(thinkingOutputOpeningText(false))
				thinkTagOpened = true
			}
			displayResponse(response.Message.Thinking, opts.WordWrap, state)
		}

1103
		content := response.Message.Content
1104
1105
1106
1107
1108
1109
1110
1111
		if thinkTagOpened && !thinkTagClosed && content != "" {
			fmt.Print(thinkingOutputClosingText(false))
			thinkTagClosed = true
		}
		// 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)
1112
1113
1114
1115
1116
1117
1118
		fullResponse.WriteString(content)

		displayResponse(content, opts.WordWrap, state)

		return nil
	}

1119
1120
1121
1122
	if opts.Format == "json" {
		opts.Format = `"` + opts.Format + `"`
	}

1123
1124
1125
	req := &api.ChatRequest{
		Model:    opts.Model,
		Messages: opts.Messages,
1126
		Format:   json.RawMessage(opts.Format),
1127
		Options:  opts.Options,
1128
		Think:    opts.Think,
1129
1130
	}

1131
1132
1133
1134
	if opts.KeepAlive != nil {
		req.KeepAlive = opts.KeepAlive
	}

1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
	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
1160
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
1161
	if err != nil {
1162
		return err
Patrick Devine's avatar
Patrick Devine committed
1163
	}
Michael Yang's avatar
Michael Yang committed
1164

Michael Yang's avatar
Michael Yang committed
1165
	p := progress.NewProgress(os.Stderr)
1166
	defer p.StopAndClear()
1167

Michael Yang's avatar
Michael Yang committed
1168
1169
1170
	spinner := progress.NewSpinner("")
	p.Add("", spinner)

1171
1172
1173
1174
1175
1176
1177
	var latest api.GenerateResponse

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

Michael Yang's avatar
Michael Yang committed
1178
	ctx, cancel := context.WithCancel(cmd.Context())
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
	defer cancel()

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

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

1189
	var state *displayResponseState = &displayResponseState{}
1190
1191
	var thinkTagOpened bool = false
	var thinkTagClosed bool = false
1192

1193
	plainText := !term.IsTerminal(int(os.Stdout.Fd()))
1194

1195
	fn := func(response api.GenerateResponse) error {
Patrick Devine's avatar
Patrick Devine committed
1196
		latest = response
1197
		content := response.Response
1198

1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
		if response.Response != "" || !opts.HideThinking {
			p.StopAndClear()
		}

		if response.Thinking != "" && !opts.HideThinking {
			if !thinkTagOpened {
				fmt.Print(thinkingOutputOpeningText(plainText))
				thinkTagOpened = true
			}
			displayResponse(response.Thinking, opts.WordWrap, state)
		}

		if thinkTagOpened && !thinkTagClosed && content != "" {
			fmt.Print(thinkingOutputClosingText(plainText))
			thinkTagClosed = true
		}

1216
		displayResponse(content, opts.WordWrap, state)
1217

Patrick Devine's avatar
Patrick Devine committed
1218
1219
		return nil
	}
1220

1221
1222
1223
1224
1225
1226
1227
	if opts.MultiModal {
		opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
		if err != nil {
			return err
		}
	}

1228
1229
1230
1231
	if opts.Format == "json" {
		opts.Format = `"` + opts.Format + `"`
	}

Michael Yang's avatar
Michael Yang committed
1232
	request := api.GenerateRequest{
1233
1234
1235
1236
		Model:     opts.Model,
		Prompt:    opts.Prompt,
		Context:   generateContext,
		Images:    opts.Images,
1237
		Format:    json.RawMessage(opts.Format),
1238
1239
1240
		System:    opts.System,
		Options:   opts.Options,
		KeepAlive: opts.KeepAlive,
1241
		Think:     opts.Think,
Michael Yang's avatar
Michael Yang committed
1242
1243
1244
	}

	if err := client.Generate(ctx, &request, fn); err != nil {
1245
		if errors.Is(err, context.Canceled) {
1246
			return nil
1247
		}
1248
		return err
Patrick Devine's avatar
Patrick Devine committed
1249
	}
1250

1251
	if opts.Prompt != "" {
Michael Yang's avatar
Michael Yang committed
1252
1253
		fmt.Println()
		fmt.Println()
Patrick Devine's avatar
Patrick Devine committed
1254
	}
1255

1256
1257
1258
1259
	if !latest.Done {
		return nil
	}

Patrick Devine's avatar
Patrick Devine committed
1260
1261
	verbose, err := cmd.Flags().GetBool("verbose")
	if err != nil {
1262
		return err
Patrick Devine's avatar
Patrick Devine committed
1263
	}
Michael Yang's avatar
Michael Yang committed
1264

Patrick Devine's avatar
Patrick Devine committed
1265
1266
	if verbose {
		latest.Summary()
Michael Yang's avatar
Michael Yang committed
1267
	}
Michael Yang's avatar
Michael Yang committed
1268

Patrick Devine's avatar
Patrick Devine committed
1269
1270
1271
	ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
	cmd.SetContext(ctx)

1272
	return nil
Michael Yang's avatar
Michael Yang committed
1273
1274
}

1275
func RunServer(_ *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1276
	if err := initializeKeypair(); err != nil {
1277
1278
1279
		return err
	}

Michael Yang's avatar
host  
Michael Yang committed
1280
	ln, err := net.Listen("tcp", envconfig.Host().Host)
1281
1282
1283
	if err != nil {
		return err
	}
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1284

1285
1286
1287
1288
1289
1290
	err = server.Serve(ln)
	if errors.Is(err, http.ErrServerClosed) {
		return nil
	}

	return err
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1291
1292
}

1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
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
1305
		cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
1306
1307
1308
1309
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1310
		privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
1311
1312
1313
1314
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1315
		if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
1316
1317
1318
			return fmt.Errorf("could not create directory %w", err)
		}

Michael Yang's avatar
Michael Yang committed
1319
		if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
1320
1321
1322
			return err
		}

Michael Yang's avatar
Michael Yang committed
1323
		sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
1324
1325
1326
1327
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1328
		publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
1329

Michael Yang's avatar
Michael Yang committed
1330
		if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
1331
1332
1333
			return err
		}

Michael Yang's avatar
Michael Yang committed
1334
		fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
1335
1336
1337
1338
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1339
func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1340
	client, err := api.ClientFromEnvironment()
1341
1342
1343
	if err != nil {
		return err
	}
Michael Yang's avatar
Michael Yang committed
1344
	if err := client.Heartbeat(cmd.Context()); err != nil {
1345
		if !(strings.Contains(err.Error(), " refused") || strings.Contains(err.Error(), "could not connect")) {
Bruce MacDonald's avatar
Bruce MacDonald committed
1346
1347
			return err
		}
1348
		if err := startApp(cmd.Context(), client); err != nil {
1349
			return fmt.Errorf("ollama server not responding - %w", err)
1350
1351
1352
1353
1354
		}
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1355
1356
1357
1358
1359
1360
1361
1362
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
1363
1364
1365
1366
1367
		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
1368
1369
	}

1370
	if serverVersion != version.Version {
Michael Yang's avatar
Michael Yang committed
1371
		fmt.Printf("Warning: client version is %s\n", version.Version)
1372
	}
Michael Yang's avatar
Michael Yang committed
1373
1374
}

1375
func appendEnvDocs(cmd *cobra.Command, envs []envconfig.EnvVar) {
1376
1377
1378
1379
1380
	if len(envs) == 0 {
		return
	}

	envUsage := `
1381
1382
Environment Variables:
`
1383
	for _, e := range envs {
1384
		envUsage += fmt.Sprintf("      %-24s   %s\n", e.Name, e.Description)
1385
1386
1387
	}

	cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
1388
1389
}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1390
1391
func NewCLI() *cobra.Command {
	log.SetFlags(log.LstdFlags | log.Lshortfile)
Michael Yang's avatar
Michael Yang committed
1392
	cobra.EnableCommandSorting = false
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1393

1394
	if runtime.GOOS == "windows" && term.IsTerminal(int(os.Stdout.Fd())) {
1395
		console.ConsoleFromFile(os.Stdin) //nolint:errcheck
1396
1397
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1398
	rootCmd := &cobra.Command{
1399
1400
1401
1402
		Use:           "ollama",
		Short:         "Large language model runner",
		SilenceUsage:  true,
		SilenceErrors: true,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1403
1404
1405
		CompletionOptions: cobra.CompletionOptions{
			DisableDefaultCmd: true,
		},
Michael Yang's avatar
Michael Yang committed
1406
1407
1408
1409
1410
1411
1412
1413
		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
1414
1415
	}

Michael Yang's avatar
Michael Yang committed
1416
	rootCmd.Flags().BoolP("version", "v", false, "Show version information")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1417

1418
	createCmd := &cobra.Command{
1419
1420
		Use:     "create MODEL",
		Short:   "Create a model from a Modelfile",
Michael Yang's avatar
Michael Yang committed
1421
		Args:    cobra.ExactArgs(1),
1422
1423
		PreRunE: checkServerHeartbeat,
		RunE:    CreateHandler,
1424
1425
	}

1426
	createCmd.Flags().StringP("file", "f", "", "Name of the Modelfile (default \"Modelfile\"")
1427
	createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_K_M)")
1428

Patrick Devine's avatar
Patrick Devine committed
1429
1430
1431
	showCmd := &cobra.Command{
		Use:     "show MODEL",
		Short:   "Show information for a model",
Michael Yang's avatar
Michael Yang committed
1432
		Args:    cobra.ExactArgs(1),
Patrick Devine's avatar
Patrick Devine committed
1433
1434
1435
1436
1437
1438
1439
1440
		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")
1441
	showCmd.Flags().Bool("system", false, "Show system message of a model")
1442
	showCmd.Flags().BoolP("verbose", "v", false, "Show detailed model information")
Patrick Devine's avatar
Patrick Devine committed
1443

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1444
	runCmd := &cobra.Command{
1445
1446
1447
1448
1449
		Use:     "run MODEL [PROMPT]",
		Short:   "Run a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    RunHandler,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1450
1451
	}

1452
	runCmd.Flags().String("keepalive", "", "Duration to keep a model loaded (e.g. 5m)")
1453
	runCmd.Flags().Bool("verbose", false, "Show timings for response")
1454
	runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
1455
	runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1456
	runCmd.Flags().String("format", "", "Response format (e.g. json)")
1457
1458
	runCmd.Flags().Bool("think", false, "Whether to use thinking mode for supported models")
	runCmd.Flags().Bool("hidethinking", false, "Hide thinking output (if provided)")
Patrick Devine's avatar
Patrick Devine committed
1459
1460
1461
1462
1463
1464
1465
1466
1467

	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
1468
1469
1470
1471
	serveCmd := &cobra.Command{
		Use:     "serve",
		Aliases: []string{"start"},
		Short:   "Start ollama",
Michael Yang's avatar
Michael Yang committed
1472
		Args:    cobra.ExactArgs(0),
Michael Yang's avatar
Michael Yang committed
1473
		RunE:    RunServer,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1474
1475
	}

1476
	pullCmd := &cobra.Command{
1477
1478
		Use:     "pull MODEL",
		Short:   "Pull a model from a registry",
Michael Yang's avatar
Michael Yang committed
1479
		Args:    cobra.ExactArgs(1),
1480
1481
		PreRunE: checkServerHeartbeat,
		RunE:    PullHandler,
1482
1483
	}

1484
1485
	pullCmd.Flags().Bool("insecure", false, "Use an insecure registry")

1486
	pushCmd := &cobra.Command{
1487
1488
		Use:     "push MODEL",
		Short:   "Push a model to a registry",
Michael Yang's avatar
Michael Yang committed
1489
		Args:    cobra.ExactArgs(1),
1490
1491
		PreRunE: checkServerHeartbeat,
		RunE:    PushHandler,
1492
1493
	}

1494
1495
	pushCmd.Flags().Bool("insecure", false, "Use an insecure registry")

Patrick Devine's avatar
Patrick Devine committed
1496
	listCmd := &cobra.Command{
1497
		Use:     "list",
Patrick Devine's avatar
Patrick Devine committed
1498
		Aliases: []string{"ls"},
1499
		Short:   "List models",
1500
		PreRunE: checkServerHeartbeat,
1501
		RunE:    ListHandler,
1502
	}
1503
1504
1505
1506
1507
1508
1509

	psCmd := &cobra.Command{
		Use:     "ps",
		Short:   "List running models",
		PreRunE: checkServerHeartbeat,
		RunE:    ListRunningHandler,
	}
Patrick Devine's avatar
Patrick Devine committed
1510
	copyCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1511
		Use:     "cp SOURCE DESTINATION",
1512
		Short:   "Copy a model",
Michael Yang's avatar
Michael Yang committed
1513
		Args:    cobra.ExactArgs(2),
1514
1515
		PreRunE: checkServerHeartbeat,
		RunE:    CopyHandler,
Patrick Devine's avatar
Patrick Devine committed
1516
1517
	}

1518
	deleteCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1519
		Use:     "rm MODEL [MODEL...]",
1520
1521
1522
1523
		Short:   "Remove a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    DeleteHandler,
Patrick Devine's avatar
Patrick Devine committed
1524
1525
	}

1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
	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:])
	})

1538
1539
1540
	envVars := envconfig.AsMap()

	envs := []envconfig.EnvVar{envVars["OLLAMA_HOST"]}
1541

1542
1543
1544
1545
	for _, cmd := range []*cobra.Command{
		createCmd,
		showCmd,
		runCmd,
Patrick Devine's avatar
Patrick Devine committed
1546
		stopCmd,
1547
1548
1549
		pullCmd,
		pushCmd,
		listCmd,
1550
		psCmd,
1551
1552
		copyCmd,
		deleteCmd,
1553
		serveCmd,
1554
	} {
1555
1556
		switch cmd {
		case runCmd:
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
			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"],
1569
				envVars["OLLAMA_SCHED_SPREAD"],
1570
				envVars["OLLAMA_FLASH_ATTENTION"],
1571
				envVars["OLLAMA_KV_CACHE_TYPE"],
1572
				envVars["OLLAMA_LLM_LIBRARY"],
1573
				envVars["OLLAMA_GPU_OVERHEAD"],
1574
				envVars["OLLAMA_LOAD_TIMEOUT"],
1575
			})
1576
1577
1578
		default:
			appendEnvDocs(cmd, envs)
		}
1579
1580
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1581
1582
	rootCmd.AddCommand(
		serveCmd,
1583
		createCmd,
Patrick Devine's avatar
Patrick Devine committed
1584
		showCmd,
1585
		runCmd,
Patrick Devine's avatar
Patrick Devine committed
1586
		stopCmd,
1587
1588
		pullCmd,
		pushCmd,
Patrick Devine's avatar
Patrick Devine committed
1589
		listCmd,
1590
		psCmd,
Patrick Devine's avatar
Patrick Devine committed
1591
		copyCmd,
1592
		deleteCmd,
1593
		runnerCmd,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1594
1595
1596
1597
	)

	return rootCmd
}
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639

// 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.
func inferThinkingOption(caps *[]model.Capability, runOpts *runOptions, explicitlySetByUser bool) (*bool, error) {
	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 {
		thinking := true
		return &thinking, nil
	}

	return nil, nil
}