cmd.go 33 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/term"
Michael Yang's avatar
Michael Yang committed
35

36
	"github.com/ollama/ollama/api"
37
	"github.com/ollama/ollama/envconfig"
38
	"github.com/ollama/ollama/format"
39
	"github.com/ollama/ollama/parser"
40
	"github.com/ollama/ollama/progress"
Jesse Gross's avatar
Jesse Gross committed
41
	"github.com/ollama/ollama/runner"
42
	"github.com/ollama/ollama/server"
43
	"github.com/ollama/ollama/types/model"
44
	"github.com/ollama/ollama/version"
Jeffrey Morgan's avatar
Jeffrey Morgan committed
45
46
)

47
var errModelfileNotFound = errors.New("specified Modelfile wasn't found")
48
49

func getModelfileName(cmd *cobra.Command) (string, error) {
50
	filename, _ := cmd.Flags().GetString("file")
51
52
53
54
55
56

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

	absName, err := filepath.Abs(filename)
57
	if err != nil {
58
		return "", err
59
60
	}

61
	_, err = os.Stat(absName)
62
	if err != nil {
63
		return "", err
64
	}
65

66
67
68
69
	return absName, nil
}

func CreateHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
70
71
72
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

73
74
75
76
77
78
79
80
81
82
	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
83
		return err
84
85
86
87
88
89
90
91
	} else {
		f, err := os.Open(filename)
		if err != nil {
			return err
		}

		reader = f
		defer f.Close()
Michael Yang's avatar
Michael Yang committed
92
93
	}

94
	modelfile, err := parser.ParseFile(reader)
Michael Yang's avatar
Michael Yang committed
95
96
97
98
	if err != nil {
		return err
	}

99
100
101
102
	status := "gathering model components"
	spinner := progress.NewSpinner(status)
	p.Add(status, spinner)

103
	req, err := modelfile.CreateRequest(filepath.Dir(filename))
Michael Yang's avatar
Michael Yang committed
104
105
106
	if err != nil {
		return err
	}
107
	spinner.Stop()
Michael Yang's avatar
Michael Yang committed
108

109
110
111
112
113
	req.Name = args[0]
	quantize, _ := cmd.Flags().GetString("quantize")
	if quantize != "" {
		req.Quantize = quantize
	}
114

115
116
117
118
119
	client, err := api.ClientFromEnvironment()
	if err != nil {
		return err
	}

120
121
122
123
	if len(req.Files) > 0 {
		fileMap := map[string]string{}
		for f, digest := range req.Files {
			if _, err := createBlob(cmd, client, f, digest, p); err != nil {
Michael Yang's avatar
Michael Yang committed
124
125
				return err
			}
126
127
128
129
			fileMap[filepath.Base(f)] = digest
		}
		req.Files = fileMap
	}
Michael Yang's avatar
Michael Yang committed
130

131
132
133
134
	if len(req.Adapters) > 0 {
		fileMap := map[string]string{}
		for f, digest := range req.Adapters {
			if _, err := createBlob(cmd, client, f, digest, p); err != nil {
Michael Yang's avatar
Michael Yang committed
135
136
				return err
			}
137
			fileMap[filepath.Base(f)] = digest
Michael Yang's avatar
Michael Yang committed
138
		}
139
		req.Adapters = fileMap
Michael Yang's avatar
Michael Yang committed
140
	}
Michael Yang's avatar
Michael Yang committed
141

Michael Yang's avatar
Michael Yang committed
142
	bars := make(map[string]*progress.Bar)
143
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
144
145
146
		if resp.Digest != "" {
			bar, ok := bars[resp.Digest]
			if !ok {
147
				bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
148
149
150
151
152
153
154
155
156
157
158
159
160
				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)
		}

161
162
163
		return nil
	}

164
	if err := client.Create(cmd.Context(), req, fn); err != nil {
165
166
167
		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")
		}
168
169
170
171
172
173
		return err
	}

	return nil
}

174
175
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
176
177
178
179
	if err != nil {
		return "", err
	}

180
	bin, err := os.Open(realPath)
181
182
183
184
185
	if err != nil {
		return "", err
	}
	defer bin.Close()

186
187
188
189
190
191
192
193
	// Get file info to retrieve the size
	fileInfo, err := bin.Stat()
	if err != nil {
		return "", err
	}
	fileSize := fileInfo.Size()

	var pw progressWriter
194
195
196
197
	status := fmt.Sprintf("copying file %s 0%%", digest)
	spinner := progress.NewSpinner(status)
	p.Add(status, spinner)
	defer spinner.Stop()
198
199
200
201
202
203
204
205
206
207

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

	go func() {
		ticker := time.NewTicker(60 * time.Millisecond)
		defer ticker.Stop()
		for {
			select {
			case <-ticker.C:
208
				spinner.SetMessage(fmt.Sprintf("copying file %s %d%%", digest, int(100*pw.n.Load()/fileSize)))
209
			case <-done:
210
				spinner.SetMessage(fmt.Sprintf("copying file %s 100%%", digest))
211
212
213
214
215
216
				return
			}
		}
	}()

	if err = client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
217
218
219
220
221
		return "", err
	}
	return digest, nil
}

222
223
224
225
226
227
228
229
230
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
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
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,
	}

	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])
		}
260
		return err
Patrick Devine's avatar
Patrick Devine committed
261
262
263
264
	}
	return nil
}

265
func RunHandler(cmd *cobra.Command, args []string) error {
266
267
268
	interactive := true

	opts := runOptions{
269
270
		Model:    args[0],
		WordWrap: os.Getenv("TERM") == "xterm-256color",
271
		Options:  map[string]any{},
272
273
274
275
276
277
278
279
	}

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

280
281
282
283
284
285
286
287
288
289
290
291
	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}
	}

292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
	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
	}
308
309
310
311
	// Be quiet if we're redirecting to a pipe or file
	if !term.IsTerminal(int(os.Stdout.Fd())) {
		interactive = false
	}
312
313
314
315
316
317
318

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

319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
	// 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
341
342
	}

343
344
345
346
347
	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
348
349
350
351
352
353
354
355
356
357
	if len(info.ProjectorInfo) != 0 {
		opts.MultiModal = true
	}
	for k := range info.ModelInfo {
		if strings.Contains(k, ".vision.") {
			opts.MultiModal = true
			break
		}
	}

358
359
360
	opts.ParentModel = info.Details.ParentModel

	if interactive {
Patrick Devine's avatar
Patrick Devine committed
361
		if err := loadOrUnloadModel(cmd, &opts); err != nil {
Michael Yang's avatar
Michael Yang committed
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
			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()
			}
		}

377
378
379
		return generateInteractive(cmd, opts)
	}
	return generate(cmd, opts)
Bruce MacDonald's avatar
Bruce MacDonald committed
380
381
}

382
func PushHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
383
	client, err := api.ClientFromEnvironment()
384
385
386
	if err != nil {
		return err
	}
387

388
389
390
391
392
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
393
394
395
396
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

	bars := make(map[string]*progress.Bar)
397
398
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
399

400
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
401
		if resp.Digest != "" {
402
403
404
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
405
406
407

			bar, ok := bars[resp.Digest]
			if !ok {
408
				bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
Michael Yang's avatar
Michael Yang committed
409
410
411
412
413
414
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
415
416
417
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
418
419
420
421
422
423

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

424
425
426
		return nil
	}

Michael Yang's avatar
Michael Yang committed
427
	request := api.PushRequest{Name: args[0], Insecure: insecure}
428
429

	n := model.ParseName(args[0])
Michael Yang's avatar
Michael Yang committed
430
	if err := client.Push(cmd.Context(), &request, fn); err != nil {
431
432
433
434
435
436
		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
437
438
439
		return err
	}

440
	p.Stop()
441
	spinner.Stop()
442
443
444
445
446
447
448
449

	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
450
	return nil
451
452
}

453
func ListHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
454
	client, err := api.ClientFromEnvironment()
455
456
457
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
458

Michael Yang's avatar
Michael Yang committed
459
	models, err := client.List(cmd.Context())
Patrick Devine's avatar
Patrick Devine committed
460
461
462
463
464
465
466
	if err != nil {
		return err
	}

	var data [][]string

	for _, m := range models.Models {
467
		if len(args) == 0 || strings.HasPrefix(strings.ToLower(m.Name), strings.ToLower(args[0])) {
468
			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
469
		}
Patrick Devine's avatar
Patrick Devine committed
470
471
472
	}

	table := tablewriter.NewWriter(os.Stdout)
Patrick Devine's avatar
Patrick Devine committed
473
	table.SetHeader([]string{"NAME", "ID", "SIZE", "MODIFIED"})
Patrick Devine's avatar
Patrick Devine committed
474
475
476
477
478
	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
479
	table.SetTablePadding("    ")
Patrick Devine's avatar
Patrick Devine committed
480
481
482
483
484
485
	table.AppendBulk(data)
	table.Render()

	return nil
}

486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
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
514
515
516
517
518
519
520
521
522

			var until string
			delta := time.Since(m.ExpiresAt)
			if delta > 0 {
				until = "Stopping..."
			} else {
				until = format.HumanTime(m.ExpiresAt, "Never")
			}
			data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), procStr, until})
523
524
525
526
527
528
529
530
531
532
		}
	}

	table := tablewriter.NewWriter(os.Stdout)
	table.SetHeader([]string{"NAME", "ID", "SIZE", "PROCESSOR", "UNTIL"})
	table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
	table.SetAlignment(tablewriter.ALIGN_LEFT)
	table.SetHeaderLine(false)
	table.SetBorder(false)
	table.SetNoWhiteSpace(true)
Michael Yang's avatar
Michael Yang committed
533
	table.SetTablePadding("    ")
534
535
536
537
538
539
	table.AppendBulk(data)
	table.Render()

	return nil
}

540
func DeleteHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
541
	client, err := api.ClientFromEnvironment()
542
543
544
	if err != nil {
		return err
	}
545

546
547
548
549
550
551
552
553
554
555
556
	// 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)
		}
	}

557
558
	for _, name := range args {
		req := api.DeleteRequest{Name: name}
Michael Yang's avatar
Michael Yang committed
559
		if err := client.Delete(cmd.Context(), &req); err != nil {
560
561
562
			return err
		}
		fmt.Printf("deleted '%s'\n", name)
563
564
565
566
	}
	return nil
}

Patrick Devine's avatar
Patrick Devine committed
567
func ShowHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
568
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
569
570
571
572
573
574
575
576
577
	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")
578
	verbose, errVerbose := cmd.Flags().GetBool("verbose")
Patrick Devine's avatar
Patrick Devine committed
579

580
	for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate, errVerbose} {
Patrick Devine's avatar
Patrick Devine committed
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
		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 {
615
		return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
616
617
	}

618
	req := api.ShowRequest{Name: args[0], Verbose: verbose}
619
620
621
622
	resp, err := client.Show(cmd.Context(), &req)
	if err != nil {
		return err
	}
623

624
	if flagsSet == 1 {
625
626
627
628
629
630
631
632
		switch showType {
		case "license":
			fmt.Println(resp.License)
		case "modelfile":
			fmt.Println(resp.Modelfile)
		case "parameters":
			fmt.Println(resp.Parameters)
		case "system":
633
			fmt.Print(resp.System)
634
		case "template":
635
			fmt.Print(resp.Template)
636
637
638
		}

		return nil
Patrick Devine's avatar
Patrick Devine committed
639
640
	}

641
	return showInfo(resp, verbose, os.Stdout)
642
643
}

644
func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
Michael Yang's avatar
Michael Yang committed
645
646
647
648
649
650
651
	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("    ")
652

Michael Yang's avatar
Michael Yang committed
653
654
655
		switch header {
		case "Template", "System", "License":
			table.SetColWidth(100)
656
657
		}

Michael Yang's avatar
Michael Yang committed
658
659
660
		table.AppendBulk(rows())
		table.Render()
		fmt.Fprintln(w)
Patrick Devine's avatar
Patrick Devine committed
661
662
	}

Michael Yang's avatar
Michael Yang committed
663
664
665
666
667
668
669
670
671
672
673
674
675
676
	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
	})
677

678
679
680
681
682
683
684
685
686
	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
687
688
689
690
691
692
693
694
695
	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
		})
696
697
	}

Michael Yang's avatar
Michael Yang committed
698
699
700
701
702
703
704
705
706
707
	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
		})
708
709
	}

710
711
712
713
714
715
716
717
718
719
720
	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) {
721
722
				case bool:
					v = fmt.Sprintf("%t", vData)
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
				case string:
					v = vData
				case float64:
					v = fmt.Sprintf("%g", vData)
				case []any:
					n := 3
					if len(vData) < n {
						n = len(vData)
					}
					v = fmt.Sprintf("%v", vData[:n])
				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
751
752
753
754
755
	head := func(s string, n int) (rows [][]string) {
		scanner := bufio.NewScanner(strings.NewReader(s))
		for scanner.Scan() && (len(rows) < n || n < 0) {
			if text := scanner.Text(); text != "" {
				rows = append(rows, []string{"", strings.TrimSpace(text)})
756
757
			}
		}
Michael Yang's avatar
Michael Yang committed
758
		return
759
760
	}

Michael Yang's avatar
Michael Yang committed
761
762
763
764
765
	if resp.System != "" {
		tableRender("System", func() [][]string {
			return head(resp.System, 2)
		})
	}
766

Michael Yang's avatar
Michael Yang committed
767
768
769
770
	if resp.License != "" {
		tableRender("License", func() [][]string {
			return head(resp.License, 2)
		})
771
	}
Michael Yang's avatar
Michael Yang committed
772
773

	return nil
774
775
}

Patrick Devine's avatar
Patrick Devine committed
776
func CopyHandler(cmd *cobra.Command, args []string) error {
Michael Yang's avatar
Michael Yang committed
777
	client, err := api.ClientFromEnvironment()
778
779
780
	if err != nil {
		return err
	}
Patrick Devine's avatar
Patrick Devine committed
781
782

	req := api.CopyRequest{Source: args[0], Destination: args[1]}
Michael Yang's avatar
Michael Yang committed
783
	if err := client.Copy(cmd.Context(), &req); err != nil {
Patrick Devine's avatar
Patrick Devine committed
784
785
786
787
788
789
		return err
	}
	fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
	return nil
}

790
func PullHandler(cmd *cobra.Command, args []string) error {
791
792
793
794
795
	insecure, err := cmd.Flags().GetBool("insecure")
	if err != nil {
		return err
	}

Michael Yang's avatar
Michael Yang committed
796
	client, err := api.ClientFromEnvironment()
797
798
799
	if err != nil {
		return err
	}
800

Michael Yang's avatar
Michael Yang committed
801
802
803
804
805
	p := progress.NewProgress(os.Stderr)
	defer p.Stop()

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

806
807
	var status string
	var spinner *progress.Spinner
Michael Yang's avatar
Michael Yang committed
808

809
	fn := func(resp api.ProgressResponse) error {
Michael Yang's avatar
Michael Yang committed
810
		if resp.Digest != "" {
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
			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
			}

831
832
833
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
834
835
836

			bar, ok := bars[resp.Digest]
			if !ok {
837
838
839
840
841
842
				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
843
844
845
846
847
848
				bars[resp.Digest] = bar
				p.Add(resp.Digest, bar)
			}

			bar.Set(resp.Completed)
		} else if status != resp.Status {
849
850
851
			if spinner != nil {
				spinner.Stop()
			}
Michael Yang's avatar
Michael Yang committed
852
853
854
855
856
857

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

858
859
		return nil
	}
860

Michael Yang's avatar
Michael Yang committed
861
	request := api.PullRequest{Name: args[0], Insecure: insecure}
862
	return client.Pull(cmd.Context(), &request, fn)
Michael Yang's avatar
Michael Yang committed
863
864
}

865
866
type generateContextKey string

867
type runOptions struct {
868
869
870
871
872
873
874
875
	Model       string
	ParentModel string
	Prompt      string
	Messages    []api.Message
	WordWrap    bool
	Format      string
	System      string
	Images      []api.ImageData
876
	Options     map[string]any
877
	MultiModal  bool
878
	KeepAlive   *api.Duration
879
880
}

881
882
883
884
885
886
887
888
889
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
890
891
			if state.lineLength+1 > termWidth-5 {
				if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
892
893
894
895
896
897
898
					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
899
900
				a := runewidth.StringWidth(state.wordBuffer)
				if a > 0 {
901
					fmt.Printf("\x1b[%dD", a)
902
903
				}
				fmt.Printf("\x1b[K\n")
904
				fmt.Printf("%s%c", state.wordBuffer, ch)
905
906
907
				chWidth := runewidth.RuneWidth(ch)

				state.lineLength = runewidth.StringWidth(state.wordBuffer) + chWidth
908
909
			} else {
				fmt.Print(string(ch))
910
911
912
913
				state.lineLength += runewidth.RuneWidth(ch)
				if runewidth.RuneWidth(ch) >= 2 {
					state.wordBuffer = ""
					continue
Josh Yan's avatar
Josh Yan committed
914
				}
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975

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

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

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

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

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

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

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

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

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

		latest = response

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

		displayResponse(content, opts.WordWrap, state)

		return nil
	}

976
977
978
979
	if opts.Format == "json" {
		opts.Format = `"` + opts.Format + `"`
	}

980
981
982
	req := &api.ChatRequest{
		Model:    opts.Model,
		Messages: opts.Messages,
983
		Format:   json.RawMessage(opts.Format),
984
985
986
		Options:  opts.Options,
	}

987
988
989
990
	if opts.KeepAlive != nil {
		req.KeepAlive = opts.KeepAlive
	}

991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
	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
1016
	client, err := api.ClientFromEnvironment()
Patrick Devine's avatar
Patrick Devine committed
1017
	if err != nil {
1018
		return err
Patrick Devine's avatar
Patrick Devine committed
1019
	}
Michael Yang's avatar
Michael Yang committed
1020

Michael Yang's avatar
Michael Yang committed
1021
	p := progress.NewProgress(os.Stderr)
1022
	defer p.StopAndClear()
1023

Michael Yang's avatar
Michael Yang committed
1024
1025
1026
	spinner := progress.NewSpinner("")
	p.Add("", spinner)

1027
1028
1029
1030
1031
1032
1033
	var latest api.GenerateResponse

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

Michael Yang's avatar
Michael Yang committed
1034
	ctx, cancel := context.WithCancel(cmd.Context())
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
	defer cancel()

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

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

1045
	var state *displayResponseState = &displayResponseState{}
1046

1047
	fn := func(response api.GenerateResponse) error {
Michael Yang's avatar
Michael Yang committed
1048
		p.StopAndClear()
1049

Patrick Devine's avatar
Patrick Devine committed
1050
		latest = response
1051
		content := response.Response
1052

1053
		displayResponse(content, opts.WordWrap, state)
1054

Patrick Devine's avatar
Patrick Devine committed
1055
1056
		return nil
	}
1057

1058
1059
1060
1061
1062
1063
1064
	if opts.MultiModal {
		opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
		if err != nil {
			return err
		}
	}

1065
1066
1067
1068
	if opts.Format == "json" {
		opts.Format = `"` + opts.Format + `"`
	}

Michael Yang's avatar
Michael Yang committed
1069
	request := api.GenerateRequest{
1070
1071
1072
1073
		Model:     opts.Model,
		Prompt:    opts.Prompt,
		Context:   generateContext,
		Images:    opts.Images,
1074
		Format:    json.RawMessage(opts.Format),
1075
1076
1077
		System:    opts.System,
		Options:   opts.Options,
		KeepAlive: opts.KeepAlive,
Michael Yang's avatar
Michael Yang committed
1078
1079
1080
	}

	if err := client.Generate(ctx, &request, fn); err != nil {
1081
		if errors.Is(err, context.Canceled) {
1082
			return nil
1083
		}
1084
		return err
Patrick Devine's avatar
Patrick Devine committed
1085
	}
1086

1087
	if opts.Prompt != "" {
Michael Yang's avatar
Michael Yang committed
1088
1089
		fmt.Println()
		fmt.Println()
Patrick Devine's avatar
Patrick Devine committed
1090
	}
1091

1092
1093
1094
1095
	if !latest.Done {
		return nil
	}

Patrick Devine's avatar
Patrick Devine committed
1096
1097
	verbose, err := cmd.Flags().GetBool("verbose")
	if err != nil {
1098
		return err
Patrick Devine's avatar
Patrick Devine committed
1099
	}
Michael Yang's avatar
Michael Yang committed
1100

Patrick Devine's avatar
Patrick Devine committed
1101
1102
	if verbose {
		latest.Summary()
Michael Yang's avatar
Michael Yang committed
1103
	}
Michael Yang's avatar
Michael Yang committed
1104

Patrick Devine's avatar
Patrick Devine committed
1105
1106
1107
	ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
	cmd.SetContext(ctx)

1108
	return nil
Michael Yang's avatar
Michael Yang committed
1109
1110
}

1111
func RunServer(_ *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1112
	if err := initializeKeypair(); err != nil {
1113
1114
1115
		return err
	}

Michael Yang's avatar
host  
Michael Yang committed
1116
	ln, err := net.Listen("tcp", envconfig.Host().Host)
1117
1118
1119
	if err != nil {
		return err
	}
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1120

1121
1122
1123
1124
1125
1126
	err = server.Serve(ln)
	if errors.Is(err, http.ErrServerClosed) {
		return nil
	}

	return err
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1127
1128
}

1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
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
1141
		cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
1142
1143
1144
1145
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1146
		privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
1147
1148
1149
1150
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1151
		if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
1152
1153
1154
			return fmt.Errorf("could not create directory %w", err)
		}

Michael Yang's avatar
Michael Yang committed
1155
		if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
1156
1157
1158
			return err
		}

Michael Yang's avatar
Michael Yang committed
1159
		sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
1160
1161
1162
1163
		if err != nil {
			return err
		}

Michael Yang's avatar
Michael Yang committed
1164
		publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
1165

Michael Yang's avatar
Michael Yang committed
1166
		if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
1167
1168
1169
			return err
		}

Michael Yang's avatar
Michael Yang committed
1170
		fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
1171
1172
1173
1174
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1175
func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
Michael Yang's avatar
Michael Yang committed
1176
	client, err := api.ClientFromEnvironment()
1177
1178
1179
	if err != nil {
		return err
	}
Michael Yang's avatar
Michael Yang committed
1180
	if err := client.Heartbeat(cmd.Context()); err != nil {
1181
		if !strings.Contains(err.Error(), " refused") {
Bruce MacDonald's avatar
Bruce MacDonald committed
1182
1183
			return err
		}
1184
		if err := startApp(cmd.Context(), client); err != nil {
Michael Yang's avatar
lint  
Michael Yang committed
1185
			return errors.New("could not connect to ollama app, is it running?")
1186
1187
1188
1189
1190
		}
	}
	return nil
}

Michael Yang's avatar
Michael Yang committed
1191
1192
1193
1194
1195
1196
1197
1198
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
1199
1200
1201
1202
1203
		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
1204
1205
	}

1206
	if serverVersion != version.Version {
Michael Yang's avatar
Michael Yang committed
1207
		fmt.Printf("Warning: client version is %s\n", version.Version)
1208
	}
Michael Yang's avatar
Michael Yang committed
1209
1210
}

1211
func appendEnvDocs(cmd *cobra.Command, envs []envconfig.EnvVar) {
1212
1213
1214
1215
1216
	if len(envs) == 0 {
		return
	}

	envUsage := `
1217
1218
Environment Variables:
`
1219
	for _, e := range envs {
1220
		envUsage += fmt.Sprintf("      %-24s   %s\n", e.Name, e.Description)
1221
1222
1223
	}

	cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
1224
1225
}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1226
1227
func NewCLI() *cobra.Command {
	log.SetFlags(log.LstdFlags | log.Lshortfile)
Michael Yang's avatar
Michael Yang committed
1228
	cobra.EnableCommandSorting = false
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1229

1230
	if runtime.GOOS == "windows" && term.IsTerminal(int(os.Stdout.Fd())) {
1231
		console.ConsoleFromFile(os.Stdin) //nolint:errcheck
1232
1233
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1234
	rootCmd := &cobra.Command{
1235
1236
1237
1238
		Use:           "ollama",
		Short:         "Large language model runner",
		SilenceUsage:  true,
		SilenceErrors: true,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1239
1240
1241
		CompletionOptions: cobra.CompletionOptions{
			DisableDefaultCmd: true,
		},
Michael Yang's avatar
Michael Yang committed
1242
1243
1244
1245
1246
1247
1248
1249
		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
1250
1251
	}

Michael Yang's avatar
Michael Yang committed
1252
	rootCmd.Flags().BoolP("version", "v", false, "Show version information")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1253

1254
	createCmd := &cobra.Command{
1255
1256
		Use:     "create MODEL",
		Short:   "Create a model from a Modelfile",
Michael Yang's avatar
Michael Yang committed
1257
		Args:    cobra.ExactArgs(1),
1258
1259
		PreRunE: checkServerHeartbeat,
		RunE:    CreateHandler,
1260
1261
	}

1262
	createCmd.Flags().StringP("file", "f", "", "Name of the Modelfile (default \"Modelfile\"")
1263
	createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_0)")
1264

Patrick Devine's avatar
Patrick Devine committed
1265
1266
1267
	showCmd := &cobra.Command{
		Use:     "show MODEL",
		Short:   "Show information for a model",
Michael Yang's avatar
Michael Yang committed
1268
		Args:    cobra.ExactArgs(1),
Patrick Devine's avatar
Patrick Devine committed
1269
1270
1271
1272
1273
1274
1275
1276
		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")
1277
	showCmd.Flags().Bool("system", false, "Show system message of a model")
1278
	showCmd.Flags().BoolP("verbose", "v", false, "Show detailed model information")
Patrick Devine's avatar
Patrick Devine committed
1279

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1280
	runCmd := &cobra.Command{
1281
1282
1283
1284
1285
		Use:     "run MODEL [PROMPT]",
		Short:   "Run a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    RunHandler,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1286
1287
	}

1288
	runCmd.Flags().String("keepalive", "", "Duration to keep a model loaded (e.g. 5m)")
1289
	runCmd.Flags().Bool("verbose", false, "Show timings for response")
1290
	runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
1291
	runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1292
	runCmd.Flags().String("format", "", "Response format (e.g. json)")
Patrick Devine's avatar
Patrick Devine committed
1293
1294
1295
1296
1297
1298
1299
1300
1301

	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
1302
1303
1304
1305
	serveCmd := &cobra.Command{
		Use:     "serve",
		Aliases: []string{"start"},
		Short:   "Start ollama",
Michael Yang's avatar
Michael Yang committed
1306
		Args:    cobra.ExactArgs(0),
Michael Yang's avatar
Michael Yang committed
1307
		RunE:    RunServer,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1308
1309
	}

1310
	pullCmd := &cobra.Command{
1311
1312
		Use:     "pull MODEL",
		Short:   "Pull a model from a registry",
Michael Yang's avatar
Michael Yang committed
1313
		Args:    cobra.ExactArgs(1),
1314
1315
		PreRunE: checkServerHeartbeat,
		RunE:    PullHandler,
1316
1317
	}

1318
1319
	pullCmd.Flags().Bool("insecure", false, "Use an insecure registry")

1320
	pushCmd := &cobra.Command{
1321
1322
		Use:     "push MODEL",
		Short:   "Push a model to a registry",
Michael Yang's avatar
Michael Yang committed
1323
		Args:    cobra.ExactArgs(1),
1324
1325
		PreRunE: checkServerHeartbeat,
		RunE:    PushHandler,
1326
1327
	}

1328
1329
	pushCmd.Flags().Bool("insecure", false, "Use an insecure registry")

Patrick Devine's avatar
Patrick Devine committed
1330
	listCmd := &cobra.Command{
1331
		Use:     "list",
Patrick Devine's avatar
Patrick Devine committed
1332
		Aliases: []string{"ls"},
1333
		Short:   "List models",
1334
		PreRunE: checkServerHeartbeat,
1335
		RunE:    ListHandler,
1336
	}
1337
1338
1339
1340
1341
1342
1343
1344

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

Patrick Devine's avatar
Patrick Devine committed
1345
	copyCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1346
		Use:     "cp SOURCE DESTINATION",
1347
		Short:   "Copy a model",
Michael Yang's avatar
Michael Yang committed
1348
		Args:    cobra.ExactArgs(2),
1349
1350
		PreRunE: checkServerHeartbeat,
		RunE:    CopyHandler,
Patrick Devine's avatar
Patrick Devine committed
1351
1352
	}

1353
	deleteCmd := &cobra.Command{
Michael Yang's avatar
Michael Yang committed
1354
		Use:     "rm MODEL [MODEL...]",
1355
1356
1357
1358
		Short:   "Remove a model",
		Args:    cobra.MinimumNArgs(1),
		PreRunE: checkServerHeartbeat,
		RunE:    DeleteHandler,
Patrick Devine's avatar
Patrick Devine committed
1359
1360
	}

1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
	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:])
	})

1373
1374
1375
	envVars := envconfig.AsMap()

	envs := []envconfig.EnvVar{envVars["OLLAMA_HOST"]}
1376

1377
1378
1379
1380
	for _, cmd := range []*cobra.Command{
		createCmd,
		showCmd,
		runCmd,
Patrick Devine's avatar
Patrick Devine committed
1381
		stopCmd,
1382
1383
1384
		pullCmd,
		pushCmd,
		listCmd,
1385
		psCmd,
1386
1387
		copyCmd,
		deleteCmd,
1388
		serveCmd,
1389
	} {
1390
1391
		switch cmd {
		case runCmd:
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
			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"],
1404
				envVars["OLLAMA_SCHED_SPREAD"],
1405
				envVars["OLLAMA_FLASH_ATTENTION"],
1406
				envVars["OLLAMA_KV_CACHE_TYPE"],
1407
				envVars["OLLAMA_LLM_LIBRARY"],
1408
				envVars["OLLAMA_GPU_OVERHEAD"],
1409
				envVars["OLLAMA_LOAD_TIMEOUT"],
1410
			})
1411
1412
1413
		default:
			appendEnvDocs(cmd, envs)
		}
1414
1415
	}

Jeffrey Morgan's avatar
Jeffrey Morgan committed
1416
1417
	rootCmd.AddCommand(
		serveCmd,
1418
		createCmd,
Patrick Devine's avatar
Patrick Devine committed
1419
		showCmd,
1420
		runCmd,
Patrick Devine's avatar
Patrick Devine committed
1421
		stopCmd,
1422
1423
		pullCmd,
		pushCmd,
Patrick Devine's avatar
Patrick Devine committed
1424
		listCmd,
1425
		psCmd,
Patrick Devine's avatar
Patrick Devine committed
1426
		copyCmd,
1427
		deleteCmd,
1428
		runnerCmd,
Jeffrey Morgan's avatar
Jeffrey Morgan committed
1429
1430
1431
1432
	)

	return rootCmd
}