cmd.go 33 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412
  1. package cmd
  2. import (
  3. "archive/zip"
  4. "bufio"
  5. "bytes"
  6. "context"
  7. "crypto/ed25519"
  8. "crypto/rand"
  9. "crypto/sha256"
  10. "encoding/pem"
  11. "errors"
  12. "fmt"
  13. "io"
  14. "log"
  15. "math"
  16. "net"
  17. "net/http"
  18. "os"
  19. "os/signal"
  20. "path/filepath"
  21. "regexp"
  22. "runtime"
  23. "slices"
  24. "strconv"
  25. "strings"
  26. "sync/atomic"
  27. "syscall"
  28. "time"
  29. "github.com/containerd/console"
  30. "github.com/mattn/go-runewidth"
  31. "github.com/olekukonko/tablewriter"
  32. "github.com/spf13/cobra"
  33. "golang.org/x/crypto/ssh"
  34. "golang.org/x/term"
  35. "github.com/ollama/ollama/api"
  36. "github.com/ollama/ollama/auth"
  37. "github.com/ollama/ollama/envconfig"
  38. "github.com/ollama/ollama/format"
  39. "github.com/ollama/ollama/parser"
  40. "github.com/ollama/ollama/progress"
  41. "github.com/ollama/ollama/server"
  42. "github.com/ollama/ollama/types/errtypes"
  43. "github.com/ollama/ollama/types/model"
  44. "github.com/ollama/ollama/version"
  45. )
  46. func CreateHandler(cmd *cobra.Command, args []string) error {
  47. filename, _ := cmd.Flags().GetString("file")
  48. filename, err := filepath.Abs(filename)
  49. if err != nil {
  50. return err
  51. }
  52. client, err := api.ClientFromEnvironment()
  53. if err != nil {
  54. return err
  55. }
  56. p := progress.NewProgress(os.Stderr)
  57. defer p.Stop()
  58. f, err := os.Open(filename)
  59. if err != nil {
  60. return err
  61. }
  62. defer f.Close()
  63. modelfile, err := parser.ParseFile(f)
  64. if err != nil {
  65. return err
  66. }
  67. home, err := os.UserHomeDir()
  68. if err != nil {
  69. return err
  70. }
  71. status := "transferring model data"
  72. spinner := progress.NewSpinner(status)
  73. p.Add(status, spinner)
  74. defer p.Stop()
  75. for i := range modelfile.Commands {
  76. switch modelfile.Commands[i].Name {
  77. case "model", "adapter":
  78. path := modelfile.Commands[i].Args
  79. if path == "~" {
  80. path = home
  81. } else if strings.HasPrefix(path, "~/") {
  82. path = filepath.Join(home, path[2:])
  83. }
  84. if !filepath.IsAbs(path) {
  85. path = filepath.Join(filepath.Dir(filename), path)
  86. }
  87. fi, err := os.Stat(path)
  88. if errors.Is(err, os.ErrNotExist) && modelfile.Commands[i].Name == "model" {
  89. continue
  90. } else if err != nil {
  91. return err
  92. }
  93. if fi.IsDir() {
  94. // this is likely a safetensors or pytorch directory
  95. // TODO make this work w/ adapters
  96. tempfile, err := tempZipFiles(path)
  97. if err != nil {
  98. return err
  99. }
  100. defer os.RemoveAll(tempfile)
  101. path = tempfile
  102. }
  103. digest, err := createBlob(cmd, client, path, spinner)
  104. if err != nil {
  105. return err
  106. }
  107. modelfile.Commands[i].Args = "@" + digest
  108. }
  109. }
  110. bars := make(map[string]*progress.Bar)
  111. fn := func(resp api.ProgressResponse) error {
  112. if resp.Digest != "" {
  113. spinner.Stop()
  114. bar, ok := bars[resp.Digest]
  115. if !ok {
  116. bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
  117. bars[resp.Digest] = bar
  118. p.Add(resp.Digest, bar)
  119. }
  120. bar.Set(resp.Completed)
  121. } else if status != resp.Status {
  122. spinner.Stop()
  123. status = resp.Status
  124. spinner = progress.NewSpinner(status)
  125. p.Add(status, spinner)
  126. }
  127. return nil
  128. }
  129. quantize, _ := cmd.Flags().GetString("quantize")
  130. request := api.CreateRequest{Name: args[0], Modelfile: modelfile.String(), Quantize: quantize}
  131. if err := client.Create(cmd.Context(), &request, fn); err != nil {
  132. return err
  133. }
  134. return nil
  135. }
  136. func tempZipFiles(path string) (string, error) {
  137. tempfile, err := os.CreateTemp("", "ollama-tf")
  138. if err != nil {
  139. return "", err
  140. }
  141. defer tempfile.Close()
  142. detectContentType := func(path string) (string, error) {
  143. f, err := os.Open(path)
  144. if err != nil {
  145. return "", err
  146. }
  147. defer f.Close()
  148. var b bytes.Buffer
  149. b.Grow(512)
  150. if _, err := io.CopyN(&b, f, 512); err != nil && !errors.Is(err, io.EOF) {
  151. return "", err
  152. }
  153. contentType, _, _ := strings.Cut(http.DetectContentType(b.Bytes()), ";")
  154. return contentType, nil
  155. }
  156. glob := func(pattern, contentType string) ([]string, error) {
  157. matches, err := filepath.Glob(pattern)
  158. if err != nil {
  159. return nil, err
  160. }
  161. for _, safetensor := range matches {
  162. if ct, err := detectContentType(safetensor); err != nil {
  163. return nil, err
  164. } else if ct != contentType {
  165. return nil, fmt.Errorf("invalid content type: expected %s for %s", ct, safetensor)
  166. }
  167. }
  168. return matches, nil
  169. }
  170. var files []string
  171. if st, _ := glob(filepath.Join(path, "model*.safetensors"), "application/octet-stream"); len(st) > 0 {
  172. // safetensors files might be unresolved git lfs references; skip if they are
  173. // covers model-x-of-y.safetensors, model.fp32-x-of-y.safetensors, model.safetensors
  174. files = append(files, st...)
  175. } else if st, _ := glob(filepath.Join(path, "adapters.safetensors"), "application/octet-stream"); len(st) > 0 {
  176. // covers adapters.safetensors
  177. files = append(files, st...)
  178. } else if st, _ := glob(filepath.Join(path, "adapter_model.safetensors"), "application/octet-stream"); len(st) > 0 {
  179. // covers adapter_model.safetensors
  180. files = append(files, st...)
  181. } else if pt, _ := glob(filepath.Join(path, "pytorch_model*.bin"), "application/zip"); len(pt) > 0 {
  182. // pytorch files might also be unresolved git lfs references; skip if they are
  183. // covers pytorch_model-x-of-y.bin, pytorch_model.fp32-x-of-y.bin, pytorch_model.bin
  184. files = append(files, pt...)
  185. } else if pt, _ := glob(filepath.Join(path, "consolidated*.pth"), "application/zip"); len(pt) > 0 {
  186. // pytorch files might also be unresolved git lfs references; skip if they are
  187. // covers consolidated.x.pth, consolidated.pth
  188. files = append(files, pt...)
  189. } else {
  190. return "", errors.New("no safetensors or torch files found")
  191. }
  192. // add configuration files, json files are detected as text/plain
  193. js, err := glob(filepath.Join(path, "*.json"), "text/plain")
  194. if err != nil {
  195. return "", err
  196. }
  197. files = append(files, js...)
  198. // bert models require a nested config.json
  199. // TODO(mxyng): merge this with the glob above
  200. js, err = glob(filepath.Join(path, "**/*.json"), "text/plain")
  201. if err != nil {
  202. return "", err
  203. }
  204. files = append(files, js...)
  205. if tks, _ := glob(filepath.Join(path, "tokenizer.model"), "application/octet-stream"); len(tks) > 0 {
  206. // add tokenizer.model if it exists, tokenizer.json is automatically picked up by the previous glob
  207. // tokenizer.model might be a unresolved git lfs reference; error if it is
  208. files = append(files, tks...)
  209. } else if tks, _ := glob(filepath.Join(path, "**/tokenizer.model"), "text/plain"); len(tks) > 0 {
  210. // some times tokenizer.model is in a subdirectory (e.g. meta-llama/Meta-Llama-3-8B)
  211. files = append(files, tks...)
  212. }
  213. zipfile := zip.NewWriter(tempfile)
  214. defer zipfile.Close()
  215. for _, file := range files {
  216. f, err := os.Open(file)
  217. if err != nil {
  218. return "", err
  219. }
  220. defer f.Close()
  221. fi, err := f.Stat()
  222. if err != nil {
  223. return "", err
  224. }
  225. zfi, err := zip.FileInfoHeader(fi)
  226. if err != nil {
  227. return "", err
  228. }
  229. zfi.Name, err = filepath.Rel(path, file)
  230. if err != nil {
  231. return "", err
  232. }
  233. zf, err := zipfile.CreateHeader(zfi)
  234. if err != nil {
  235. return "", err
  236. }
  237. if _, err := io.Copy(zf, f); err != nil {
  238. return "", err
  239. }
  240. }
  241. return tempfile.Name(), nil
  242. }
  243. func createBlob(cmd *cobra.Command, client *api.Client, path string, spinner *progress.Spinner) (string, error) {
  244. bin, err := os.Open(path)
  245. if err != nil {
  246. return "", err
  247. }
  248. defer bin.Close()
  249. // Get file info to retrieve the size
  250. fileInfo, err := bin.Stat()
  251. if err != nil {
  252. return "", err
  253. }
  254. fileSize := fileInfo.Size()
  255. hash := sha256.New()
  256. if _, err := io.Copy(hash, bin); err != nil {
  257. return "", err
  258. }
  259. if _, err := bin.Seek(0, io.SeekStart); err != nil {
  260. return "", err
  261. }
  262. var pw progressWriter
  263. status := "transferring model data 0%"
  264. spinner.SetMessage(status)
  265. done := make(chan struct{})
  266. defer close(done)
  267. go func() {
  268. ticker := time.NewTicker(60 * time.Millisecond)
  269. defer ticker.Stop()
  270. for {
  271. select {
  272. case <-ticker.C:
  273. spinner.SetMessage(fmt.Sprintf("transferring model data %d%%", int(100*pw.n.Load()/fileSize)))
  274. case <-done:
  275. spinner.SetMessage("transferring model data 100%")
  276. return
  277. }
  278. }
  279. }()
  280. digest := fmt.Sprintf("sha256:%x", hash.Sum(nil))
  281. if err = client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
  282. return "", err
  283. }
  284. return digest, nil
  285. }
  286. type progressWriter struct {
  287. n atomic.Int64
  288. }
  289. func (w *progressWriter) Write(p []byte) (n int, err error) {
  290. w.n.Add(int64(len(p)))
  291. return len(p), nil
  292. }
  293. func RunHandler(cmd *cobra.Command, args []string) error {
  294. interactive := true
  295. opts := runOptions{
  296. Model: args[0],
  297. WordWrap: os.Getenv("TERM") == "xterm-256color",
  298. Options: map[string]interface{}{},
  299. }
  300. format, err := cmd.Flags().GetString("format")
  301. if err != nil {
  302. return err
  303. }
  304. opts.Format = format
  305. keepAlive, err := cmd.Flags().GetString("keepalive")
  306. if err != nil {
  307. return err
  308. }
  309. if keepAlive != "" {
  310. d, err := time.ParseDuration(keepAlive)
  311. if err != nil {
  312. return err
  313. }
  314. opts.KeepAlive = &api.Duration{Duration: d}
  315. }
  316. prompts := args[1:]
  317. // prepend stdin to the prompt if provided
  318. if !term.IsTerminal(int(os.Stdin.Fd())) {
  319. in, err := io.ReadAll(os.Stdin)
  320. if err != nil {
  321. return err
  322. }
  323. prompts = append([]string{string(in)}, prompts...)
  324. opts.WordWrap = false
  325. interactive = false
  326. }
  327. opts.Prompt = strings.Join(prompts, " ")
  328. if len(prompts) > 0 {
  329. interactive = false
  330. }
  331. nowrap, err := cmd.Flags().GetBool("nowordwrap")
  332. if err != nil {
  333. return err
  334. }
  335. opts.WordWrap = !nowrap
  336. // Fill out the rest of the options based on information about the
  337. // model.
  338. client, err := api.ClientFromEnvironment()
  339. if err != nil {
  340. return err
  341. }
  342. name := args[0]
  343. info, err := func() (*api.ShowResponse, error) {
  344. showReq := &api.ShowRequest{Name: name}
  345. info, err := client.Show(cmd.Context(), showReq)
  346. var se api.StatusError
  347. if errors.As(err, &se) && se.StatusCode == http.StatusNotFound {
  348. if err := PullHandler(cmd, []string{name}); err != nil {
  349. return nil, err
  350. }
  351. return client.Show(cmd.Context(), &api.ShowRequest{Name: name})
  352. }
  353. return info, err
  354. }()
  355. if err != nil {
  356. return err
  357. }
  358. opts.MultiModal = slices.Contains(info.Details.Families, "clip")
  359. opts.ParentModel = info.Details.ParentModel
  360. if interactive {
  361. if err := loadModel(cmd, &opts); err != nil {
  362. return err
  363. }
  364. for _, msg := range info.Messages {
  365. switch msg.Role {
  366. case "user":
  367. fmt.Printf(">>> %s\n", msg.Content)
  368. case "assistant":
  369. state := &displayResponseState{}
  370. displayResponse(msg.Content, opts.WordWrap, state)
  371. fmt.Println()
  372. fmt.Println()
  373. }
  374. }
  375. return generateInteractive(cmd, opts)
  376. }
  377. return generate(cmd, opts)
  378. }
  379. func errFromUnknownKey(unknownKeyErr error) error {
  380. // find SSH public key in the error message
  381. sshKeyPattern := `ssh-\w+ [^\s"]+`
  382. re := regexp.MustCompile(sshKeyPattern)
  383. matches := re.FindStringSubmatch(unknownKeyErr.Error())
  384. if len(matches) > 0 {
  385. serverPubKey := matches[0]
  386. localPubKey, err := auth.GetPublicKey()
  387. if err != nil {
  388. return unknownKeyErr
  389. }
  390. if runtime.GOOS == "linux" && serverPubKey != localPubKey {
  391. // try the ollama service public key
  392. svcPubKey, err := os.ReadFile("/usr/share/ollama/.ollama/id_ed25519.pub")
  393. if err != nil {
  394. return unknownKeyErr
  395. }
  396. localPubKey = strings.TrimSpace(string(svcPubKey))
  397. }
  398. // check if the returned public key matches the local public key, this prevents adding a remote key to the user's account
  399. if serverPubKey != localPubKey {
  400. return unknownKeyErr
  401. }
  402. var msg strings.Builder
  403. msg.WriteString(unknownKeyErr.Error())
  404. msg.WriteString("\n\nYour ollama key is:\n")
  405. msg.WriteString(localPubKey)
  406. msg.WriteString("\nAdd your key at:\n")
  407. msg.WriteString("https://ollama.com/settings/keys")
  408. return errors.New(msg.String())
  409. }
  410. return unknownKeyErr
  411. }
  412. func PushHandler(cmd *cobra.Command, args []string) error {
  413. client, err := api.ClientFromEnvironment()
  414. if err != nil {
  415. return err
  416. }
  417. insecure, err := cmd.Flags().GetBool("insecure")
  418. if err != nil {
  419. return err
  420. }
  421. p := progress.NewProgress(os.Stderr)
  422. defer p.Stop()
  423. bars := make(map[string]*progress.Bar)
  424. var status string
  425. var spinner *progress.Spinner
  426. fn := func(resp api.ProgressResponse) error {
  427. if resp.Digest != "" {
  428. if spinner != nil {
  429. spinner.Stop()
  430. }
  431. bar, ok := bars[resp.Digest]
  432. if !ok {
  433. bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
  434. bars[resp.Digest] = bar
  435. p.Add(resp.Digest, bar)
  436. }
  437. bar.Set(resp.Completed)
  438. } else if status != resp.Status {
  439. if spinner != nil {
  440. spinner.Stop()
  441. }
  442. status = resp.Status
  443. spinner = progress.NewSpinner(status)
  444. p.Add(status, spinner)
  445. }
  446. return nil
  447. }
  448. request := api.PushRequest{Name: args[0], Insecure: insecure}
  449. if err := client.Push(cmd.Context(), &request, fn); err != nil {
  450. if spinner != nil {
  451. spinner.Stop()
  452. }
  453. if strings.Contains(err.Error(), "access denied") {
  454. return errors.New("you are not authorized to push to this namespace, create the model under a namespace you own")
  455. }
  456. host := model.ParseName(args[0]).Host
  457. isOllamaHost := strings.HasSuffix(host, ".ollama.ai") || strings.HasSuffix(host, ".ollama.com")
  458. if strings.Contains(err.Error(), errtypes.UnknownOllamaKeyErrMsg) && isOllamaHost {
  459. // the user has not added their ollama key to ollama.com
  460. // re-throw an error with a more user-friendly message
  461. return errFromUnknownKey(err)
  462. }
  463. return err
  464. }
  465. spinner.Stop()
  466. return nil
  467. }
  468. func ListHandler(cmd *cobra.Command, args []string) error {
  469. client, err := api.ClientFromEnvironment()
  470. if err != nil {
  471. return err
  472. }
  473. models, err := client.List(cmd.Context())
  474. if err != nil {
  475. return err
  476. }
  477. var data [][]string
  478. for _, m := range models.Models {
  479. if len(args) == 0 || strings.HasPrefix(m.Name, args[0]) {
  480. data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), format.HumanTime(m.ModifiedAt, "Never")})
  481. }
  482. }
  483. table := tablewriter.NewWriter(os.Stdout)
  484. table.SetHeader([]string{"NAME", "ID", "SIZE", "MODIFIED"})
  485. table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
  486. table.SetAlignment(tablewriter.ALIGN_LEFT)
  487. table.SetHeaderLine(false)
  488. table.SetBorder(false)
  489. table.SetNoWhiteSpace(true)
  490. table.SetTablePadding(" ")
  491. table.AppendBulk(data)
  492. table.Render()
  493. return nil
  494. }
  495. func ListRunningHandler(cmd *cobra.Command, args []string) error {
  496. client, err := api.ClientFromEnvironment()
  497. if err != nil {
  498. return err
  499. }
  500. models, err := client.ListRunning(cmd.Context())
  501. if err != nil {
  502. return err
  503. }
  504. var data [][]string
  505. for _, m := range models.Models {
  506. if len(args) == 0 || strings.HasPrefix(m.Name, args[0]) {
  507. var procStr string
  508. switch {
  509. case m.SizeVRAM == 0:
  510. procStr = "100% CPU"
  511. case m.SizeVRAM == m.Size:
  512. procStr = "100% GPU"
  513. case m.SizeVRAM > m.Size || m.Size == 0:
  514. procStr = "Unknown"
  515. default:
  516. sizeCPU := m.Size - m.SizeVRAM
  517. cpuPercent := math.Round(float64(sizeCPU) / float64(m.Size) * 100)
  518. procStr = fmt.Sprintf("%d%%/%d%% CPU/GPU", int(cpuPercent), int(100-cpuPercent))
  519. }
  520. data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), procStr, format.HumanTime(m.ExpiresAt, "Never")})
  521. }
  522. }
  523. table := tablewriter.NewWriter(os.Stdout)
  524. table.SetHeader([]string{"NAME", "ID", "SIZE", "PROCESSOR", "UNTIL"})
  525. table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
  526. table.SetAlignment(tablewriter.ALIGN_LEFT)
  527. table.SetHeaderLine(false)
  528. table.SetBorder(false)
  529. table.SetNoWhiteSpace(true)
  530. table.SetTablePadding(" ")
  531. table.AppendBulk(data)
  532. table.Render()
  533. return nil
  534. }
  535. func DeleteHandler(cmd *cobra.Command, args []string) error {
  536. client, err := api.ClientFromEnvironment()
  537. if err != nil {
  538. return err
  539. }
  540. for _, name := range args {
  541. req := api.DeleteRequest{Name: name}
  542. if err := client.Delete(cmd.Context(), &req); err != nil {
  543. return err
  544. }
  545. fmt.Printf("deleted '%s'\n", name)
  546. }
  547. return nil
  548. }
  549. func ShowHandler(cmd *cobra.Command, args []string) error {
  550. client, err := api.ClientFromEnvironment()
  551. if err != nil {
  552. return err
  553. }
  554. license, errLicense := cmd.Flags().GetBool("license")
  555. modelfile, errModelfile := cmd.Flags().GetBool("modelfile")
  556. parameters, errParams := cmd.Flags().GetBool("parameters")
  557. system, errSystem := cmd.Flags().GetBool("system")
  558. template, errTemplate := cmd.Flags().GetBool("template")
  559. for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate} {
  560. if boolErr != nil {
  561. return errors.New("error retrieving flags")
  562. }
  563. }
  564. flagsSet := 0
  565. showType := ""
  566. if license {
  567. flagsSet++
  568. showType = "license"
  569. }
  570. if modelfile {
  571. flagsSet++
  572. showType = "modelfile"
  573. }
  574. if parameters {
  575. flagsSet++
  576. showType = "parameters"
  577. }
  578. if system {
  579. flagsSet++
  580. showType = "system"
  581. }
  582. if template {
  583. flagsSet++
  584. showType = "template"
  585. }
  586. if flagsSet > 1 {
  587. return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
  588. }
  589. req := api.ShowRequest{Name: args[0]}
  590. resp, err := client.Show(cmd.Context(), &req)
  591. if err != nil {
  592. return err
  593. }
  594. if flagsSet == 1 {
  595. switch showType {
  596. case "license":
  597. fmt.Println(resp.License)
  598. case "modelfile":
  599. fmt.Println(resp.Modelfile)
  600. case "parameters":
  601. fmt.Println(resp.Parameters)
  602. case "system":
  603. fmt.Println(resp.System)
  604. case "template":
  605. fmt.Println(resp.Template)
  606. }
  607. return nil
  608. }
  609. return showInfo(resp, os.Stdout)
  610. }
  611. func showInfo(resp *api.ShowResponse, w io.Writer) error {
  612. tableRender := func(header string, rows func() [][]string) {
  613. fmt.Fprintln(w, " ", header)
  614. table := tablewriter.NewWriter(w)
  615. table.SetAlignment(tablewriter.ALIGN_LEFT)
  616. table.SetBorder(false)
  617. table.SetNoWhiteSpace(true)
  618. table.SetTablePadding(" ")
  619. switch header {
  620. case "Template", "System", "License":
  621. table.SetColWidth(100)
  622. }
  623. table.AppendBulk(rows())
  624. table.Render()
  625. fmt.Fprintln(w)
  626. }
  627. tableRender("Model", func() (rows [][]string) {
  628. if resp.ModelInfo != nil {
  629. arch := resp.ModelInfo["general.architecture"].(string)
  630. rows = append(rows, []string{"", "architecture", arch})
  631. rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ModelInfo["general.parameter_count"].(float64)))})
  632. rows = append(rows, []string{"", "context length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)].(float64), 'f', -1, 64)})
  633. rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.embedding_length", arch)].(float64), 'f', -1, 64)})
  634. } else {
  635. rows = append(rows, []string{"", "architecture", resp.Details.Family})
  636. rows = append(rows, []string{"", "parameters", resp.Details.ParameterSize})
  637. }
  638. rows = append(rows, []string{"", "quantization", resp.Details.QuantizationLevel})
  639. return
  640. })
  641. if resp.ProjectorInfo != nil {
  642. tableRender("Projector", func() (rows [][]string) {
  643. arch := resp.ProjectorInfo["general.architecture"].(string)
  644. rows = append(rows, []string{"", "architecture", arch})
  645. rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ProjectorInfo["general.parameter_count"].(float64)))})
  646. rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.embedding_length", arch)].(float64), 'f', -1, 64)})
  647. rows = append(rows, []string{"", "dimensions", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.projection_dim", arch)].(float64), 'f', -1, 64)})
  648. return
  649. })
  650. }
  651. if resp.Parameters != "" {
  652. tableRender("Parameters", func() (rows [][]string) {
  653. scanner := bufio.NewScanner(strings.NewReader(resp.Parameters))
  654. for scanner.Scan() {
  655. if text := scanner.Text(); text != "" {
  656. rows = append(rows, append([]string{""}, strings.Fields(text)...))
  657. }
  658. }
  659. return
  660. })
  661. }
  662. head := func(s string, n int) (rows [][]string) {
  663. scanner := bufio.NewScanner(strings.NewReader(s))
  664. for scanner.Scan() && (len(rows) < n || n < 0) {
  665. if text := scanner.Text(); text != "" {
  666. rows = append(rows, []string{"", strings.TrimSpace(text)})
  667. }
  668. }
  669. return
  670. }
  671. if resp.System != "" {
  672. tableRender("System", func() [][]string {
  673. return head(resp.System, 2)
  674. })
  675. }
  676. if resp.License != "" {
  677. tableRender("License", func() [][]string {
  678. return head(resp.License, 2)
  679. })
  680. }
  681. return nil
  682. }
  683. func CopyHandler(cmd *cobra.Command, args []string) error {
  684. client, err := api.ClientFromEnvironment()
  685. if err != nil {
  686. return err
  687. }
  688. req := api.CopyRequest{Source: args[0], Destination: args[1]}
  689. if err := client.Copy(cmd.Context(), &req); err != nil {
  690. return err
  691. }
  692. fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
  693. return nil
  694. }
  695. func PullHandler(cmd *cobra.Command, args []string) error {
  696. insecure, err := cmd.Flags().GetBool("insecure")
  697. if err != nil {
  698. return err
  699. }
  700. client, err := api.ClientFromEnvironment()
  701. if err != nil {
  702. return err
  703. }
  704. p := progress.NewProgress(os.Stderr)
  705. defer p.Stop()
  706. bars := make(map[string]*progress.Bar)
  707. var status string
  708. var spinner *progress.Spinner
  709. fn := func(resp api.ProgressResponse) error {
  710. if resp.Digest != "" {
  711. if spinner != nil {
  712. spinner.Stop()
  713. }
  714. bar, ok := bars[resp.Digest]
  715. if !ok {
  716. bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
  717. bars[resp.Digest] = bar
  718. p.Add(resp.Digest, bar)
  719. }
  720. bar.Set(resp.Completed)
  721. } else if status != resp.Status {
  722. if spinner != nil {
  723. spinner.Stop()
  724. }
  725. status = resp.Status
  726. spinner = progress.NewSpinner(status)
  727. p.Add(status, spinner)
  728. }
  729. return nil
  730. }
  731. request := api.PullRequest{Name: args[0], Insecure: insecure}
  732. if err := client.Pull(cmd.Context(), &request, fn); err != nil {
  733. return err
  734. }
  735. return nil
  736. }
  737. type generateContextKey string
  738. type runOptions struct {
  739. Model string
  740. ParentModel string
  741. Prompt string
  742. Messages []api.Message
  743. WordWrap bool
  744. Format string
  745. System string
  746. Images []api.ImageData
  747. Options map[string]interface{}
  748. MultiModal bool
  749. KeepAlive *api.Duration
  750. }
  751. type displayResponseState struct {
  752. lineLength int
  753. wordBuffer string
  754. }
  755. func displayResponse(content string, wordWrap bool, state *displayResponseState) {
  756. termWidth, _, _ := term.GetSize(int(os.Stdout.Fd()))
  757. if wordWrap && termWidth >= 10 {
  758. for _, ch := range content {
  759. if state.lineLength+1 > termWidth-5 {
  760. if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
  761. fmt.Printf("%s%c", state.wordBuffer, ch)
  762. state.wordBuffer = ""
  763. state.lineLength = 0
  764. continue
  765. }
  766. // backtrack the length of the last word and clear to the end of the line
  767. a := runewidth.StringWidth(state.wordBuffer)
  768. if a > 0 {
  769. fmt.Printf("\x1b[%dD", a)
  770. }
  771. fmt.Printf("\x1b[K\n")
  772. fmt.Printf("%s%c", state.wordBuffer, ch)
  773. chWidth := runewidth.RuneWidth(ch)
  774. state.lineLength = runewidth.StringWidth(state.wordBuffer) + chWidth
  775. } else {
  776. fmt.Print(string(ch))
  777. state.lineLength += runewidth.RuneWidth(ch)
  778. if runewidth.RuneWidth(ch) >= 2 {
  779. state.wordBuffer = ""
  780. continue
  781. }
  782. switch ch {
  783. case ' ':
  784. state.wordBuffer = ""
  785. case '\n':
  786. state.lineLength = 0
  787. default:
  788. state.wordBuffer += string(ch)
  789. }
  790. }
  791. }
  792. } else {
  793. fmt.Printf("%s%s", state.wordBuffer, content)
  794. if len(state.wordBuffer) > 0 {
  795. state.wordBuffer = ""
  796. }
  797. }
  798. }
  799. func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
  800. client, err := api.ClientFromEnvironment()
  801. if err != nil {
  802. return nil, err
  803. }
  804. p := progress.NewProgress(os.Stderr)
  805. defer p.StopAndClear()
  806. spinner := progress.NewSpinner("")
  807. p.Add("", spinner)
  808. cancelCtx, cancel := context.WithCancel(cmd.Context())
  809. defer cancel()
  810. sigChan := make(chan os.Signal, 1)
  811. signal.Notify(sigChan, syscall.SIGINT)
  812. go func() {
  813. <-sigChan
  814. cancel()
  815. }()
  816. var state *displayResponseState = &displayResponseState{}
  817. var latest api.ChatResponse
  818. var fullResponse strings.Builder
  819. var role string
  820. fn := func(response api.ChatResponse) error {
  821. p.StopAndClear()
  822. latest = response
  823. role = response.Message.Role
  824. content := response.Message.Content
  825. fullResponse.WriteString(content)
  826. displayResponse(content, opts.WordWrap, state)
  827. return nil
  828. }
  829. req := &api.ChatRequest{
  830. Model: opts.Model,
  831. Messages: opts.Messages,
  832. Format: opts.Format,
  833. Options: opts.Options,
  834. }
  835. if opts.KeepAlive != nil {
  836. req.KeepAlive = opts.KeepAlive
  837. }
  838. if err := client.Chat(cancelCtx, req, fn); err != nil {
  839. if errors.Is(err, context.Canceled) {
  840. return nil, nil
  841. }
  842. return nil, err
  843. }
  844. if len(opts.Messages) > 0 {
  845. fmt.Println()
  846. fmt.Println()
  847. }
  848. verbose, err := cmd.Flags().GetBool("verbose")
  849. if err != nil {
  850. return nil, err
  851. }
  852. if verbose {
  853. latest.Summary()
  854. }
  855. return &api.Message{Role: role, Content: fullResponse.String()}, nil
  856. }
  857. func generate(cmd *cobra.Command, opts runOptions) error {
  858. client, err := api.ClientFromEnvironment()
  859. if err != nil {
  860. return err
  861. }
  862. p := progress.NewProgress(os.Stderr)
  863. defer p.StopAndClear()
  864. spinner := progress.NewSpinner("")
  865. p.Add("", spinner)
  866. var latest api.GenerateResponse
  867. generateContext, ok := cmd.Context().Value(generateContextKey("context")).([]int)
  868. if !ok {
  869. generateContext = []int{}
  870. }
  871. ctx, cancel := context.WithCancel(cmd.Context())
  872. defer cancel()
  873. sigChan := make(chan os.Signal, 1)
  874. signal.Notify(sigChan, syscall.SIGINT)
  875. go func() {
  876. <-sigChan
  877. cancel()
  878. }()
  879. var state *displayResponseState = &displayResponseState{}
  880. fn := func(response api.GenerateResponse) error {
  881. p.StopAndClear()
  882. latest = response
  883. content := response.Response
  884. displayResponse(content, opts.WordWrap, state)
  885. return nil
  886. }
  887. if opts.MultiModal {
  888. opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
  889. if err != nil {
  890. return err
  891. }
  892. }
  893. request := api.GenerateRequest{
  894. Model: opts.Model,
  895. Prompt: opts.Prompt,
  896. Context: generateContext,
  897. Images: opts.Images,
  898. Format: opts.Format,
  899. System: opts.System,
  900. Options: opts.Options,
  901. KeepAlive: opts.KeepAlive,
  902. }
  903. if err := client.Generate(ctx, &request, fn); err != nil {
  904. if errors.Is(err, context.Canceled) {
  905. return nil
  906. }
  907. return err
  908. }
  909. if opts.Prompt != "" {
  910. fmt.Println()
  911. fmt.Println()
  912. }
  913. if !latest.Done {
  914. return nil
  915. }
  916. verbose, err := cmd.Flags().GetBool("verbose")
  917. if err != nil {
  918. return err
  919. }
  920. if verbose {
  921. latest.Summary()
  922. }
  923. ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
  924. cmd.SetContext(ctx)
  925. return nil
  926. }
  927. func RunServer(_ *cobra.Command, _ []string) error {
  928. if err := initializeKeypair(); err != nil {
  929. return err
  930. }
  931. ln, err := net.Listen("tcp", envconfig.Host().Host)
  932. if err != nil {
  933. return err
  934. }
  935. err = server.Serve(ln)
  936. if errors.Is(err, http.ErrServerClosed) {
  937. return nil
  938. }
  939. return err
  940. }
  941. func initializeKeypair() error {
  942. home, err := os.UserHomeDir()
  943. if err != nil {
  944. return err
  945. }
  946. privKeyPath := filepath.Join(home, ".ollama", "id_ed25519")
  947. pubKeyPath := filepath.Join(home, ".ollama", "id_ed25519.pub")
  948. _, err = os.Stat(privKeyPath)
  949. if os.IsNotExist(err) {
  950. fmt.Printf("Couldn't find '%s'. Generating new private key.\n", privKeyPath)
  951. cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
  952. if err != nil {
  953. return err
  954. }
  955. privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
  956. if err != nil {
  957. return err
  958. }
  959. if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
  960. return fmt.Errorf("could not create directory %w", err)
  961. }
  962. if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
  963. return err
  964. }
  965. sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
  966. if err != nil {
  967. return err
  968. }
  969. publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
  970. if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
  971. return err
  972. }
  973. fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
  974. }
  975. return nil
  976. }
  977. func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
  978. client, err := api.ClientFromEnvironment()
  979. if err != nil {
  980. return err
  981. }
  982. if err := client.Heartbeat(cmd.Context()); err != nil {
  983. if !strings.Contains(err.Error(), " refused") {
  984. return err
  985. }
  986. if err := startApp(cmd.Context(), client); err != nil {
  987. return errors.New("could not connect to ollama app, is it running?")
  988. }
  989. }
  990. return nil
  991. }
  992. func versionHandler(cmd *cobra.Command, _ []string) {
  993. client, err := api.ClientFromEnvironment()
  994. if err != nil {
  995. return
  996. }
  997. serverVersion, err := client.Version(cmd.Context())
  998. if err != nil {
  999. fmt.Println("Warning: could not connect to a running Ollama instance")
  1000. }
  1001. if serverVersion != "" {
  1002. fmt.Printf("ollama version is %s\n", serverVersion)
  1003. }
  1004. if serverVersion != version.Version {
  1005. fmt.Printf("Warning: client version is %s\n", version.Version)
  1006. }
  1007. }
  1008. func appendEnvDocs(cmd *cobra.Command, envs []envconfig.EnvVar) {
  1009. if len(envs) == 0 {
  1010. return
  1011. }
  1012. envUsage := `
  1013. Environment Variables:
  1014. `
  1015. for _, e := range envs {
  1016. envUsage += fmt.Sprintf(" %-24s %s\n", e.Name, e.Description)
  1017. }
  1018. cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
  1019. }
  1020. func NewCLI() *cobra.Command {
  1021. log.SetFlags(log.LstdFlags | log.Lshortfile)
  1022. cobra.EnableCommandSorting = false
  1023. if runtime.GOOS == "windows" {
  1024. console.ConsoleFromFile(os.Stdin) //nolint:errcheck
  1025. }
  1026. rootCmd := &cobra.Command{
  1027. Use: "ollama",
  1028. Short: "Large language model runner",
  1029. SilenceUsage: true,
  1030. SilenceErrors: true,
  1031. CompletionOptions: cobra.CompletionOptions{
  1032. DisableDefaultCmd: true,
  1033. },
  1034. Run: func(cmd *cobra.Command, args []string) {
  1035. if version, _ := cmd.Flags().GetBool("version"); version {
  1036. versionHandler(cmd, args)
  1037. return
  1038. }
  1039. cmd.Print(cmd.UsageString())
  1040. },
  1041. }
  1042. rootCmd.Flags().BoolP("version", "v", false, "Show version information")
  1043. createCmd := &cobra.Command{
  1044. Use: "create MODEL",
  1045. Short: "Create a model from a Modelfile",
  1046. Args: cobra.ExactArgs(1),
  1047. PreRunE: checkServerHeartbeat,
  1048. RunE: CreateHandler,
  1049. }
  1050. createCmd.Flags().StringP("file", "f", "Modelfile", "Name of the Modelfile")
  1051. createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_0)")
  1052. showCmd := &cobra.Command{
  1053. Use: "show MODEL",
  1054. Short: "Show information for a model",
  1055. Args: cobra.ExactArgs(1),
  1056. PreRunE: checkServerHeartbeat,
  1057. RunE: ShowHandler,
  1058. }
  1059. showCmd.Flags().Bool("license", false, "Show license of a model")
  1060. showCmd.Flags().Bool("modelfile", false, "Show Modelfile of a model")
  1061. showCmd.Flags().Bool("parameters", false, "Show parameters of a model")
  1062. showCmd.Flags().Bool("template", false, "Show template of a model")
  1063. showCmd.Flags().Bool("system", false, "Show system message of a model")
  1064. runCmd := &cobra.Command{
  1065. Use: "run MODEL [PROMPT]",
  1066. Short: "Run a model",
  1067. Args: cobra.MinimumNArgs(1),
  1068. PreRunE: checkServerHeartbeat,
  1069. RunE: RunHandler,
  1070. }
  1071. runCmd.Flags().String("keepalive", "", "Duration to keep a model loaded (e.g. 5m)")
  1072. runCmd.Flags().Bool("verbose", false, "Show timings for response")
  1073. runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
  1074. runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
  1075. runCmd.Flags().String("format", "", "Response format (e.g. json)")
  1076. serveCmd := &cobra.Command{
  1077. Use: "serve",
  1078. Aliases: []string{"start"},
  1079. Short: "Start ollama",
  1080. Args: cobra.ExactArgs(0),
  1081. RunE: RunServer,
  1082. }
  1083. pullCmd := &cobra.Command{
  1084. Use: "pull MODEL",
  1085. Short: "Pull a model from a registry",
  1086. Args: cobra.ExactArgs(1),
  1087. PreRunE: checkServerHeartbeat,
  1088. RunE: PullHandler,
  1089. }
  1090. pullCmd.Flags().Bool("insecure", false, "Use an insecure registry")
  1091. pushCmd := &cobra.Command{
  1092. Use: "push MODEL",
  1093. Short: "Push a model to a registry",
  1094. Args: cobra.ExactArgs(1),
  1095. PreRunE: checkServerHeartbeat,
  1096. RunE: PushHandler,
  1097. }
  1098. pushCmd.Flags().Bool("insecure", false, "Use an insecure registry")
  1099. listCmd := &cobra.Command{
  1100. Use: "list",
  1101. Aliases: []string{"ls"},
  1102. Short: "List models",
  1103. PreRunE: checkServerHeartbeat,
  1104. RunE: ListHandler,
  1105. }
  1106. psCmd := &cobra.Command{
  1107. Use: "ps",
  1108. Short: "List running models",
  1109. PreRunE: checkServerHeartbeat,
  1110. RunE: ListRunningHandler,
  1111. }
  1112. copyCmd := &cobra.Command{
  1113. Use: "cp SOURCE DESTINATION",
  1114. Short: "Copy a model",
  1115. Args: cobra.ExactArgs(2),
  1116. PreRunE: checkServerHeartbeat,
  1117. RunE: CopyHandler,
  1118. }
  1119. deleteCmd := &cobra.Command{
  1120. Use: "rm MODEL [MODEL...]",
  1121. Short: "Remove a model",
  1122. Args: cobra.MinimumNArgs(1),
  1123. PreRunE: checkServerHeartbeat,
  1124. RunE: DeleteHandler,
  1125. }
  1126. envVars := envconfig.AsMap()
  1127. envs := []envconfig.EnvVar{envVars["OLLAMA_HOST"]}
  1128. for _, cmd := range []*cobra.Command{
  1129. createCmd,
  1130. showCmd,
  1131. runCmd,
  1132. pullCmd,
  1133. pushCmd,
  1134. listCmd,
  1135. psCmd,
  1136. copyCmd,
  1137. deleteCmd,
  1138. serveCmd,
  1139. } {
  1140. switch cmd {
  1141. case runCmd:
  1142. appendEnvDocs(cmd, []envconfig.EnvVar{envVars["OLLAMA_HOST"], envVars["OLLAMA_NOHISTORY"]})
  1143. case serveCmd:
  1144. appendEnvDocs(cmd, []envconfig.EnvVar{
  1145. envVars["OLLAMA_DEBUG"],
  1146. envVars["OLLAMA_HOST"],
  1147. envVars["OLLAMA_KEEP_ALIVE"],
  1148. envVars["OLLAMA_MAX_LOADED_MODELS"],
  1149. envVars["OLLAMA_MAX_QUEUE"],
  1150. envVars["OLLAMA_MODELS"],
  1151. envVars["OLLAMA_NUM_PARALLEL"],
  1152. envVars["OLLAMA_NOPRUNE"],
  1153. envVars["OLLAMA_ORIGINS"],
  1154. envVars["OLLAMA_SCHED_SPREAD"],
  1155. envVars["OLLAMA_TMPDIR"],
  1156. envVars["OLLAMA_FLASH_ATTENTION"],
  1157. envVars["OLLAMA_LLM_LIBRARY"],
  1158. envVars["OLLAMA_GPU_OVERHEAD"],
  1159. envVars["OLLAMA_LOAD_TIMEOUT"],
  1160. })
  1161. default:
  1162. appendEnvDocs(cmd, envs)
  1163. }
  1164. }
  1165. rootCmd.AddCommand(
  1166. serveCmd,
  1167. createCmd,
  1168. showCmd,
  1169. runCmd,
  1170. pullCmd,
  1171. pushCmd,
  1172. listCmd,
  1173. psCmd,
  1174. copyCmd,
  1175. deleteCmd,
  1176. )
  1177. return rootCmd
  1178. }