cmd.go 31 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355
  1. package cmd
  2. import (
  3. "bufio"
  4. "context"
  5. "crypto/ed25519"
  6. "crypto/rand"
  7. "encoding/json"
  8. "encoding/pem"
  9. "errors"
  10. "fmt"
  11. "io"
  12. "log"
  13. "math"
  14. "net"
  15. "net/http"
  16. "os"
  17. "os/signal"
  18. "path/filepath"
  19. "runtime"
  20. "strconv"
  21. "strings"
  22. "sync/atomic"
  23. "syscall"
  24. "time"
  25. "github.com/containerd/console"
  26. "github.com/mattn/go-runewidth"
  27. "github.com/olekukonko/tablewriter"
  28. "github.com/spf13/cobra"
  29. "golang.org/x/crypto/ssh"
  30. "golang.org/x/term"
  31. "github.com/ollama/ollama/api"
  32. "github.com/ollama/ollama/envconfig"
  33. "github.com/ollama/ollama/format"
  34. "github.com/ollama/ollama/llama"
  35. "github.com/ollama/ollama/parser"
  36. "github.com/ollama/ollama/progress"
  37. "github.com/ollama/ollama/runner"
  38. "github.com/ollama/ollama/server"
  39. "github.com/ollama/ollama/types/model"
  40. "github.com/ollama/ollama/version"
  41. )
  42. var errModelfileNotFound = errors.New("specified Modelfile wasn't found")
  43. func getModelfileName(cmd *cobra.Command) (string, error) {
  44. filename, _ := cmd.Flags().GetString("file")
  45. if filename == "" {
  46. filename = "Modelfile"
  47. }
  48. absName, err := filepath.Abs(filename)
  49. if err != nil {
  50. return "", err
  51. }
  52. _, err = os.Stat(absName)
  53. if err != nil {
  54. return "", err
  55. }
  56. return absName, nil
  57. }
  58. func CreateHandler(cmd *cobra.Command, args []string) error {
  59. p := progress.NewProgress(os.Stderr)
  60. defer p.Stop()
  61. var reader io.Reader
  62. filename, err := getModelfileName(cmd)
  63. if os.IsNotExist(err) {
  64. if filename == "" {
  65. reader = strings.NewReader("FROM .\n")
  66. } else {
  67. return errModelfileNotFound
  68. }
  69. } else if err != nil {
  70. return err
  71. } else {
  72. f, err := os.Open(filename)
  73. if err != nil {
  74. return err
  75. }
  76. reader = f
  77. defer f.Close()
  78. }
  79. modelfile, err := parser.ParseFile(reader)
  80. if err != nil {
  81. return err
  82. }
  83. status := "gathering model components"
  84. spinner := progress.NewSpinner(status)
  85. p.Add(status, spinner)
  86. req, err := modelfile.CreateRequest(filepath.Dir(filename))
  87. if err != nil {
  88. return err
  89. }
  90. spinner.Stop()
  91. req.Name = args[0]
  92. quantize, _ := cmd.Flags().GetString("quantize")
  93. if quantize != "" {
  94. req.Quantize = quantize
  95. }
  96. client, err := api.ClientFromEnvironment()
  97. if err != nil {
  98. return err
  99. }
  100. if len(req.Files) > 0 {
  101. fileMap := map[string]string{}
  102. for f, digest := range req.Files {
  103. if _, err := createBlob(cmd, client, f, digest, p); err != nil {
  104. return err
  105. }
  106. fileMap[filepath.Base(f)] = digest
  107. }
  108. req.Files = fileMap
  109. }
  110. if len(req.Adapters) > 0 {
  111. fileMap := map[string]string{}
  112. for f, digest := range req.Adapters {
  113. if _, err := createBlob(cmd, client, f, digest, p); err != nil {
  114. return err
  115. }
  116. fileMap[filepath.Base(f)] = digest
  117. }
  118. req.Adapters = fileMap
  119. }
  120. bars := make(map[string]*progress.Bar)
  121. fn := func(resp api.ProgressResponse) error {
  122. if resp.Digest != "" {
  123. bar, ok := bars[resp.Digest]
  124. if !ok {
  125. bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
  126. bars[resp.Digest] = bar
  127. p.Add(resp.Digest, bar)
  128. }
  129. bar.Set(resp.Completed)
  130. } else if status != resp.Status {
  131. spinner.Stop()
  132. status = resp.Status
  133. spinner = progress.NewSpinner(status)
  134. p.Add(status, spinner)
  135. }
  136. return nil
  137. }
  138. if err := client.Create(cmd.Context(), req, fn); err != nil {
  139. if strings.Contains(err.Error(), "path or Modelfile are required") {
  140. return fmt.Errorf("the ollama server must be updated to use `ollama create` with this client")
  141. }
  142. return err
  143. }
  144. return nil
  145. }
  146. func createBlob(cmd *cobra.Command, client *api.Client, path string, digest string, p *progress.Progress) (string, error) {
  147. realPath, err := filepath.EvalSymlinks(path)
  148. if err != nil {
  149. return "", err
  150. }
  151. bin, err := os.Open(realPath)
  152. if err != nil {
  153. return "", err
  154. }
  155. defer bin.Close()
  156. // Get file info to retrieve the size
  157. fileInfo, err := bin.Stat()
  158. if err != nil {
  159. return "", err
  160. }
  161. fileSize := fileInfo.Size()
  162. var pw progressWriter
  163. status := fmt.Sprintf("copying file %s 0%%", digest)
  164. spinner := progress.NewSpinner(status)
  165. p.Add(status, spinner)
  166. defer spinner.Stop()
  167. done := make(chan struct{})
  168. defer close(done)
  169. go func() {
  170. ticker := time.NewTicker(60 * time.Millisecond)
  171. defer ticker.Stop()
  172. for {
  173. select {
  174. case <-ticker.C:
  175. spinner.SetMessage(fmt.Sprintf("copying file %s %d%%", digest, int(100*pw.n.Load()/fileSize)))
  176. case <-done:
  177. spinner.SetMessage(fmt.Sprintf("copying file %s 100%%", digest))
  178. return
  179. }
  180. }
  181. }()
  182. if err = client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
  183. return "", err
  184. }
  185. return digest, nil
  186. }
  187. type progressWriter struct {
  188. n atomic.Int64
  189. }
  190. func (w *progressWriter) Write(p []byte) (n int, err error) {
  191. w.n.Add(int64(len(p)))
  192. return len(p), nil
  193. }
  194. func loadOrUnloadModel(cmd *cobra.Command, opts *runOptions) error {
  195. p := progress.NewProgress(os.Stderr)
  196. defer p.StopAndClear()
  197. spinner := progress.NewSpinner("")
  198. p.Add("", spinner)
  199. client, err := api.ClientFromEnvironment()
  200. if err != nil {
  201. return err
  202. }
  203. req := &api.GenerateRequest{
  204. Model: opts.Model,
  205. KeepAlive: opts.KeepAlive,
  206. }
  207. return client.Generate(cmd.Context(), req, func(api.GenerateResponse) error { return nil })
  208. }
  209. func StopHandler(cmd *cobra.Command, args []string) error {
  210. opts := &runOptions{
  211. Model: args[0],
  212. KeepAlive: &api.Duration{Duration: 0},
  213. }
  214. if err := loadOrUnloadModel(cmd, opts); err != nil {
  215. if strings.Contains(err.Error(), "not found") {
  216. return fmt.Errorf("couldn't find model \"%s\" to stop", args[0])
  217. }
  218. return err
  219. }
  220. return nil
  221. }
  222. func RunHandler(cmd *cobra.Command, args []string) error {
  223. interactive := true
  224. opts := runOptions{
  225. Model: args[0],
  226. WordWrap: os.Getenv("TERM") == "xterm-256color",
  227. Options: map[string]interface{}{},
  228. }
  229. format, err := cmd.Flags().GetString("format")
  230. if err != nil {
  231. return err
  232. }
  233. opts.Format = format
  234. keepAlive, err := cmd.Flags().GetString("keepalive")
  235. if err != nil {
  236. return err
  237. }
  238. if keepAlive != "" {
  239. d, err := time.ParseDuration(keepAlive)
  240. if err != nil {
  241. return err
  242. }
  243. opts.KeepAlive = &api.Duration{Duration: d}
  244. }
  245. prompts := args[1:]
  246. // prepend stdin to the prompt if provided
  247. if !term.IsTerminal(int(os.Stdin.Fd())) {
  248. in, err := io.ReadAll(os.Stdin)
  249. if err != nil {
  250. return err
  251. }
  252. prompts = append([]string{string(in)}, prompts...)
  253. opts.WordWrap = false
  254. interactive = false
  255. }
  256. opts.Prompt = strings.Join(prompts, " ")
  257. if len(prompts) > 0 {
  258. interactive = false
  259. }
  260. // Be quiet if we're redirecting to a pipe or file
  261. if !term.IsTerminal(int(os.Stdout.Fd())) {
  262. interactive = false
  263. }
  264. nowrap, err := cmd.Flags().GetBool("nowordwrap")
  265. if err != nil {
  266. return err
  267. }
  268. opts.WordWrap = !nowrap
  269. // Fill out the rest of the options based on information about the
  270. // model.
  271. client, err := api.ClientFromEnvironment()
  272. if err != nil {
  273. return err
  274. }
  275. name := args[0]
  276. info, err := func() (*api.ShowResponse, error) {
  277. showReq := &api.ShowRequest{Name: name}
  278. info, err := client.Show(cmd.Context(), showReq)
  279. var se api.StatusError
  280. if errors.As(err, &se) && se.StatusCode == http.StatusNotFound {
  281. if err := PullHandler(cmd, []string{name}); err != nil {
  282. return nil, err
  283. }
  284. return client.Show(cmd.Context(), &api.ShowRequest{Name: name})
  285. }
  286. return info, err
  287. }()
  288. if err != nil {
  289. return err
  290. }
  291. if len(info.ProjectorInfo) != 0 {
  292. opts.MultiModal = true
  293. }
  294. for k := range info.ModelInfo {
  295. if strings.Contains(k, ".vision.") {
  296. opts.MultiModal = true
  297. break
  298. }
  299. }
  300. opts.ParentModel = info.Details.ParentModel
  301. if interactive {
  302. if err := loadOrUnloadModel(cmd, &opts); err != nil {
  303. return err
  304. }
  305. for _, msg := range info.Messages {
  306. switch msg.Role {
  307. case "user":
  308. fmt.Printf(">>> %s\n", msg.Content)
  309. case "assistant":
  310. state := &displayResponseState{}
  311. displayResponse(msg.Content, opts.WordWrap, state)
  312. fmt.Println()
  313. fmt.Println()
  314. }
  315. }
  316. return generateInteractive(cmd, opts)
  317. }
  318. return generate(cmd, opts)
  319. }
  320. func PushHandler(cmd *cobra.Command, args []string) error {
  321. client, err := api.ClientFromEnvironment()
  322. if err != nil {
  323. return err
  324. }
  325. insecure, err := cmd.Flags().GetBool("insecure")
  326. if err != nil {
  327. return err
  328. }
  329. p := progress.NewProgress(os.Stderr)
  330. defer p.Stop()
  331. bars := make(map[string]*progress.Bar)
  332. var status string
  333. var spinner *progress.Spinner
  334. fn := func(resp api.ProgressResponse) error {
  335. if resp.Digest != "" {
  336. if spinner != nil {
  337. spinner.Stop()
  338. }
  339. bar, ok := bars[resp.Digest]
  340. if !ok {
  341. bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
  342. bars[resp.Digest] = bar
  343. p.Add(resp.Digest, bar)
  344. }
  345. bar.Set(resp.Completed)
  346. } else if status != resp.Status {
  347. if spinner != nil {
  348. spinner.Stop()
  349. }
  350. status = resp.Status
  351. spinner = progress.NewSpinner(status)
  352. p.Add(status, spinner)
  353. }
  354. return nil
  355. }
  356. request := api.PushRequest{Name: args[0], Insecure: insecure}
  357. n := model.ParseName(args[0])
  358. if err := client.Push(cmd.Context(), &request, fn); err != nil {
  359. if spinner != nil {
  360. spinner.Stop()
  361. }
  362. if strings.Contains(err.Error(), "access denied") {
  363. return errors.New("you are not authorized to push to this namespace, create the model under a namespace you own")
  364. }
  365. return err
  366. }
  367. p.Stop()
  368. spinner.Stop()
  369. destination := n.String()
  370. if strings.HasSuffix(n.Host, ".ollama.ai") || strings.HasSuffix(n.Host, ".ollama.com") {
  371. destination = "https://ollama.com/" + strings.TrimSuffix(n.DisplayShortest(), ":latest")
  372. }
  373. fmt.Printf("\nYou can find your model at:\n\n")
  374. fmt.Printf("\t%s\n", destination)
  375. return nil
  376. }
  377. func ListHandler(cmd *cobra.Command, args []string) error {
  378. client, err := api.ClientFromEnvironment()
  379. if err != nil {
  380. return err
  381. }
  382. models, err := client.List(cmd.Context())
  383. if err != nil {
  384. return err
  385. }
  386. var data [][]string
  387. for _, m := range models.Models {
  388. if len(args) == 0 || strings.HasPrefix(strings.ToLower(m.Name), strings.ToLower(args[0])) {
  389. data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), format.HumanTime(m.ModifiedAt, "Never")})
  390. }
  391. }
  392. table := tablewriter.NewWriter(os.Stdout)
  393. table.SetHeader([]string{"NAME", "ID", "SIZE", "MODIFIED"})
  394. table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
  395. table.SetAlignment(tablewriter.ALIGN_LEFT)
  396. table.SetHeaderLine(false)
  397. table.SetBorder(false)
  398. table.SetNoWhiteSpace(true)
  399. table.SetTablePadding(" ")
  400. table.AppendBulk(data)
  401. table.Render()
  402. return nil
  403. }
  404. func ListRunningHandler(cmd *cobra.Command, args []string) error {
  405. client, err := api.ClientFromEnvironment()
  406. if err != nil {
  407. return err
  408. }
  409. models, err := client.ListRunning(cmd.Context())
  410. if err != nil {
  411. return err
  412. }
  413. var data [][]string
  414. for _, m := range models.Models {
  415. if len(args) == 0 || strings.HasPrefix(m.Name, args[0]) {
  416. var procStr string
  417. switch {
  418. case m.SizeVRAM == 0:
  419. procStr = "100% CPU"
  420. case m.SizeVRAM == m.Size:
  421. procStr = "100% GPU"
  422. case m.SizeVRAM > m.Size || m.Size == 0:
  423. procStr = "Unknown"
  424. default:
  425. sizeCPU := m.Size - m.SizeVRAM
  426. cpuPercent := math.Round(float64(sizeCPU) / float64(m.Size) * 100)
  427. procStr = fmt.Sprintf("%d%%/%d%% CPU/GPU", int(cpuPercent), int(100-cpuPercent))
  428. }
  429. var until string
  430. delta := time.Since(m.ExpiresAt)
  431. if delta > 0 {
  432. until = "Stopping..."
  433. } else {
  434. until = format.HumanTime(m.ExpiresAt, "Never")
  435. }
  436. data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), procStr, until})
  437. }
  438. }
  439. table := tablewriter.NewWriter(os.Stdout)
  440. table.SetHeader([]string{"NAME", "ID", "SIZE", "PROCESSOR", "UNTIL"})
  441. table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
  442. table.SetAlignment(tablewriter.ALIGN_LEFT)
  443. table.SetHeaderLine(false)
  444. table.SetBorder(false)
  445. table.SetNoWhiteSpace(true)
  446. table.SetTablePadding(" ")
  447. table.AppendBulk(data)
  448. table.Render()
  449. return nil
  450. }
  451. func DeleteHandler(cmd *cobra.Command, args []string) error {
  452. client, err := api.ClientFromEnvironment()
  453. if err != nil {
  454. return err
  455. }
  456. // Unload the model if it's running before deletion
  457. opts := &runOptions{
  458. Model: args[0],
  459. KeepAlive: &api.Duration{Duration: 0},
  460. }
  461. if err := loadOrUnloadModel(cmd, opts); err != nil {
  462. if !strings.Contains(err.Error(), "not found") {
  463. return fmt.Errorf("unable to stop existing running model \"%s\": %s", args[0], err)
  464. }
  465. }
  466. for _, name := range args {
  467. req := api.DeleteRequest{Name: name}
  468. if err := client.Delete(cmd.Context(), &req); err != nil {
  469. return err
  470. }
  471. fmt.Printf("deleted '%s'\n", name)
  472. }
  473. return nil
  474. }
  475. func ShowHandler(cmd *cobra.Command, args []string) error {
  476. client, err := api.ClientFromEnvironment()
  477. if err != nil {
  478. return err
  479. }
  480. license, errLicense := cmd.Flags().GetBool("license")
  481. modelfile, errModelfile := cmd.Flags().GetBool("modelfile")
  482. parameters, errParams := cmd.Flags().GetBool("parameters")
  483. system, errSystem := cmd.Flags().GetBool("system")
  484. template, errTemplate := cmd.Flags().GetBool("template")
  485. for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate} {
  486. if boolErr != nil {
  487. return errors.New("error retrieving flags")
  488. }
  489. }
  490. flagsSet := 0
  491. showType := ""
  492. if license {
  493. flagsSet++
  494. showType = "license"
  495. }
  496. if modelfile {
  497. flagsSet++
  498. showType = "modelfile"
  499. }
  500. if parameters {
  501. flagsSet++
  502. showType = "parameters"
  503. }
  504. if system {
  505. flagsSet++
  506. showType = "system"
  507. }
  508. if template {
  509. flagsSet++
  510. showType = "template"
  511. }
  512. if flagsSet > 1 {
  513. return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
  514. }
  515. req := api.ShowRequest{Name: args[0]}
  516. resp, err := client.Show(cmd.Context(), &req)
  517. if err != nil {
  518. return err
  519. }
  520. if flagsSet == 1 {
  521. switch showType {
  522. case "license":
  523. fmt.Println(resp.License)
  524. case "modelfile":
  525. fmt.Println(resp.Modelfile)
  526. case "parameters":
  527. fmt.Println(resp.Parameters)
  528. case "system":
  529. fmt.Print(resp.System)
  530. case "template":
  531. fmt.Print(resp.Template)
  532. }
  533. return nil
  534. }
  535. return showInfo(resp, os.Stdout)
  536. }
  537. func showInfo(resp *api.ShowResponse, w io.Writer) error {
  538. tableRender := func(header string, rows func() [][]string) {
  539. fmt.Fprintln(w, " ", header)
  540. table := tablewriter.NewWriter(w)
  541. table.SetAlignment(tablewriter.ALIGN_LEFT)
  542. table.SetBorder(false)
  543. table.SetNoWhiteSpace(true)
  544. table.SetTablePadding(" ")
  545. switch header {
  546. case "Template", "System", "License":
  547. table.SetColWidth(100)
  548. }
  549. table.AppendBulk(rows())
  550. table.Render()
  551. fmt.Fprintln(w)
  552. }
  553. tableRender("Model", func() (rows [][]string) {
  554. if resp.ModelInfo != nil {
  555. arch := resp.ModelInfo["general.architecture"].(string)
  556. rows = append(rows, []string{"", "architecture", arch})
  557. rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ModelInfo["general.parameter_count"].(float64)))})
  558. rows = append(rows, []string{"", "context length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)].(float64), 'f', -1, 64)})
  559. rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.embedding_length", arch)].(float64), 'f', -1, 64)})
  560. } else {
  561. rows = append(rows, []string{"", "architecture", resp.Details.Family})
  562. rows = append(rows, []string{"", "parameters", resp.Details.ParameterSize})
  563. }
  564. rows = append(rows, []string{"", "quantization", resp.Details.QuantizationLevel})
  565. return
  566. })
  567. if resp.ProjectorInfo != nil {
  568. tableRender("Projector", func() (rows [][]string) {
  569. arch := resp.ProjectorInfo["general.architecture"].(string)
  570. rows = append(rows, []string{"", "architecture", arch})
  571. rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ProjectorInfo["general.parameter_count"].(float64)))})
  572. rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.embedding_length", arch)].(float64), 'f', -1, 64)})
  573. rows = append(rows, []string{"", "dimensions", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.projection_dim", arch)].(float64), 'f', -1, 64)})
  574. return
  575. })
  576. }
  577. if resp.Parameters != "" {
  578. tableRender("Parameters", func() (rows [][]string) {
  579. scanner := bufio.NewScanner(strings.NewReader(resp.Parameters))
  580. for scanner.Scan() {
  581. if text := scanner.Text(); text != "" {
  582. rows = append(rows, append([]string{""}, strings.Fields(text)...))
  583. }
  584. }
  585. return
  586. })
  587. }
  588. head := func(s string, n int) (rows [][]string) {
  589. scanner := bufio.NewScanner(strings.NewReader(s))
  590. for scanner.Scan() && (len(rows) < n || n < 0) {
  591. if text := scanner.Text(); text != "" {
  592. rows = append(rows, []string{"", strings.TrimSpace(text)})
  593. }
  594. }
  595. return
  596. }
  597. if resp.System != "" {
  598. tableRender("System", func() [][]string {
  599. return head(resp.System, 2)
  600. })
  601. }
  602. if resp.License != "" {
  603. tableRender("License", func() [][]string {
  604. return head(resp.License, 2)
  605. })
  606. }
  607. return nil
  608. }
  609. func CopyHandler(cmd *cobra.Command, args []string) error {
  610. client, err := api.ClientFromEnvironment()
  611. if err != nil {
  612. return err
  613. }
  614. req := api.CopyRequest{Source: args[0], Destination: args[1]}
  615. if err := client.Copy(cmd.Context(), &req); err != nil {
  616. return err
  617. }
  618. fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
  619. return nil
  620. }
  621. func PullHandler(cmd *cobra.Command, args []string) error {
  622. insecure, err := cmd.Flags().GetBool("insecure")
  623. if err != nil {
  624. return err
  625. }
  626. client, err := api.ClientFromEnvironment()
  627. if err != nil {
  628. return err
  629. }
  630. p := progress.NewProgress(os.Stderr)
  631. defer p.Stop()
  632. bars := make(map[string]*progress.Bar)
  633. var status string
  634. var spinner *progress.Spinner
  635. fn := func(resp api.ProgressResponse) error {
  636. if resp.Digest != "" {
  637. if spinner != nil {
  638. spinner.Stop()
  639. }
  640. bar, ok := bars[resp.Digest]
  641. if !ok {
  642. bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
  643. bars[resp.Digest] = bar
  644. p.Add(resp.Digest, bar)
  645. }
  646. bar.Set(resp.Completed)
  647. } else if status != resp.Status {
  648. if spinner != nil {
  649. spinner.Stop()
  650. }
  651. status = resp.Status
  652. spinner = progress.NewSpinner(status)
  653. p.Add(status, spinner)
  654. }
  655. return nil
  656. }
  657. request := api.PullRequest{Name: args[0], Insecure: insecure}
  658. if err := client.Pull(cmd.Context(), &request, fn); err != nil {
  659. return err
  660. }
  661. return nil
  662. }
  663. type generateContextKey string
  664. type runOptions struct {
  665. Model string
  666. ParentModel string
  667. Prompt string
  668. Messages []api.Message
  669. WordWrap bool
  670. Format string
  671. System string
  672. Images []api.ImageData
  673. Options map[string]interface{}
  674. MultiModal bool
  675. KeepAlive *api.Duration
  676. }
  677. type displayResponseState struct {
  678. lineLength int
  679. wordBuffer string
  680. }
  681. func displayResponse(content string, wordWrap bool, state *displayResponseState) {
  682. termWidth, _, _ := term.GetSize(int(os.Stdout.Fd()))
  683. if wordWrap && termWidth >= 10 {
  684. for _, ch := range content {
  685. if state.lineLength+1 > termWidth-5 {
  686. if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
  687. fmt.Printf("%s%c", state.wordBuffer, ch)
  688. state.wordBuffer = ""
  689. state.lineLength = 0
  690. continue
  691. }
  692. // backtrack the length of the last word and clear to the end of the line
  693. a := runewidth.StringWidth(state.wordBuffer)
  694. if a > 0 {
  695. fmt.Printf("\x1b[%dD", a)
  696. }
  697. fmt.Printf("\x1b[K\n")
  698. fmt.Printf("%s%c", state.wordBuffer, ch)
  699. chWidth := runewidth.RuneWidth(ch)
  700. state.lineLength = runewidth.StringWidth(state.wordBuffer) + chWidth
  701. } else {
  702. fmt.Print(string(ch))
  703. state.lineLength += runewidth.RuneWidth(ch)
  704. if runewidth.RuneWidth(ch) >= 2 {
  705. state.wordBuffer = ""
  706. continue
  707. }
  708. switch ch {
  709. case ' ':
  710. state.wordBuffer = ""
  711. case '\n':
  712. state.lineLength = 0
  713. default:
  714. state.wordBuffer += string(ch)
  715. }
  716. }
  717. }
  718. } else {
  719. fmt.Printf("%s%s", state.wordBuffer, content)
  720. if len(state.wordBuffer) > 0 {
  721. state.wordBuffer = ""
  722. }
  723. }
  724. }
  725. func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
  726. client, err := api.ClientFromEnvironment()
  727. if err != nil {
  728. return nil, err
  729. }
  730. p := progress.NewProgress(os.Stderr)
  731. defer p.StopAndClear()
  732. spinner := progress.NewSpinner("")
  733. p.Add("", spinner)
  734. cancelCtx, cancel := context.WithCancel(cmd.Context())
  735. defer cancel()
  736. sigChan := make(chan os.Signal, 1)
  737. signal.Notify(sigChan, syscall.SIGINT)
  738. go func() {
  739. <-sigChan
  740. cancel()
  741. }()
  742. var state *displayResponseState = &displayResponseState{}
  743. var latest api.ChatResponse
  744. var fullResponse strings.Builder
  745. var role string
  746. fn := func(response api.ChatResponse) error {
  747. p.StopAndClear()
  748. latest = response
  749. role = response.Message.Role
  750. content := response.Message.Content
  751. fullResponse.WriteString(content)
  752. displayResponse(content, opts.WordWrap, state)
  753. return nil
  754. }
  755. if opts.Format == "json" {
  756. opts.Format = `"` + opts.Format + `"`
  757. }
  758. req := &api.ChatRequest{
  759. Model: opts.Model,
  760. Messages: opts.Messages,
  761. Format: json.RawMessage(opts.Format),
  762. Options: opts.Options,
  763. }
  764. if opts.KeepAlive != nil {
  765. req.KeepAlive = opts.KeepAlive
  766. }
  767. if err := client.Chat(cancelCtx, req, fn); err != nil {
  768. if errors.Is(err, context.Canceled) {
  769. return nil, nil
  770. }
  771. return nil, err
  772. }
  773. if len(opts.Messages) > 0 {
  774. fmt.Println()
  775. fmt.Println()
  776. }
  777. verbose, err := cmd.Flags().GetBool("verbose")
  778. if err != nil {
  779. return nil, err
  780. }
  781. if verbose {
  782. latest.Summary()
  783. }
  784. return &api.Message{Role: role, Content: fullResponse.String()}, nil
  785. }
  786. func generate(cmd *cobra.Command, opts runOptions) error {
  787. client, err := api.ClientFromEnvironment()
  788. if err != nil {
  789. return err
  790. }
  791. p := progress.NewProgress(os.Stderr)
  792. defer p.StopAndClear()
  793. spinner := progress.NewSpinner("")
  794. p.Add("", spinner)
  795. var latest api.GenerateResponse
  796. generateContext, ok := cmd.Context().Value(generateContextKey("context")).([]int)
  797. if !ok {
  798. generateContext = []int{}
  799. }
  800. ctx, cancel := context.WithCancel(cmd.Context())
  801. defer cancel()
  802. sigChan := make(chan os.Signal, 1)
  803. signal.Notify(sigChan, syscall.SIGINT)
  804. go func() {
  805. <-sigChan
  806. cancel()
  807. }()
  808. var state *displayResponseState = &displayResponseState{}
  809. fn := func(response api.GenerateResponse) error {
  810. p.StopAndClear()
  811. latest = response
  812. content := response.Response
  813. displayResponse(content, opts.WordWrap, state)
  814. return nil
  815. }
  816. if opts.MultiModal {
  817. opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
  818. if err != nil {
  819. return err
  820. }
  821. }
  822. if opts.Format == "json" {
  823. opts.Format = `"` + opts.Format + `"`
  824. }
  825. request := api.GenerateRequest{
  826. Model: opts.Model,
  827. Prompt: opts.Prompt,
  828. Context: generateContext,
  829. Images: opts.Images,
  830. Format: json.RawMessage(opts.Format),
  831. System: opts.System,
  832. Options: opts.Options,
  833. KeepAlive: opts.KeepAlive,
  834. }
  835. if err := client.Generate(ctx, &request, fn); err != nil {
  836. if errors.Is(err, context.Canceled) {
  837. return nil
  838. }
  839. return err
  840. }
  841. if opts.Prompt != "" {
  842. fmt.Println()
  843. fmt.Println()
  844. }
  845. if !latest.Done {
  846. return nil
  847. }
  848. verbose, err := cmd.Flags().GetBool("verbose")
  849. if err != nil {
  850. return err
  851. }
  852. if verbose {
  853. latest.Summary()
  854. }
  855. ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
  856. cmd.SetContext(ctx)
  857. return nil
  858. }
  859. func RunServer(_ *cobra.Command, _ []string) error {
  860. if err := initializeKeypair(); err != nil {
  861. return err
  862. }
  863. ln, err := net.Listen("tcp", envconfig.Host().Host)
  864. if err != nil {
  865. return err
  866. }
  867. err = server.Serve(ln)
  868. if errors.Is(err, http.ErrServerClosed) {
  869. return nil
  870. }
  871. return err
  872. }
  873. func initializeKeypair() error {
  874. home, err := os.UserHomeDir()
  875. if err != nil {
  876. return err
  877. }
  878. privKeyPath := filepath.Join(home, ".ollama", "id_ed25519")
  879. pubKeyPath := filepath.Join(home, ".ollama", "id_ed25519.pub")
  880. _, err = os.Stat(privKeyPath)
  881. if os.IsNotExist(err) {
  882. fmt.Printf("Couldn't find '%s'. Generating new private key.\n", privKeyPath)
  883. cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
  884. if err != nil {
  885. return err
  886. }
  887. privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
  888. if err != nil {
  889. return err
  890. }
  891. if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
  892. return fmt.Errorf("could not create directory %w", err)
  893. }
  894. if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
  895. return err
  896. }
  897. sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
  898. if err != nil {
  899. return err
  900. }
  901. publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
  902. if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
  903. return err
  904. }
  905. fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
  906. }
  907. return nil
  908. }
  909. func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
  910. client, err := api.ClientFromEnvironment()
  911. if err != nil {
  912. return err
  913. }
  914. if err := client.Heartbeat(cmd.Context()); err != nil {
  915. if !strings.Contains(err.Error(), " refused") {
  916. return err
  917. }
  918. if err := startApp(cmd.Context(), client); err != nil {
  919. return errors.New("could not connect to ollama app, is it running?")
  920. }
  921. }
  922. return nil
  923. }
  924. func versionHandler(cmd *cobra.Command, _ []string) {
  925. client, err := api.ClientFromEnvironment()
  926. if err != nil {
  927. return
  928. }
  929. serverVersion, err := client.Version(cmd.Context())
  930. if err != nil {
  931. fmt.Println("Warning: could not connect to a running Ollama instance")
  932. }
  933. if serverVersion != "" {
  934. fmt.Printf("ollama version is %s\n", serverVersion)
  935. }
  936. if serverVersion != version.Version {
  937. fmt.Printf("Warning: client version is %s\n", version.Version)
  938. }
  939. }
  940. func appendEnvDocs(cmd *cobra.Command, envs []envconfig.EnvVar) {
  941. if len(envs) == 0 {
  942. return
  943. }
  944. envUsage := `
  945. Environment Variables:
  946. `
  947. for _, e := range envs {
  948. envUsage += fmt.Sprintf(" %-24s %s\n", e.Name, e.Description)
  949. }
  950. cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
  951. }
  952. func NewCLI() *cobra.Command {
  953. log.SetFlags(log.LstdFlags | log.Lshortfile)
  954. cobra.EnableCommandSorting = false
  955. if runtime.GOOS == "windows" && term.IsTerminal(int(os.Stdout.Fd())) {
  956. console.ConsoleFromFile(os.Stdin) //nolint:errcheck
  957. }
  958. rootCmd := &cobra.Command{
  959. Use: "ollama",
  960. Short: "Large language model runner",
  961. SilenceUsage: true,
  962. SilenceErrors: true,
  963. CompletionOptions: cobra.CompletionOptions{
  964. DisableDefaultCmd: true,
  965. },
  966. Run: func(cmd *cobra.Command, args []string) {
  967. if version, _ := cmd.Flags().GetBool("version"); version {
  968. versionHandler(cmd, args)
  969. return
  970. }
  971. cmd.Print(cmd.UsageString())
  972. },
  973. }
  974. rootCmd.Flags().BoolP("version", "v", false, "Show version information")
  975. createCmd := &cobra.Command{
  976. Use: "create MODEL",
  977. Short: "Create a model from a Modelfile",
  978. Args: cobra.ExactArgs(1),
  979. PreRunE: checkServerHeartbeat,
  980. RunE: CreateHandler,
  981. }
  982. createCmd.Flags().StringP("file", "f", "", "Name of the Modelfile (default \"Modelfile\"")
  983. createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_0)")
  984. showCmd := &cobra.Command{
  985. Use: "show MODEL",
  986. Short: "Show information for a model",
  987. Args: cobra.ExactArgs(1),
  988. PreRunE: checkServerHeartbeat,
  989. RunE: ShowHandler,
  990. }
  991. showCmd.Flags().Bool("license", false, "Show license of a model")
  992. showCmd.Flags().Bool("modelfile", false, "Show Modelfile of a model")
  993. showCmd.Flags().Bool("parameters", false, "Show parameters of a model")
  994. showCmd.Flags().Bool("template", false, "Show template of a model")
  995. showCmd.Flags().Bool("system", false, "Show system message of a model")
  996. runCmd := &cobra.Command{
  997. Use: "run MODEL [PROMPT]",
  998. Short: "Run a model",
  999. Args: cobra.MinimumNArgs(1),
  1000. PreRunE: checkServerHeartbeat,
  1001. RunE: RunHandler,
  1002. }
  1003. runCmd.Flags().String("keepalive", "", "Duration to keep a model loaded (e.g. 5m)")
  1004. runCmd.Flags().Bool("verbose", false, "Show timings for response")
  1005. runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
  1006. runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
  1007. runCmd.Flags().String("format", "", "Response format (e.g. json)")
  1008. stopCmd := &cobra.Command{
  1009. Use: "stop MODEL",
  1010. Short: "Stop a running model",
  1011. Args: cobra.ExactArgs(1),
  1012. PreRunE: checkServerHeartbeat,
  1013. RunE: StopHandler,
  1014. }
  1015. serveCmd := &cobra.Command{
  1016. Use: "serve",
  1017. Aliases: []string{"start"},
  1018. Short: "Start ollama",
  1019. Args: cobra.ExactArgs(0),
  1020. RunE: RunServer,
  1021. }
  1022. pullCmd := &cobra.Command{
  1023. Use: "pull MODEL",
  1024. Short: "Pull a model from a registry",
  1025. Args: cobra.ExactArgs(1),
  1026. PreRunE: checkServerHeartbeat,
  1027. RunE: PullHandler,
  1028. }
  1029. pullCmd.Flags().Bool("insecure", false, "Use an insecure registry")
  1030. pushCmd := &cobra.Command{
  1031. Use: "push MODEL",
  1032. Short: "Push a model to a registry",
  1033. Args: cobra.ExactArgs(1),
  1034. PreRunE: checkServerHeartbeat,
  1035. RunE: PushHandler,
  1036. }
  1037. pushCmd.Flags().Bool("insecure", false, "Use an insecure registry")
  1038. listCmd := &cobra.Command{
  1039. Use: "list",
  1040. Aliases: []string{"ls"},
  1041. Short: "List models",
  1042. PreRunE: checkServerHeartbeat,
  1043. RunE: ListHandler,
  1044. }
  1045. psCmd := &cobra.Command{
  1046. Use: "ps",
  1047. Short: "List running models",
  1048. PreRunE: checkServerHeartbeat,
  1049. RunE: ListRunningHandler,
  1050. }
  1051. copyCmd := &cobra.Command{
  1052. Use: "cp SOURCE DESTINATION",
  1053. Short: "Copy a model",
  1054. Args: cobra.ExactArgs(2),
  1055. PreRunE: checkServerHeartbeat,
  1056. RunE: CopyHandler,
  1057. }
  1058. deleteCmd := &cobra.Command{
  1059. Use: "rm MODEL [MODEL...]",
  1060. Short: "Remove a model",
  1061. Args: cobra.MinimumNArgs(1),
  1062. PreRunE: checkServerHeartbeat,
  1063. RunE: DeleteHandler,
  1064. }
  1065. runnerCmd := &cobra.Command{
  1066. Use: "runner",
  1067. Short: llama.PrintSystemInfo(),
  1068. Hidden: true,
  1069. RunE: func(cmd *cobra.Command, args []string) error {
  1070. return runner.Execute(os.Args[1:])
  1071. },
  1072. FParseErrWhitelist: cobra.FParseErrWhitelist{UnknownFlags: true},
  1073. }
  1074. runnerCmd.SetHelpFunc(func(cmd *cobra.Command, args []string) {
  1075. _ = runner.Execute(args[1:])
  1076. })
  1077. envVars := envconfig.AsMap()
  1078. envs := []envconfig.EnvVar{envVars["OLLAMA_HOST"]}
  1079. for _, cmd := range []*cobra.Command{
  1080. createCmd,
  1081. showCmd,
  1082. runCmd,
  1083. stopCmd,
  1084. pullCmd,
  1085. pushCmd,
  1086. listCmd,
  1087. psCmd,
  1088. copyCmd,
  1089. deleteCmd,
  1090. serveCmd,
  1091. } {
  1092. switch cmd {
  1093. case runCmd:
  1094. appendEnvDocs(cmd, []envconfig.EnvVar{envVars["OLLAMA_HOST"], envVars["OLLAMA_NOHISTORY"]})
  1095. case serveCmd:
  1096. appendEnvDocs(cmd, []envconfig.EnvVar{
  1097. envVars["OLLAMA_DEBUG"],
  1098. envVars["OLLAMA_HOST"],
  1099. envVars["OLLAMA_KEEP_ALIVE"],
  1100. envVars["OLLAMA_MAX_LOADED_MODELS"],
  1101. envVars["OLLAMA_MAX_QUEUE"],
  1102. envVars["OLLAMA_MODELS"],
  1103. envVars["OLLAMA_NUM_PARALLEL"],
  1104. envVars["OLLAMA_NOPRUNE"],
  1105. envVars["OLLAMA_ORIGINS"],
  1106. envVars["OLLAMA_SCHED_SPREAD"],
  1107. envVars["OLLAMA_TMPDIR"],
  1108. envVars["OLLAMA_FLASH_ATTENTION"],
  1109. envVars["OLLAMA_KV_CACHE_TYPE"],
  1110. envVars["OLLAMA_LLM_LIBRARY"],
  1111. envVars["OLLAMA_GPU_OVERHEAD"],
  1112. envVars["OLLAMA_LOAD_TIMEOUT"],
  1113. })
  1114. default:
  1115. appendEnvDocs(cmd, envs)
  1116. }
  1117. }
  1118. rootCmd.AddCommand(
  1119. serveCmd,
  1120. createCmd,
  1121. showCmd,
  1122. runCmd,
  1123. stopCmd,
  1124. pullCmd,
  1125. pushCmd,
  1126. listCmd,
  1127. psCmd,
  1128. copyCmd,
  1129. deleteCmd,
  1130. runnerCmd,
  1131. )
  1132. return rootCmd
  1133. }