images.go 28 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149
  1. package server
  2. import (
  3. "bytes"
  4. "cmp"
  5. "context"
  6. "crypto/sha256"
  7. "encoding/hex"
  8. "encoding/json"
  9. "errors"
  10. "fmt"
  11. "io"
  12. "log"
  13. "log/slog"
  14. "net"
  15. "net/http"
  16. "net/url"
  17. "os"
  18. "path/filepath"
  19. "runtime"
  20. "slices"
  21. "strconv"
  22. "strings"
  23. "github.com/ollama/ollama/api"
  24. "github.com/ollama/ollama/envconfig"
  25. "github.com/ollama/ollama/format"
  26. "github.com/ollama/ollama/llama"
  27. "github.com/ollama/ollama/llm"
  28. "github.com/ollama/ollama/parser"
  29. "github.com/ollama/ollama/template"
  30. "github.com/ollama/ollama/types/model"
  31. "github.com/ollama/ollama/version"
  32. )
  33. var (
  34. errCapabilities = errors.New("does not support")
  35. errCapabilityCompletion = errors.New("completion")
  36. errCapabilityTools = errors.New("tools")
  37. errCapabilityInsert = errors.New("insert")
  38. )
  39. type Capability string
  40. const (
  41. CapabilityCompletion = Capability("completion")
  42. CapabilityTools = Capability("tools")
  43. CapabilityInsert = Capability("insert")
  44. )
  45. type registryOptions struct {
  46. Insecure bool
  47. Username string
  48. Password string
  49. Token string
  50. CheckRedirect func(req *http.Request, via []*http.Request) error
  51. }
  52. type Model struct {
  53. Name string `json:"name"`
  54. Config ConfigV2
  55. ShortName string
  56. ModelPath string
  57. ParentModel string
  58. AdapterPaths []string
  59. ProjectorPaths []string
  60. System string
  61. License []string
  62. Digest string
  63. Options map[string]interface{}
  64. Messages []api.Message
  65. Template *template.Template
  66. }
  67. // CheckCapabilities checks if the model has the specified capabilities returning an error describing
  68. // any missing or unknown capabilities
  69. func (m *Model) CheckCapabilities(caps ...Capability) error {
  70. var errs []error
  71. for _, cap := range caps {
  72. switch cap {
  73. case CapabilityCompletion:
  74. f, err := os.Open(m.ModelPath)
  75. if err != nil {
  76. slog.Error("couldn't open model file", "error", err)
  77. continue
  78. }
  79. defer f.Close()
  80. // TODO(mxyng): decode the GGML into model to avoid doing this multiple times
  81. ggml, _, err := llm.DecodeGGML(f, 0)
  82. if err != nil {
  83. slog.Error("couldn't decode ggml", "error", err)
  84. continue
  85. }
  86. if _, ok := ggml.KV()[fmt.Sprintf("%s.pooling_type", ggml.KV().Architecture())]; ok {
  87. errs = append(errs, errCapabilityCompletion)
  88. }
  89. case CapabilityTools:
  90. if !slices.Contains(m.Template.Vars(), "tools") {
  91. errs = append(errs, errCapabilityTools)
  92. }
  93. case CapabilityInsert:
  94. vars := m.Template.Vars()
  95. if !slices.Contains(vars, "suffix") {
  96. errs = append(errs, errCapabilityInsert)
  97. }
  98. default:
  99. slog.Error("unknown capability", "capability", cap)
  100. return fmt.Errorf("unknown capability: %s", cap)
  101. }
  102. }
  103. if err := errors.Join(errs...); err != nil {
  104. return fmt.Errorf("%w %w", errCapabilities, errors.Join(errs...))
  105. }
  106. return nil
  107. }
  108. func (m *Model) String() string {
  109. var modelfile parser.File
  110. modelfile.Commands = append(modelfile.Commands, parser.Command{
  111. Name: "model",
  112. Args: m.ModelPath,
  113. })
  114. for _, adapter := range m.AdapterPaths {
  115. modelfile.Commands = append(modelfile.Commands, parser.Command{
  116. Name: "adapter",
  117. Args: adapter,
  118. })
  119. }
  120. for _, projector := range m.ProjectorPaths {
  121. modelfile.Commands = append(modelfile.Commands, parser.Command{
  122. Name: "model",
  123. Args: projector,
  124. })
  125. }
  126. if m.Template != nil {
  127. modelfile.Commands = append(modelfile.Commands, parser.Command{
  128. Name: "template",
  129. Args: m.Template.String(),
  130. })
  131. }
  132. if m.System != "" {
  133. modelfile.Commands = append(modelfile.Commands, parser.Command{
  134. Name: "system",
  135. Args: m.System,
  136. })
  137. }
  138. for k, v := range m.Options {
  139. switch v := v.(type) {
  140. case []any:
  141. for _, s := range v {
  142. modelfile.Commands = append(modelfile.Commands, parser.Command{
  143. Name: k,
  144. Args: fmt.Sprintf("%v", s),
  145. })
  146. }
  147. default:
  148. modelfile.Commands = append(modelfile.Commands, parser.Command{
  149. Name: k,
  150. Args: fmt.Sprintf("%v", v),
  151. })
  152. }
  153. }
  154. for _, license := range m.License {
  155. modelfile.Commands = append(modelfile.Commands, parser.Command{
  156. Name: "license",
  157. Args: license,
  158. })
  159. }
  160. for _, msg := range m.Messages {
  161. modelfile.Commands = append(modelfile.Commands, parser.Command{
  162. Name: "message",
  163. Args: fmt.Sprintf("%s: %s", msg.Role, msg.Content),
  164. })
  165. }
  166. return modelfile.String()
  167. }
  168. type ConfigV2 struct {
  169. ModelFormat string `json:"model_format"`
  170. ModelFamily string `json:"model_family"`
  171. ModelFamilies []string `json:"model_families"`
  172. ModelType string `json:"model_type"`
  173. FileType string `json:"file_type"`
  174. // required by spec
  175. Architecture string `json:"architecture"`
  176. OS string `json:"os"`
  177. RootFS RootFS `json:"rootfs"`
  178. }
  179. type RootFS struct {
  180. Type string `json:"type"`
  181. DiffIDs []string `json:"diff_ids"`
  182. }
  183. func GetManifest(mp ModelPath) (*Manifest, string, error) {
  184. fp, err := mp.GetManifestPath()
  185. if err != nil {
  186. return nil, "", err
  187. }
  188. f, err := os.Open(fp)
  189. if err != nil {
  190. return nil, "", err
  191. }
  192. defer f.Close()
  193. sha256sum := sha256.New()
  194. var manifest Manifest
  195. if err := json.NewDecoder(io.TeeReader(f, sha256sum)).Decode(&manifest); err != nil {
  196. return nil, "", err
  197. }
  198. return &manifest, hex.EncodeToString(sha256sum.Sum(nil)), nil
  199. }
  200. func GetModel(name string) (*Model, error) {
  201. mp := ParseModelPath(name)
  202. manifest, digest, err := GetManifest(mp)
  203. if err != nil {
  204. return nil, err
  205. }
  206. model := &Model{
  207. Name: mp.GetFullTagname(),
  208. ShortName: mp.GetShortTagname(),
  209. Digest: digest,
  210. Template: template.DefaultTemplate,
  211. }
  212. if manifest.Config.Digest != "" {
  213. filename, err := GetBlobsPath(manifest.Config.Digest)
  214. if err != nil {
  215. return nil, err
  216. }
  217. configFile, err := os.Open(filename)
  218. if err != nil {
  219. return nil, err
  220. }
  221. defer configFile.Close()
  222. if err := json.NewDecoder(configFile).Decode(&model.Config); err != nil {
  223. return nil, err
  224. }
  225. }
  226. for _, layer := range manifest.Layers {
  227. filename, err := GetBlobsPath(layer.Digest)
  228. if err != nil {
  229. return nil, err
  230. }
  231. switch layer.MediaType {
  232. case "application/vnd.ollama.image.model":
  233. model.ModelPath = filename
  234. model.ParentModel = layer.From
  235. case "application/vnd.ollama.image.embed":
  236. // Deprecated in versions > 0.1.2
  237. // TODO: remove this warning in a future version
  238. slog.Info("WARNING: model contains embeddings, but embeddings in modelfiles have been deprecated and will be ignored.")
  239. case "application/vnd.ollama.image.adapter":
  240. model.AdapterPaths = append(model.AdapterPaths, filename)
  241. case "application/vnd.ollama.image.projector":
  242. model.ProjectorPaths = append(model.ProjectorPaths, filename)
  243. case "application/vnd.ollama.image.prompt",
  244. "application/vnd.ollama.image.template":
  245. bts, err := os.ReadFile(filename)
  246. if err != nil {
  247. return nil, err
  248. }
  249. model.Template, err = template.Parse(string(bts))
  250. if err != nil {
  251. return nil, err
  252. }
  253. case "application/vnd.ollama.image.system":
  254. bts, err := os.ReadFile(filename)
  255. if err != nil {
  256. return nil, err
  257. }
  258. model.System = string(bts)
  259. case "application/vnd.ollama.image.params":
  260. params, err := os.Open(filename)
  261. if err != nil {
  262. return nil, err
  263. }
  264. defer params.Close()
  265. // parse model options parameters into a map so that we can see which fields have been specified explicitly
  266. if err = json.NewDecoder(params).Decode(&model.Options); err != nil {
  267. return nil, err
  268. }
  269. case "application/vnd.ollama.image.messages":
  270. msgs, err := os.Open(filename)
  271. if err != nil {
  272. return nil, err
  273. }
  274. defer msgs.Close()
  275. if err = json.NewDecoder(msgs).Decode(&model.Messages); err != nil {
  276. return nil, err
  277. }
  278. case "application/vnd.ollama.image.license":
  279. bts, err := os.ReadFile(filename)
  280. if err != nil {
  281. return nil, err
  282. }
  283. model.License = append(model.License, string(bts))
  284. }
  285. }
  286. return model, nil
  287. }
  288. func realpath(rel, from string) string {
  289. abspath, err := filepath.Abs(from)
  290. if err != nil {
  291. return from
  292. }
  293. home, err := os.UserHomeDir()
  294. if err != nil {
  295. return abspath
  296. }
  297. if from == "~" {
  298. return home
  299. } else if strings.HasPrefix(from, "~/") {
  300. return filepath.Join(home, from[2:])
  301. }
  302. if _, err := os.Stat(filepath.Join(rel, from)); err == nil {
  303. // this is a file relative to the Modelfile
  304. return filepath.Join(rel, from)
  305. }
  306. return abspath
  307. }
  308. func CreateModel(ctx context.Context, name model.Name, modelFileDir, quantization string, modelfile *parser.File, fn func(resp api.ProgressResponse)) (err error) {
  309. config := ConfigV2{
  310. OS: "linux",
  311. Architecture: "amd64",
  312. RootFS: RootFS{
  313. Type: "layers",
  314. },
  315. }
  316. var messages []*api.Message
  317. parameters := make(map[string]any)
  318. var layers []Layer
  319. var baseLayers []*layerGGML
  320. for _, c := range modelfile.Commands {
  321. mediatype := fmt.Sprintf("application/vnd.ollama.image.%s", c.Name)
  322. command := c.Name
  323. switch command {
  324. case "model", "adapter":
  325. if name := model.ParseName(c.Args); name.IsValid() && command == "model" {
  326. baseLayers, err = parseFromModel(ctx, name, fn)
  327. if err != nil {
  328. return err
  329. }
  330. } else if strings.HasPrefix(c.Args, "@") {
  331. digest := strings.TrimPrefix(c.Args, "@")
  332. if ib, ok := intermediateBlobs[digest]; ok {
  333. p, err := GetBlobsPath(ib)
  334. if err != nil {
  335. return err
  336. }
  337. if _, err := os.Stat(p); errors.Is(err, os.ErrNotExist) {
  338. // pass
  339. } else if err != nil {
  340. return err
  341. } else {
  342. fn(api.ProgressResponse{Status: fmt.Sprintf("using cached layer %s", ib)})
  343. digest = ib
  344. }
  345. }
  346. blobpath, err := GetBlobsPath(digest)
  347. if err != nil {
  348. return err
  349. }
  350. blob, err := os.Open(blobpath)
  351. if err != nil {
  352. return err
  353. }
  354. defer blob.Close()
  355. baseLayers, err = parseFromFile(ctx, command, baseLayers, blob, digest, fn)
  356. if err != nil {
  357. return err
  358. }
  359. } else if file, err := os.Open(realpath(modelFileDir, c.Args)); err == nil {
  360. defer file.Close()
  361. baseLayers, err = parseFromFile(ctx, command, baseLayers, file, "", fn)
  362. if err != nil {
  363. return err
  364. }
  365. } else {
  366. return fmt.Errorf("invalid model reference: %s", c.Args)
  367. }
  368. for _, baseLayer := range baseLayers {
  369. if quantization != "" &&
  370. baseLayer.MediaType == "application/vnd.ollama.image.model" &&
  371. baseLayer.GGML != nil &&
  372. baseLayer.GGML.Name() == "gguf" {
  373. want, err := llm.ParseFileType(quantization)
  374. if err != nil {
  375. return err
  376. }
  377. ft := baseLayer.GGML.KV().FileType()
  378. if !slices.Contains([]string{"F16", "F32"}, ft.String()) {
  379. return errors.New("quantization is only supported for F16 and F32 models")
  380. } else if want != ft {
  381. fn(api.ProgressResponse{Status: fmt.Sprintf("quantizing %s model to %s", ft, quantization)})
  382. blob, err := GetBlobsPath(baseLayer.Digest)
  383. if err != nil {
  384. return err
  385. }
  386. temp, err := os.CreateTemp(filepath.Dir(blob), quantization)
  387. if err != nil {
  388. return err
  389. }
  390. defer temp.Close()
  391. defer os.Remove(temp.Name())
  392. if err := llama.Quantize(blob, temp.Name(), uint32(want)); err != nil {
  393. return err
  394. }
  395. layer, err := NewLayer(temp, baseLayer.MediaType)
  396. if err != nil {
  397. return err
  398. }
  399. if _, err := temp.Seek(0, io.SeekStart); err != nil {
  400. return err
  401. }
  402. ggml, _, err := llm.DecodeGGML(temp, 0)
  403. if err != nil {
  404. return err
  405. }
  406. baseLayer.Layer = layer
  407. baseLayer.GGML = ggml
  408. }
  409. }
  410. if baseLayer.GGML != nil {
  411. config.ModelFormat = cmp.Or(config.ModelFormat, baseLayer.GGML.Name())
  412. config.ModelFamily = cmp.Or(config.ModelFamily, baseLayer.GGML.KV().Architecture())
  413. config.ModelType = cmp.Or(config.ModelType, format.HumanNumber(baseLayer.GGML.KV().ParameterCount()))
  414. config.FileType = cmp.Or(config.FileType, baseLayer.GGML.KV().FileType().String())
  415. config.ModelFamilies = append(config.ModelFamilies, baseLayer.GGML.KV().Architecture())
  416. }
  417. layers = append(layers, baseLayer.Layer)
  418. }
  419. case "license", "template", "system":
  420. if c.Name == "template" {
  421. if _, err := template.Parse(c.Args); err != nil {
  422. return fmt.Errorf("%w: %s", errBadTemplate, err)
  423. }
  424. }
  425. if c.Name != "license" {
  426. // replace
  427. layers = slices.DeleteFunc(layers, func(layer Layer) bool {
  428. if layer.MediaType != mediatype {
  429. return false
  430. }
  431. if err := layer.Remove(); err != nil {
  432. return false
  433. }
  434. return true
  435. })
  436. }
  437. blob := strings.NewReader(c.Args)
  438. layer, err := NewLayer(blob, mediatype)
  439. if err != nil {
  440. return err
  441. }
  442. layers = append(layers, layer)
  443. case "message":
  444. role, content, ok := strings.Cut(c.Args, ": ")
  445. if !ok {
  446. return fmt.Errorf("invalid message: %s", c.Args)
  447. }
  448. messages = append(messages, &api.Message{Role: role, Content: content})
  449. default:
  450. ps, err := api.FormatParams(map[string][]string{c.Name: {c.Args}})
  451. if err != nil {
  452. return err
  453. }
  454. for k, v := range ps {
  455. if ks, ok := parameters[k].([]string); ok {
  456. parameters[k] = append(ks, v.([]string)...)
  457. } else if vs, ok := v.([]string); ok {
  458. parameters[k] = vs
  459. } else {
  460. parameters[k] = v
  461. }
  462. }
  463. }
  464. }
  465. var err2 error
  466. layers = slices.DeleteFunc(layers, func(layer Layer) bool {
  467. switch layer.MediaType {
  468. case "application/vnd.ollama.image.message":
  469. // if there are new messages, remove the inherited ones
  470. if len(messages) > 0 {
  471. return true
  472. }
  473. return false
  474. case "application/vnd.ollama.image.params":
  475. // merge inherited parameters with new ones
  476. r, err := layer.Open()
  477. if err != nil {
  478. err2 = err
  479. return false
  480. }
  481. defer r.Close()
  482. var ps map[string]any
  483. if err := json.NewDecoder(r).Decode(&ps); err != nil {
  484. err2 = err
  485. return false
  486. }
  487. for k, v := range ps {
  488. if _, ok := parameters[k]; !ok {
  489. parameters[k] = v
  490. }
  491. }
  492. return true
  493. default:
  494. return false
  495. }
  496. })
  497. if err2 != nil {
  498. return err2
  499. }
  500. if len(messages) > 0 {
  501. var b bytes.Buffer
  502. if err := json.NewEncoder(&b).Encode(messages); err != nil {
  503. return err
  504. }
  505. layer, err := NewLayer(&b, "application/vnd.ollama.image.messages")
  506. if err != nil {
  507. return err
  508. }
  509. layers = append(layers, layer)
  510. }
  511. if len(parameters) > 0 {
  512. var b bytes.Buffer
  513. if err := json.NewEncoder(&b).Encode(parameters); err != nil {
  514. return err
  515. }
  516. layer, err := NewLayer(&b, "application/vnd.ollama.image.params")
  517. if err != nil {
  518. return err
  519. }
  520. layers = append(layers, layer)
  521. }
  522. digests := make([]string, len(layers))
  523. for i, layer := range layers {
  524. digests[i] = layer.Digest
  525. }
  526. config.RootFS.DiffIDs = digests
  527. var b bytes.Buffer
  528. if err := json.NewEncoder(&b).Encode(config); err != nil {
  529. return err
  530. }
  531. configLayer, err := NewLayer(&b, "application/vnd.docker.container.image.v1+json")
  532. if err != nil {
  533. return err
  534. }
  535. for _, layer := range append(layers, configLayer) {
  536. if layer.status != "" {
  537. fn(api.ProgressResponse{Status: layer.status})
  538. }
  539. }
  540. old, _ := ParseNamedManifest(name)
  541. fn(api.ProgressResponse{Status: "writing manifest"})
  542. if err := WriteManifest(name, configLayer, layers); err != nil {
  543. return err
  544. }
  545. if !envconfig.NoPrune() && old != nil {
  546. if err := old.RemoveLayers(); err != nil {
  547. return err
  548. }
  549. }
  550. fn(api.ProgressResponse{Status: "success"})
  551. return nil
  552. }
  553. func CopyModel(src, dst model.Name) error {
  554. if !dst.IsFullyQualified() {
  555. return model.Unqualified(dst)
  556. }
  557. if !src.IsFullyQualified() {
  558. return model.Unqualified(src)
  559. }
  560. if src.Filepath() == dst.Filepath() {
  561. return nil
  562. }
  563. manifests, err := GetManifestPath()
  564. if err != nil {
  565. return err
  566. }
  567. dstpath := filepath.Join(manifests, dst.Filepath())
  568. if err := os.MkdirAll(filepath.Dir(dstpath), 0o755); err != nil {
  569. return err
  570. }
  571. srcpath := filepath.Join(manifests, src.Filepath())
  572. srcfile, err := os.Open(srcpath)
  573. if err != nil {
  574. return err
  575. }
  576. defer srcfile.Close()
  577. dstfile, err := os.Create(dstpath)
  578. if err != nil {
  579. return err
  580. }
  581. defer dstfile.Close()
  582. _, err = io.Copy(dstfile, srcfile)
  583. return err
  584. }
  585. func deleteUnusedLayers(deleteMap map[string]struct{}) error {
  586. // Ignore corrupt manifests to avoid blocking deletion of layers that are freshly orphaned
  587. manifests, err := Manifests(true)
  588. if err != nil {
  589. return err
  590. }
  591. for _, manifest := range manifests {
  592. for _, layer := range manifest.Layers {
  593. delete(deleteMap, layer.Digest)
  594. }
  595. delete(deleteMap, manifest.Config.Digest)
  596. }
  597. // only delete the files which are still in the deleteMap
  598. for k := range deleteMap {
  599. fp, err := GetBlobsPath(k)
  600. if err != nil {
  601. slog.Info(fmt.Sprintf("couldn't get file path for '%s': %v", k, err))
  602. continue
  603. }
  604. if err := os.Remove(fp); err != nil {
  605. slog.Info(fmt.Sprintf("couldn't remove file '%s': %v", fp, err))
  606. continue
  607. }
  608. }
  609. return nil
  610. }
  611. func PruneLayers() error {
  612. deleteMap := make(map[string]struct{})
  613. p, err := GetBlobsPath("")
  614. if err != nil {
  615. return err
  616. }
  617. blobs, err := os.ReadDir(p)
  618. if err != nil {
  619. slog.Info(fmt.Sprintf("couldn't read dir '%s': %v", p, err))
  620. return err
  621. }
  622. for _, blob := range blobs {
  623. name := blob.Name()
  624. name = strings.ReplaceAll(name, "-", ":")
  625. _, err := GetBlobsPath(name)
  626. if err != nil {
  627. if errors.Is(err, ErrInvalidDigestFormat) {
  628. // remove invalid blobs (e.g. partial downloads)
  629. if err := os.Remove(filepath.Join(p, blob.Name())); err != nil {
  630. slog.Error("couldn't remove blob", "blob", blob.Name(), "error", err)
  631. }
  632. }
  633. continue
  634. }
  635. deleteMap[name] = struct{}{}
  636. }
  637. slog.Info(fmt.Sprintf("total blobs: %d", len(deleteMap)))
  638. if err := deleteUnusedLayers(deleteMap); err != nil {
  639. slog.Error(fmt.Sprintf("couldn't remove unused layers: %v", err))
  640. return nil
  641. }
  642. slog.Info(fmt.Sprintf("total unused blobs removed: %d", len(deleteMap)))
  643. return nil
  644. }
  645. func PruneDirectory(path string) error {
  646. info, err := os.Lstat(path)
  647. if err != nil {
  648. return err
  649. }
  650. if info.IsDir() && info.Mode()&os.ModeSymlink == 0 {
  651. entries, err := os.ReadDir(path)
  652. if err != nil {
  653. return err
  654. }
  655. for _, entry := range entries {
  656. if err := PruneDirectory(filepath.Join(path, entry.Name())); err != nil {
  657. return err
  658. }
  659. }
  660. entries, err = os.ReadDir(path)
  661. if err != nil {
  662. return err
  663. }
  664. if len(entries) > 0 {
  665. return nil
  666. }
  667. return os.Remove(path)
  668. }
  669. return nil
  670. }
  671. func PushModel(ctx context.Context, name string, regOpts *registryOptions, fn func(api.ProgressResponse)) error {
  672. mp := ParseModelPath(name)
  673. fn(api.ProgressResponse{Status: "retrieving manifest"})
  674. if mp.ProtocolScheme == "http" && !regOpts.Insecure {
  675. return errors.New("insecure protocol http")
  676. }
  677. if mp.Namespace != strings.ToLower(mp.Namespace) {
  678. return fmt.Errorf("namespace must be lowercase, but is %s", mp.Namespace)
  679. }
  680. if mp.Repository != strings.ToLower(mp.Repository) {
  681. return fmt.Errorf("model name must be lowercase, but is %s", mp.Repository)
  682. }
  683. manifest, _, err := GetManifest(mp)
  684. if err != nil {
  685. fn(api.ProgressResponse{Status: "couldn't retrieve manifest"})
  686. return err
  687. }
  688. var layers []Layer
  689. layers = append(layers, manifest.Layers...)
  690. if manifest.Config.Digest != "" {
  691. layers = append(layers, manifest.Config)
  692. }
  693. for _, layer := range layers {
  694. if err := uploadBlob(ctx, mp, layer, regOpts, fn); err != nil {
  695. slog.Info(fmt.Sprintf("error uploading blob: %v", err))
  696. return err
  697. }
  698. }
  699. fn(api.ProgressResponse{Status: "pushing manifest"})
  700. requestURL := mp.BaseURL()
  701. requestURL = requestURL.JoinPath("v2", mp.GetNamespaceRepository(), "manifests", mp.Tag)
  702. manifestJSON, err := json.Marshal(manifest)
  703. if err != nil {
  704. return err
  705. }
  706. headers := make(http.Header)
  707. headers.Set("Content-Type", "application/vnd.docker.distribution.manifest.v2+json")
  708. resp, err := makeRequestWithRetry(ctx, http.MethodPut, requestURL, headers, bytes.NewReader(manifestJSON), regOpts)
  709. if err != nil {
  710. return err
  711. }
  712. defer resp.Body.Close()
  713. fn(api.ProgressResponse{Status: "success"})
  714. return nil
  715. }
  716. func PullModel(ctx context.Context, name string, regOpts *registryOptions, fn func(api.ProgressResponse)) error {
  717. mp := ParseModelPath(name)
  718. // build deleteMap to prune unused layers
  719. deleteMap := make(map[string]struct{})
  720. manifest, _, err := GetManifest(mp)
  721. if errors.Is(err, os.ErrNotExist) {
  722. // noop
  723. } else if err != nil {
  724. slog.Warn("pulling model with bad existing manifest", "name", name, "error", err)
  725. } else {
  726. for _, l := range manifest.Layers {
  727. deleteMap[l.Digest] = struct{}{}
  728. }
  729. if manifest.Config.Digest != "" {
  730. deleteMap[manifest.Config.Digest] = struct{}{}
  731. }
  732. }
  733. if mp.ProtocolScheme == "http" && !regOpts.Insecure {
  734. return errors.New("insecure protocol http")
  735. }
  736. fn(api.ProgressResponse{Status: "pulling manifest"})
  737. manifest, err = pullModelManifest(ctx, mp, regOpts)
  738. if err != nil {
  739. return fmt.Errorf("pull model manifest: %s", err)
  740. }
  741. var layers []Layer
  742. layers = append(layers, manifest.Layers...)
  743. if manifest.Config.Digest != "" {
  744. layers = append(layers, manifest.Config)
  745. }
  746. skipVerify := make(map[string]bool)
  747. for _, layer := range layers {
  748. cacheHit, err := downloadBlob(ctx, downloadOpts{
  749. mp: mp,
  750. digest: layer.Digest,
  751. regOpts: regOpts,
  752. fn: fn,
  753. })
  754. if err != nil {
  755. return err
  756. }
  757. skipVerify[layer.Digest] = cacheHit
  758. delete(deleteMap, layer.Digest)
  759. }
  760. delete(deleteMap, manifest.Config.Digest)
  761. fn(api.ProgressResponse{Status: "verifying sha256 digest"})
  762. for _, layer := range layers {
  763. if skipVerify[layer.Digest] {
  764. continue
  765. }
  766. if err := verifyBlob(layer.Digest); err != nil {
  767. if errors.Is(err, errDigestMismatch) {
  768. // something went wrong, delete the blob
  769. fp, err := GetBlobsPath(layer.Digest)
  770. if err != nil {
  771. return err
  772. }
  773. if err := os.Remove(fp); err != nil {
  774. // log this, but return the original error
  775. slog.Info(fmt.Sprintf("couldn't remove file with digest mismatch '%s': %v", fp, err))
  776. }
  777. }
  778. return err
  779. }
  780. }
  781. fn(api.ProgressResponse{Status: "writing manifest"})
  782. manifestJSON, err := json.Marshal(manifest)
  783. if err != nil {
  784. return err
  785. }
  786. fp, err := mp.GetManifestPath()
  787. if err != nil {
  788. return err
  789. }
  790. if err := os.MkdirAll(filepath.Dir(fp), 0o755); err != nil {
  791. return err
  792. }
  793. err = os.WriteFile(fp, manifestJSON, 0o644)
  794. if err != nil {
  795. slog.Info(fmt.Sprintf("couldn't write to %s", fp))
  796. return err
  797. }
  798. if !envconfig.NoPrune() && len(deleteMap) > 0 {
  799. fn(api.ProgressResponse{Status: "removing unused layers"})
  800. if err := deleteUnusedLayers(deleteMap); err != nil {
  801. fn(api.ProgressResponse{Status: fmt.Sprintf("couldn't remove unused layers: %v", err)})
  802. }
  803. }
  804. fn(api.ProgressResponse{Status: "success"})
  805. return nil
  806. }
  807. func pullModelManifest(ctx context.Context, mp ModelPath, regOpts *registryOptions) (*Manifest, error) {
  808. requestURL := mp.BaseURL().JoinPath("v2", mp.GetNamespaceRepository(), "manifests", mp.Tag)
  809. headers := make(http.Header)
  810. headers.Set("Accept", "application/vnd.docker.distribution.manifest.v2+json")
  811. resp, err := makeRequestWithRetry(ctx, http.MethodGet, requestURL, headers, nil, regOpts)
  812. if err != nil {
  813. return nil, err
  814. }
  815. defer resp.Body.Close()
  816. var m Manifest
  817. if err := json.NewDecoder(resp.Body).Decode(&m); err != nil {
  818. return nil, err
  819. }
  820. return &m, err
  821. }
  822. // GetSHA256Digest returns the SHA256 hash of a given buffer and returns it, and the size of buffer
  823. func GetSHA256Digest(r io.Reader) (string, int64) {
  824. h := sha256.New()
  825. n, err := io.Copy(h, r)
  826. if err != nil {
  827. log.Fatal(err)
  828. }
  829. return fmt.Sprintf("sha256:%x", h.Sum(nil)), n
  830. }
  831. var errUnauthorized = errors.New("unauthorized: access denied")
  832. func makeRequestWithRetry(ctx context.Context, method string, requestURL *url.URL, headers http.Header, body io.ReadSeeker, regOpts *registryOptions) (*http.Response, error) {
  833. for range 2 {
  834. resp, err := makeRequest(ctx, method, requestURL, headers, body, regOpts)
  835. if err != nil {
  836. if !errors.Is(err, context.Canceled) {
  837. slog.Info(fmt.Sprintf("request failed: %v", err))
  838. }
  839. return nil, err
  840. }
  841. switch {
  842. case resp.StatusCode == http.StatusUnauthorized:
  843. resp.Body.Close()
  844. // Handle authentication error with one retry
  845. challenge := parseRegistryChallenge(resp.Header.Get("www-authenticate"))
  846. token, err := getAuthorizationToken(ctx, challenge)
  847. if err != nil {
  848. return nil, err
  849. }
  850. regOpts.Token = token
  851. if body != nil {
  852. _, err = body.Seek(0, io.SeekStart)
  853. if err != nil {
  854. return nil, err
  855. }
  856. }
  857. case resp.StatusCode == http.StatusNotFound:
  858. resp.Body.Close()
  859. return nil, os.ErrNotExist
  860. case resp.StatusCode >= http.StatusBadRequest:
  861. defer resp.Body.Close()
  862. responseBody, err := io.ReadAll(resp.Body)
  863. if err != nil {
  864. return nil, fmt.Errorf("%d: %s", resp.StatusCode, err)
  865. }
  866. return nil, fmt.Errorf("%d: %s", resp.StatusCode, responseBody)
  867. default:
  868. return resp, nil
  869. }
  870. }
  871. return nil, errUnauthorized
  872. }
  873. // testMakeRequestDialContext specifies the dial function for the http client in
  874. // makeRequest. It can be used to resolve hosts in model names to local
  875. // addresses for testing. For example, the model name ("example.com/my/model")
  876. // can be directed to push/pull from "127.0.0.1:1234".
  877. //
  878. // This is not safe to set across goroutines. It should be set in
  879. // the main test goroutine, and not by tests marked to run in parallel with
  880. // t.Parallel().
  881. //
  882. // It should be cleared after use, otherwise it will affect other tests.
  883. //
  884. // Ideally we would have some set this up the stack, but the code is not
  885. // structured in a way that makes this easy, so this will have to do for now.
  886. var testMakeRequestDialContext func(ctx context.Context, network, addr string) (net.Conn, error)
  887. func makeRequest(ctx context.Context, method string, requestURL *url.URL, headers http.Header, body io.Reader, regOpts *registryOptions) (*http.Response, error) {
  888. if requestURL.Scheme != "http" && regOpts != nil && regOpts.Insecure {
  889. requestURL.Scheme = "http"
  890. }
  891. req, err := http.NewRequestWithContext(ctx, method, requestURL.String(), body)
  892. if err != nil {
  893. return nil, err
  894. }
  895. if headers != nil {
  896. req.Header = headers
  897. }
  898. if regOpts != nil {
  899. if regOpts.Token != "" {
  900. req.Header.Set("Authorization", "Bearer "+regOpts.Token)
  901. } else if regOpts.Username != "" && regOpts.Password != "" {
  902. req.SetBasicAuth(regOpts.Username, regOpts.Password)
  903. }
  904. }
  905. req.Header.Set("User-Agent", fmt.Sprintf("ollama/%s (%s %s) Go/%s", version.Version, runtime.GOARCH, runtime.GOOS, runtime.Version()))
  906. if s := req.Header.Get("Content-Length"); s != "" {
  907. contentLength, err := strconv.ParseInt(s, 10, 64)
  908. if err != nil {
  909. return nil, err
  910. }
  911. req.ContentLength = contentLength
  912. }
  913. c := &http.Client{
  914. CheckRedirect: regOpts.CheckRedirect,
  915. }
  916. if testMakeRequestDialContext != nil {
  917. tr := http.DefaultTransport.(*http.Transport).Clone()
  918. tr.DialContext = testMakeRequestDialContext
  919. c.Transport = tr
  920. }
  921. return c.Do(req)
  922. }
  923. func getValue(header, key string) string {
  924. startIdx := strings.Index(header, key+"=")
  925. if startIdx == -1 {
  926. return ""
  927. }
  928. // Move the index to the starting quote after the key.
  929. startIdx += len(key) + 2
  930. endIdx := startIdx
  931. for endIdx < len(header) {
  932. if header[endIdx] == '"' {
  933. if endIdx+1 < len(header) && header[endIdx+1] != ',' { // If the next character isn't a comma, continue
  934. endIdx++
  935. continue
  936. }
  937. break
  938. }
  939. endIdx++
  940. }
  941. return header[startIdx:endIdx]
  942. }
  943. func parseRegistryChallenge(authStr string) registryChallenge {
  944. authStr = strings.TrimPrefix(authStr, "Bearer ")
  945. return registryChallenge{
  946. Realm: getValue(authStr, "realm"),
  947. Service: getValue(authStr, "service"),
  948. Scope: getValue(authStr, "scope"),
  949. }
  950. }
  951. var errDigestMismatch = errors.New("digest mismatch, file must be downloaded again")
  952. func verifyBlob(digest string) error {
  953. fp, err := GetBlobsPath(digest)
  954. if err != nil {
  955. return err
  956. }
  957. f, err := os.Open(fp)
  958. if err != nil {
  959. return err
  960. }
  961. defer f.Close()
  962. fileDigest, _ := GetSHA256Digest(f)
  963. if digest != fileDigest {
  964. return fmt.Errorf("%w: want %s, got %s", errDigestMismatch, digest, fileDigest)
  965. }
  966. return nil
  967. }