model.go 6.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291
  1. package server
  2. import (
  3. "archive/zip"
  4. "bytes"
  5. "context"
  6. "errors"
  7. "fmt"
  8. "io"
  9. "log/slog"
  10. "net/http"
  11. "os"
  12. "path/filepath"
  13. "github.com/ollama/ollama/api"
  14. "github.com/ollama/ollama/convert"
  15. "github.com/ollama/ollama/llm"
  16. "github.com/ollama/ollama/templates"
  17. "github.com/ollama/ollama/types/model"
  18. )
  19. var intermediateBlobs map[string]string = make(map[string]string)
  20. type layerGGML struct {
  21. *Layer
  22. *llm.GGML
  23. }
  24. func parseFromModel(ctx context.Context, name model.Name, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
  25. m, err := ParseNamedManifest(name)
  26. switch {
  27. case errors.Is(err, os.ErrNotExist):
  28. if err := PullModel(ctx, name.String(), &registryOptions{}, fn); err != nil {
  29. return nil, err
  30. }
  31. m, err = ParseNamedManifest(name)
  32. if err != nil {
  33. return nil, err
  34. }
  35. case err != nil:
  36. return nil, err
  37. }
  38. for _, layer := range m.Layers {
  39. layer, err := NewLayerFromLayer(layer.Digest, layer.MediaType, name.DisplayShortest())
  40. if err != nil {
  41. return nil, err
  42. }
  43. switch layer.MediaType {
  44. case "application/vnd.ollama.image.model",
  45. "application/vnd.ollama.image.projector",
  46. "application/vnd.ollama.image.adapter":
  47. blobpath, err := GetBlobsPath(layer.Digest)
  48. if err != nil {
  49. return nil, err
  50. }
  51. blob, err := os.Open(blobpath)
  52. if err != nil {
  53. return nil, err
  54. }
  55. defer blob.Close()
  56. ggml, _, err := llm.DecodeGGML(blob, 0)
  57. if err != nil {
  58. return nil, err
  59. }
  60. layers = append(layers, &layerGGML{layer, ggml})
  61. default:
  62. layers = append(layers, &layerGGML{layer, nil})
  63. }
  64. }
  65. return layers, nil
  66. }
  67. func extractFromZipFile(p string, file *os.File, fn func(api.ProgressResponse)) error {
  68. stat, err := file.Stat()
  69. if err != nil {
  70. return err
  71. }
  72. r, err := zip.NewReader(file, stat.Size())
  73. if err != nil {
  74. return err
  75. }
  76. fn(api.ProgressResponse{Status: "unpacking model metadata"})
  77. for _, f := range r.File {
  78. if !filepath.IsLocal(f.Name) {
  79. return fmt.Errorf("%w: %s", zip.ErrInsecurePath, f.Name)
  80. }
  81. n := filepath.Join(p, f.Name)
  82. if err := os.MkdirAll(filepath.Dir(n), 0o750); err != nil {
  83. return err
  84. }
  85. // TODO(mxyng): this should not write out all files to disk
  86. outfile, err := os.Create(n)
  87. if err != nil {
  88. return err
  89. }
  90. defer outfile.Close()
  91. infile, err := f.Open()
  92. if err != nil {
  93. return err
  94. }
  95. defer infile.Close()
  96. if _, err = io.Copy(outfile, infile); err != nil {
  97. return err
  98. }
  99. if err := outfile.Close(); err != nil {
  100. return err
  101. }
  102. if err := infile.Close(); err != nil {
  103. return err
  104. }
  105. }
  106. return nil
  107. }
  108. func parseFromZipFile(_ context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
  109. tempDir, err := os.MkdirTemp(filepath.Dir(file.Name()), "")
  110. if err != nil {
  111. return nil, err
  112. }
  113. defer os.RemoveAll(tempDir)
  114. if err := extractFromZipFile(tempDir, file, fn); err != nil {
  115. return nil, err
  116. }
  117. mf, err := convert.GetModelFormat(tempDir)
  118. if err != nil {
  119. return nil, err
  120. }
  121. params, err := mf.GetParams(tempDir)
  122. if err != nil {
  123. return nil, err
  124. }
  125. mArch, err := mf.GetModelArch("", tempDir, params)
  126. if err != nil {
  127. return nil, err
  128. }
  129. fn(api.ProgressResponse{Status: "processing tensors"})
  130. if err := mArch.GetTensors(); err != nil {
  131. return nil, err
  132. }
  133. if err := mArch.LoadVocab(); err != nil {
  134. return nil, err
  135. }
  136. fn(api.ProgressResponse{Status: "converting model"})
  137. // TODO(mxyng): this should write directly into a layer
  138. // e.g. NewLayer(arch.Reader(), "application/vnd.ollama.image.model")
  139. temp, err := os.CreateTemp(tempDir, "fp16")
  140. if err != nil {
  141. return nil, err
  142. }
  143. defer temp.Close()
  144. defer os.Remove(temp.Name())
  145. if err = mArch.WriteGGUF(temp); err != nil {
  146. return nil, err
  147. }
  148. if _, err := temp.Seek(0, io.SeekStart); err != nil {
  149. return nil, err
  150. }
  151. layer, err := NewLayer(temp, "application/vnd.ollama.image.model")
  152. if err != nil {
  153. return nil, err
  154. }
  155. bin, err := layer.Open()
  156. if err != nil {
  157. return nil, err
  158. }
  159. defer bin.Close()
  160. ggml, _, err := llm.DecodeGGML(bin, 0)
  161. if err != nil {
  162. return nil, err
  163. }
  164. layers = append(layers, &layerGGML{layer, ggml})
  165. intermediateBlobs[digest] = layer.Digest
  166. return detectChatTemplate(layers)
  167. }
  168. func parseFromFile(ctx context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
  169. sr := io.NewSectionReader(file, 0, 512)
  170. contentType, err := detectContentType(sr)
  171. if err != nil {
  172. return nil, err
  173. }
  174. switch contentType {
  175. case "gguf", "ggla":
  176. // noop
  177. case "application/zip":
  178. return parseFromZipFile(ctx, file, digest, fn)
  179. default:
  180. return nil, fmt.Errorf("unsupported content type: %s", contentType)
  181. }
  182. stat, err := file.Stat()
  183. if err != nil {
  184. return nil, err
  185. }
  186. var offset int64
  187. for offset < stat.Size() {
  188. ggml, n, err := llm.DecodeGGML(file, 0)
  189. if errors.Is(err, io.EOF) {
  190. break
  191. } else if err != nil {
  192. return nil, err
  193. }
  194. mediatype := "application/vnd.ollama.image.model"
  195. if ggml.Name() == "ggla" {
  196. mediatype = "application/vnd.ollama.image.adapter"
  197. } else if ggml.KV().Architecture() == "clip" {
  198. mediatype = "application/vnd.ollama.image.projector"
  199. }
  200. layer, err := NewLayer(io.NewSectionReader(file, offset, n), mediatype)
  201. if err != nil {
  202. return nil, err
  203. }
  204. layers = append(layers, &layerGGML{layer, ggml})
  205. offset = n
  206. }
  207. return detectChatTemplate(layers)
  208. }
  209. func detectChatTemplate(layers []*layerGGML) ([]*layerGGML, error) {
  210. for _, layer := range layers {
  211. if s := layer.GGML.KV().ChatTemplate(); s != "" {
  212. if t, err := templates.NamedTemplate(s); err != nil {
  213. slog.Debug("template detection", "error", err)
  214. } else {
  215. tmpl, err := NewLayer(t.Reader(), "application/vnd.ollama.image.template")
  216. if err != nil {
  217. return nil, err
  218. }
  219. tmpl.status = fmt.Sprintf("using autodetected template %s", t.Name)
  220. layers = append(layers, &layerGGML{tmpl, nil})
  221. }
  222. }
  223. }
  224. return layers, nil
  225. }
  226. func detectContentType(r io.Reader) (string, error) {
  227. var b bytes.Buffer
  228. if _, err := io.Copy(&b, r); err != nil {
  229. return "", err
  230. }
  231. if contentType := llm.DetectGGMLType(b.Bytes()); contentType != "" {
  232. return contentType, nil
  233. }
  234. if contentType := http.DetectContentType(b.Bytes()); contentType != "application/octet-stream" {
  235. return contentType, nil
  236. }
  237. return "unknown", nil
  238. }