model.go 5.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256
  1. package server
  2. import (
  3. "archive/zip"
  4. "bytes"
  5. "context"
  6. "errors"
  7. "fmt"
  8. "io"
  9. "net/http"
  10. "os"
  11. "path/filepath"
  12. "sync"
  13. "github.com/ollama/ollama/api"
  14. "github.com/ollama/ollama/convert"
  15. "github.com/ollama/ollama/llm"
  16. "github.com/ollama/ollama/types/model"
  17. )
  18. var intermediateBlobs sync.Map
  19. type layerWithGGML struct {
  20. *Layer
  21. *llm.GGML
  22. }
  23. func parseFromModel(ctx context.Context, name model.Name, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
  24. modelpath := ParseModelPath(name.String())
  25. manifest, _, err := GetManifest(modelpath)
  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. modelpath = ParseModelPath(name.String())
  32. manifest, _, err = GetManifest(modelpath)
  33. if err != nil {
  34. return nil, err
  35. }
  36. case err != nil:
  37. return nil, err
  38. }
  39. for _, layer := range manifest.Layers {
  40. layer, err := NewLayerFromLayer(layer.Digest, layer.MediaType, modelpath.GetShortTagname())
  41. if err != nil {
  42. return nil, err
  43. }
  44. switch layer.MediaType {
  45. case "application/vnd.ollama.image.model",
  46. "application/vnd.ollama.image.projector",
  47. "application/vnd.ollama.image.adapter":
  48. blobpath, err := GetBlobsPath(layer.Digest)
  49. if err != nil {
  50. return nil, err
  51. }
  52. blob, err := os.Open(blobpath)
  53. if err != nil {
  54. return nil, err
  55. }
  56. defer blob.Close()
  57. ggml, _, err := llm.DecodeGGML(blob)
  58. if err != nil {
  59. return nil, err
  60. }
  61. layers = append(layers, &layerWithGGML{layer, ggml})
  62. default:
  63. layers = append(layers, &layerWithGGML{layer, nil})
  64. }
  65. }
  66. return layers, nil
  67. }
  68. func parseFromZipFile(_ context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
  69. stat, err := file.Stat()
  70. if err != nil {
  71. return nil, err
  72. }
  73. r, err := zip.NewReader(file, stat.Size())
  74. if err != nil {
  75. return nil, err
  76. }
  77. tempdir, err := os.MkdirTemp(filepath.Dir(file.Name()), "")
  78. if err != nil {
  79. return nil, err
  80. }
  81. defer os.RemoveAll(tempdir)
  82. fn(api.ProgressResponse{Status: "unpacking model metadata"})
  83. for _, f := range r.File {
  84. // TODO(mxyng): this should not write out all files to disk
  85. outfile, err := os.Create(filepath.Join(tempdir, f.Name))
  86. if err != nil {
  87. return nil, err
  88. }
  89. defer outfile.Close()
  90. infile, err := f.Open()
  91. if err != nil {
  92. return nil, err
  93. }
  94. defer infile.Close()
  95. if _, err = io.Copy(outfile, infile); err != nil {
  96. return nil, err
  97. }
  98. if err := outfile.Close(); err != nil {
  99. return nil, err
  100. }
  101. if err := infile.Close(); err != nil {
  102. return nil, err
  103. }
  104. }
  105. mf, err := convert.GetModelFormat(tempdir)
  106. if err != nil {
  107. return nil, err
  108. }
  109. params, err := mf.GetParams(tempdir)
  110. if err != nil {
  111. return nil, err
  112. }
  113. mArch, err := mf.GetModelArch("", tempdir, params)
  114. if err != nil {
  115. return nil, err
  116. }
  117. fn(api.ProgressResponse{Status: "processing tensors"})
  118. if err := mArch.GetTensors(); err != nil {
  119. return nil, err
  120. }
  121. if err := mArch.LoadVocab(); err != nil {
  122. return nil, err
  123. }
  124. fn(api.ProgressResponse{Status: "converting model"})
  125. // TODO(mxyng): this should write directly into a layer
  126. // e.g. NewLayer(arch.Reader(), "application/vnd.ollama.image.model")
  127. temp, err := os.CreateTemp(tempdir, "fp16")
  128. if err != nil {
  129. return nil, err
  130. }
  131. defer temp.Close()
  132. defer os.Remove(temp.Name())
  133. if err = mArch.WriteGGUF(temp); err != nil {
  134. return nil, err
  135. }
  136. if _, err := temp.Seek(0, io.SeekStart); err != nil {
  137. return nil, err
  138. }
  139. layer, err := NewLayer(temp, "application/vnd.ollama.image.model")
  140. if err != nil {
  141. return nil, fmt.Errorf("aaa: %w", err)
  142. }
  143. bin, err := layer.Open()
  144. if err != nil {
  145. return nil, err
  146. }
  147. defer bin.Close()
  148. ggml, _, err := llm.DecodeGGML(bin)
  149. if err != nil {
  150. return nil, err
  151. }
  152. layers = append(layers, &layerWithGGML{layer, ggml})
  153. intermediateBlobs.Store(digest, layer.Digest)
  154. return layers, nil
  155. }
  156. func parseFromFile(ctx context.Context, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerWithGGML, err error) {
  157. sr := io.NewSectionReader(file, 0, 512)
  158. contentType, err := detectContentType(sr)
  159. if err != nil {
  160. return nil, err
  161. }
  162. switch contentType {
  163. case "gguf", "ggla":
  164. // noop
  165. case "application/zip":
  166. return parseFromZipFile(ctx, file, digest, fn)
  167. default:
  168. return nil, fmt.Errorf("unsupported content type: %s", contentType)
  169. }
  170. stat, err := file.Stat()
  171. if err != nil {
  172. return nil, err
  173. }
  174. var offset int64
  175. for offset < stat.Size() {
  176. ggml, n, err := llm.DecodeGGML(file)
  177. if errors.Is(err, io.EOF) {
  178. break
  179. } else if err != nil {
  180. return nil, err
  181. }
  182. mediatype := "application/vnd.ollama.image.model"
  183. if ggml.Name() == "ggla" {
  184. mediatype = "application/vnd.ollama.image.adapter"
  185. } else if ggml.KV().Architecture() == "clip" {
  186. mediatype = "application/vnd.ollama.image.projector"
  187. }
  188. layer, err := NewLayer(io.NewSectionReader(file, offset, n), mediatype)
  189. if err != nil {
  190. return nil, err
  191. }
  192. layers = append(layers, &layerWithGGML{layer, ggml})
  193. offset = n
  194. }
  195. return layers, nil
  196. }
  197. func detectContentType(r io.Reader) (string, error) {
  198. var b bytes.Buffer
  199. if _, err := io.Copy(&b, r); err != nil {
  200. return "", err
  201. }
  202. if contentType := llm.DetectGGMLType(b.Bytes()); contentType != "" {
  203. return contentType, nil
  204. }
  205. if contentType := http.DetectContentType(b.Bytes()); contentType != "application/octet-stream" {
  206. return contentType, nil
  207. }
  208. return "unknown", nil
  209. }