model.go 8.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374
  1. package server
  2. import (
  3. "archive/zip"
  4. "bytes"
  5. "context"
  6. "encoding/json"
  7. "errors"
  8. "fmt"
  9. "io"
  10. "log/slog"
  11. "net/http"
  12. "os"
  13. "path/filepath"
  14. "slices"
  15. "strings"
  16. "text/template/parse"
  17. "github.com/ollama/ollama/api"
  18. "github.com/ollama/ollama/convert"
  19. "github.com/ollama/ollama/llm"
  20. "github.com/ollama/ollama/template"
  21. "github.com/ollama/ollama/types/model"
  22. )
  23. var intermediateBlobs map[string]string = make(map[string]string)
  24. type layerGGML struct {
  25. Layer
  26. *llm.GGML
  27. }
  28. func parseFromModel(ctx context.Context, name model.Name, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
  29. m, err := ParseNamedManifest(name)
  30. switch {
  31. case errors.Is(err, os.ErrNotExist):
  32. if err := PullModel(ctx, name.String(), &registryOptions{}, fn); err != nil {
  33. return nil, err
  34. }
  35. m, err = ParseNamedManifest(name)
  36. if err != nil {
  37. return nil, err
  38. }
  39. case err != nil:
  40. return nil, err
  41. }
  42. for _, layer := range m.Layers {
  43. layer, err := NewLayerFromLayer(layer.Digest, layer.MediaType, name.DisplayShortest())
  44. if err != nil {
  45. return nil, err
  46. }
  47. switch layer.MediaType {
  48. case "application/vnd.ollama.image.model",
  49. "application/vnd.ollama.image.projector",
  50. "application/vnd.ollama.image.adapter":
  51. blobpath, err := GetBlobsPath(layer.Digest)
  52. if err != nil {
  53. return nil, err
  54. }
  55. blob, err := os.Open(blobpath)
  56. if err != nil {
  57. return nil, err
  58. }
  59. defer blob.Close()
  60. ggml, _, err := llm.DecodeGGML(blob, 0)
  61. if err != nil {
  62. return nil, err
  63. }
  64. layers = append(layers, &layerGGML{layer, ggml})
  65. default:
  66. layers = append(layers, &layerGGML{layer, nil})
  67. }
  68. }
  69. return layers, nil
  70. }
  71. func parseFromZipFile(_ context.Context, command string, baseLayers []*layerGGML, f *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
  72. fi, err := f.Stat()
  73. if err != nil {
  74. return nil, err
  75. }
  76. r, err := zip.NewReader(f, fi.Size())
  77. if err != nil {
  78. return nil, err
  79. }
  80. p, err := os.MkdirTemp(filepath.Dir(f.Name()), "")
  81. if err != nil {
  82. return nil, err
  83. }
  84. defer os.RemoveAll(p)
  85. fn(api.ProgressResponse{Status: "converting model"})
  86. // TODO(mxyng): this should write directly into a layer
  87. // e.g. NewLayer(arch.Reader(), "application/vnd.ollama.image.model")
  88. t, err := os.CreateTemp(p, "fp16")
  89. if err != nil {
  90. return nil, err
  91. }
  92. defer t.Close()
  93. defer os.Remove(t.Name())
  94. var layerType string
  95. switch command {
  96. case "adapter":
  97. var baseModel *llm.GGML
  98. for _, l := range baseLayers {
  99. if l.GGML != nil {
  100. baseModel = l.GGML
  101. break
  102. }
  103. }
  104. if baseModel == nil {
  105. return nil, fmt.Errorf("no base model specified for the adapter")
  106. }
  107. if err := convert.ConvertAdapter(convert.NewZipReader(r, p, 32<<20), t, baseModel.KV()); err != nil {
  108. return nil, err
  109. }
  110. layerType = "application/vnd.ollama.image.adapter"
  111. case "model":
  112. if err := convert.ConvertModel(convert.NewZipReader(r, p, 32<<20), t); err != nil {
  113. return nil, err
  114. }
  115. layerType = "application/vnd.ollama.image.model"
  116. }
  117. if _, err := t.Seek(0, io.SeekStart); err != nil {
  118. return nil, err
  119. }
  120. layer, err := NewLayer(t, layerType)
  121. if err != nil {
  122. return nil, err
  123. }
  124. bin, err := layer.Open()
  125. if err != nil {
  126. return nil, err
  127. }
  128. defer bin.Close()
  129. ggml, _, err := llm.DecodeGGML(bin, 0)
  130. if err != nil {
  131. return nil, err
  132. }
  133. layers = append(layers, &layerGGML{layer, ggml})
  134. intermediateBlobs[digest] = layer.Digest
  135. return detectChatTemplate(layers)
  136. }
  137. func parseFromFile(ctx context.Context, command string, baseLayers []*layerGGML, file *os.File, digest string, fn func(api.ProgressResponse)) (layers []*layerGGML, err error) {
  138. sr := io.NewSectionReader(file, 0, 512)
  139. contentType, err := detectContentType(sr)
  140. if err != nil {
  141. return nil, err
  142. }
  143. switch contentType {
  144. case "gguf", "ggla":
  145. // noop
  146. case "application/zip":
  147. return parseFromZipFile(ctx, command, baseLayers, file, digest, fn)
  148. default:
  149. return nil, fmt.Errorf("unsupported content type: %s", contentType)
  150. }
  151. ggml, _, err := llm.DecodeGGML(file, 0)
  152. if err != nil {
  153. return nil, err
  154. }
  155. mediatype := "application/vnd.ollama.image.model"
  156. if ggml.Name() == "ggla" || ggml.KV().Kind() == "adapter" {
  157. mediatype = "application/vnd.ollama.image.adapter"
  158. }
  159. if _, ok := ggml.KV()[fmt.Sprintf("%s.vision.block_count", ggml.KV().Architecture())]; ok || ggml.KV().Kind() == "projector" {
  160. mediatype = "application/vnd.ollama.image.projector"
  161. }
  162. layer, err := NewLayerFromLayer(digest, mediatype, file.Name())
  163. if err != nil {
  164. slog.Debug("could not create new layer from layer", "error", err)
  165. }
  166. layers = append(layers, &layerGGML{layer, ggml})
  167. return detectChatTemplate(layers)
  168. }
  169. func detectChatTemplate(layers []*layerGGML) ([]*layerGGML, error) {
  170. for _, layer := range layers {
  171. if s := layer.GGML.KV().ChatTemplate(); s != "" {
  172. if t, err := template.Named(s); err != nil {
  173. slog.Debug("template detection", "error", err)
  174. } else {
  175. layer, err := NewLayer(t.Reader(), "application/vnd.ollama.image.template")
  176. if err != nil {
  177. return nil, err
  178. }
  179. layer.status = fmt.Sprintf("using autodetected template %s", t.Name)
  180. layers = append(layers, &layerGGML{layer, nil})
  181. if t.Parameters != nil {
  182. var b bytes.Buffer
  183. if err := json.NewEncoder(&b).Encode(t.Parameters); err != nil {
  184. return nil, err
  185. }
  186. layer, err := NewLayer(&b, "application/vnd.ollama.image.params")
  187. if err != nil {
  188. return nil, err
  189. }
  190. layers = append(layers, &layerGGML{layer, nil})
  191. }
  192. }
  193. }
  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. }
  210. func parseObjects(s string) []map[string]any {
  211. var objs []map[string]any
  212. for offset := 0; offset < len(s); {
  213. var obj map[string]any
  214. decoder := json.NewDecoder(strings.NewReader(s[offset:]))
  215. if err := decoder.Decode(&obj); errors.Is(err, io.EOF) || errors.Is(err, io.ErrUnexpectedEOF) {
  216. break
  217. } else if syntax := &(json.SyntaxError{}); errors.As(err, &syntax) {
  218. // skip over any syntax errors
  219. offset += int(syntax.Offset)
  220. } else if unmarshalType := &(json.UnmarshalTypeError{}); errors.As(err, &unmarshalType) {
  221. // skip over any unmarshalable types
  222. offset += int(unmarshalType.Offset)
  223. } else if err != nil {
  224. return nil
  225. } else {
  226. offset += int(decoder.InputOffset())
  227. objs = append(objs, obj)
  228. }
  229. }
  230. return objs
  231. }
  232. // parseToolCalls attempts to parse a JSON string into a slice of ToolCalls.
  233. // mxyng: this only really works if the input contains tool calls in some JSON format
  234. func (m *Model) parseToolCalls(s string) ([]api.ToolCall, bool) {
  235. // create a subtree from the node that ranges over .ToolCalls
  236. tmpl := m.Template.Subtree(func(n parse.Node) bool {
  237. if t, ok := n.(*parse.RangeNode); ok {
  238. return slices.Contains(template.Identifiers(t.Pipe), "ToolCalls")
  239. }
  240. return false
  241. })
  242. if tmpl == nil {
  243. return nil, false
  244. }
  245. var b bytes.Buffer
  246. if err := tmpl.Execute(&b, map[string][]api.ToolCall{
  247. "ToolCalls": {
  248. {
  249. Function: api.ToolCallFunction{
  250. Name: "@@name@@",
  251. Arguments: api.ToolCallFunctionArguments{
  252. "@@argument@@": 1,
  253. },
  254. },
  255. },
  256. },
  257. }); err != nil {
  258. return nil, false
  259. }
  260. templateObjects := parseObjects(b.String())
  261. if len(templateObjects) == 0 {
  262. return nil, false
  263. }
  264. // find the keys that correspond to the name and arguments fields
  265. var name, arguments string
  266. for k, v := range templateObjects[0] {
  267. switch v.(type) {
  268. case string:
  269. name = k
  270. case map[string]any:
  271. arguments = k
  272. }
  273. }
  274. if name == "" || arguments == "" {
  275. return nil, false
  276. }
  277. responseObjects := parseObjects(s)
  278. if len(responseObjects) == 0 {
  279. return nil, false
  280. }
  281. // collect all nested objects
  282. var collect func(any) []map[string]any
  283. collect = func(obj any) (all []map[string]any) {
  284. switch o := obj.(type) {
  285. case map[string]any:
  286. all = append(all, o)
  287. for _, v := range o {
  288. all = append(all, collect(v)...)
  289. }
  290. case []any:
  291. for _, v := range o {
  292. all = append(all, collect(v)...)
  293. }
  294. }
  295. return all
  296. }
  297. var objs []map[string]any
  298. for _, p := range responseObjects {
  299. objs = append(objs, collect(p)...)
  300. }
  301. var toolCalls []api.ToolCall
  302. for _, kv := range objs {
  303. n, nok := kv[name].(string)
  304. a, aok := kv[arguments].(map[string]any)
  305. if nok && aok {
  306. toolCalls = append(toolCalls, api.ToolCall{
  307. Function: api.ToolCallFunction{
  308. Name: n,
  309. Arguments: a,
  310. },
  311. })
  312. }
  313. }
  314. return toolCalls, len(toolCalls) > 0
  315. }