routes.go 7.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359
  1. package server
  2. import (
  3. "encoding/json"
  4. "errors"
  5. "fmt"
  6. "io"
  7. "log"
  8. "net"
  9. "net/http"
  10. "os"
  11. "path/filepath"
  12. "reflect"
  13. "strings"
  14. "sync"
  15. "time"
  16. "github.com/gin-contrib/cors"
  17. "github.com/gin-gonic/gin"
  18. "github.com/jmorganca/ollama/api"
  19. "github.com/jmorganca/ollama/llama"
  20. )
  21. var loaded struct {
  22. mu sync.Mutex
  23. llm *llama.LLM
  24. expireAt time.Time
  25. expireTimer *time.Timer
  26. digest string
  27. options api.Options
  28. }
  29. func GenerateHandler(c *gin.Context) {
  30. loaded.mu.Lock()
  31. defer loaded.mu.Unlock()
  32. checkpointStart := time.Now()
  33. var req api.GenerateRequest
  34. if err := c.ShouldBindJSON(&req); err != nil {
  35. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  36. return
  37. }
  38. model, err := GetModel(req.Model)
  39. if err != nil {
  40. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  41. return
  42. }
  43. if model.Digest != loaded.digest || !reflect.DeepEqual(loaded.options, req.Options) {
  44. if loaded.llm != nil {
  45. loaded.llm.Close()
  46. loaded.llm = nil
  47. }
  48. opts := api.DefaultOptions()
  49. if err := opts.FromMap(model.Options); err != nil {
  50. log.Printf("could not load model options: %v", err)
  51. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  52. return
  53. }
  54. if err := opts.FromMap(req.Options); err != nil {
  55. log.Printf("could not merge model options: %v", err)
  56. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  57. return
  58. }
  59. llm, err := llama.New(model.ModelPath, opts)
  60. if err != nil {
  61. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  62. return
  63. }
  64. loaded.llm = llm
  65. loaded.digest = model.Digest
  66. }
  67. sessionDuration := 5 * time.Minute
  68. loaded.expireAt = time.Now().Add(sessionDuration)
  69. if loaded.expireTimer == nil {
  70. loaded.expireTimer = time.AfterFunc(sessionDuration, func() {
  71. loaded.mu.Lock()
  72. defer loaded.mu.Unlock()
  73. if time.Now().Before(loaded.expireAt) {
  74. return
  75. }
  76. if loaded.llm == nil {
  77. return
  78. }
  79. loaded.llm.Close()
  80. loaded.llm = nil
  81. })
  82. }
  83. loaded.expireTimer.Reset(sessionDuration)
  84. checkpointLoaded := time.Now()
  85. prompt, err := model.Prompt(req)
  86. if err != nil {
  87. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  88. return
  89. }
  90. ch := make(chan any)
  91. go func() {
  92. defer close(ch)
  93. fn := func(r api.GenerateResponse) {
  94. loaded.expireAt = time.Now().Add(sessionDuration)
  95. loaded.expireTimer.Reset(sessionDuration)
  96. r.Model = req.Model
  97. r.CreatedAt = time.Now().UTC()
  98. if r.Done {
  99. r.TotalDuration = time.Since(checkpointStart)
  100. r.LoadDuration = checkpointLoaded.Sub(checkpointStart)
  101. }
  102. ch <- r
  103. }
  104. if err := loaded.llm.Predict(req.Context, prompt, fn); err != nil {
  105. ch <- gin.H{"error": err.Error()}
  106. }
  107. }()
  108. streamResponse(c, ch)
  109. }
  110. func PullModelHandler(c *gin.Context) {
  111. var req api.PullRequest
  112. if err := c.ShouldBindJSON(&req); err != nil {
  113. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  114. return
  115. }
  116. ch := make(chan any)
  117. go func() {
  118. defer close(ch)
  119. fn := func(r api.ProgressResponse) {
  120. ch <- r
  121. }
  122. regOpts := &RegistryOptions{
  123. Insecure: req.Insecure,
  124. Username: req.Username,
  125. Password: req.Password,
  126. }
  127. if err := PullModel(req.Name, regOpts, fn); err != nil {
  128. ch <- gin.H{"error": err.Error()}
  129. }
  130. }()
  131. streamResponse(c, ch)
  132. }
  133. func PushModelHandler(c *gin.Context) {
  134. var req api.PushRequest
  135. if err := c.ShouldBindJSON(&req); err != nil {
  136. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  137. return
  138. }
  139. ch := make(chan any)
  140. go func() {
  141. defer close(ch)
  142. fn := func(r api.ProgressResponse) {
  143. ch <- r
  144. }
  145. regOpts := &RegistryOptions{
  146. Insecure: req.Insecure,
  147. Username: req.Username,
  148. Password: req.Password,
  149. }
  150. if err := PushModel(req.Name, regOpts, fn); err != nil {
  151. ch <- gin.H{"error": err.Error()}
  152. }
  153. }()
  154. streamResponse(c, ch)
  155. }
  156. func CreateModelHandler(c *gin.Context) {
  157. var req api.CreateRequest
  158. if err := c.ShouldBindJSON(&req); err != nil {
  159. c.JSON(http.StatusBadRequest, gin.H{"message": err.Error()})
  160. return
  161. }
  162. ch := make(chan any)
  163. go func() {
  164. defer close(ch)
  165. fn := func(resp api.ProgressResponse) {
  166. ch <- resp
  167. }
  168. if err := CreateModel(req.Name, req.Path, fn); err != nil {
  169. ch <- gin.H{"error": err.Error()}
  170. }
  171. }()
  172. streamResponse(c, ch)
  173. }
  174. func DeleteModelHandler(c *gin.Context) {
  175. var req api.DeleteRequest
  176. if err := c.ShouldBindJSON(&req); err != nil {
  177. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  178. return
  179. }
  180. if err := DeleteModel(req.Name); err != nil {
  181. if os.IsNotExist(err) {
  182. c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Name)})
  183. } else {
  184. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  185. }
  186. return
  187. }
  188. }
  189. func ListModelsHandler(c *gin.Context) {
  190. var models []api.ListResponseModel
  191. fp, err := GetManifestPath()
  192. if err != nil {
  193. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  194. return
  195. }
  196. err = filepath.Walk(fp, func(path string, info os.FileInfo, err error) error {
  197. if err != nil {
  198. if errors.Is(err, os.ErrNotExist) {
  199. log.Printf("manifest file does not exist: %s", fp)
  200. return nil
  201. }
  202. return err
  203. }
  204. if !info.IsDir() {
  205. fi, err := os.Stat(path)
  206. if err != nil {
  207. log.Printf("skipping file: %s", fp)
  208. return nil
  209. }
  210. path := path[len(fp)+1:]
  211. slashIndex := strings.LastIndex(path, "/")
  212. if slashIndex == -1 {
  213. return nil
  214. }
  215. tag := path[:slashIndex] + ":" + path[slashIndex+1:]
  216. mp := ParseModelPath(tag)
  217. manifest, err := GetManifest(mp)
  218. if err != nil {
  219. log.Printf("skipping file: %s", fp)
  220. return nil
  221. }
  222. model := api.ListResponseModel{
  223. Name: mp.GetShortTagname(),
  224. Size: manifest.GetTotalSize(),
  225. ModifiedAt: fi.ModTime(),
  226. }
  227. models = append(models, model)
  228. }
  229. return nil
  230. })
  231. if err != nil {
  232. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  233. return
  234. }
  235. c.JSON(http.StatusOK, api.ListResponse{Models: models})
  236. }
  237. func CopyModelHandler(c *gin.Context) {
  238. var req api.CopyRequest
  239. if err := c.ShouldBindJSON(&req); err != nil {
  240. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  241. return
  242. }
  243. if err := CopyModel(req.Source, req.Destination); err != nil {
  244. if os.IsNotExist(err) {
  245. c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Source)})
  246. } else {
  247. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  248. }
  249. return
  250. }
  251. }
  252. func Serve(ln net.Listener) error {
  253. config := cors.DefaultConfig()
  254. config.AllowWildcard = true
  255. // only allow http/https from localhost
  256. config.AllowOrigins = []string{
  257. "http://localhost",
  258. "http://localhost:*",
  259. "https://localhost",
  260. "https://localhost:*",
  261. "http://127.0.0.1",
  262. "http://127.0.0.1:*",
  263. "https://127.0.0.1",
  264. "https://127.0.0.1:*",
  265. }
  266. r := gin.Default()
  267. r.Use(cors.New(config))
  268. r.GET("/", func(c *gin.Context) {
  269. c.String(http.StatusOK, "Ollama is running")
  270. })
  271. r.POST("/api/pull", PullModelHandler)
  272. r.POST("/api/generate", GenerateHandler)
  273. r.POST("/api/create", CreateModelHandler)
  274. r.POST("/api/push", PushModelHandler)
  275. r.POST("/api/copy", CopyModelHandler)
  276. r.GET("/api/tags", ListModelsHandler)
  277. r.DELETE("/api/delete", DeleteModelHandler)
  278. log.Printf("Listening on %s", ln.Addr())
  279. s := &http.Server{
  280. Handler: r,
  281. }
  282. return s.Serve(ln)
  283. }
  284. func streamResponse(c *gin.Context, ch chan any) {
  285. c.Stream(func(w io.Writer) bool {
  286. val, ok := <-ch
  287. if !ok {
  288. return false
  289. }
  290. bts, err := json.Marshal(val)
  291. if err != nil {
  292. log.Printf("streamResponse: json.Marshal failed with %s", err)
  293. return false
  294. }
  295. bts = append(bts, '\n')
  296. if _, err := w.Write(bts); err != nil {
  297. log.Printf("streamResponse: w.Write failed with %s", err)
  298. return false
  299. }
  300. return true
  301. })
  302. }