routes.go 7.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366
  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. opts := api.DefaultOptions()
  44. if err := opts.FromMap(model.Options); err != nil {
  45. log.Printf("could not load model options: %v", err)
  46. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  47. return
  48. }
  49. if err := opts.FromMap(req.Options); err != nil {
  50. log.Printf("could not merge model options: %v", err)
  51. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  52. return
  53. }
  54. if model.Digest != loaded.digest || !reflect.DeepEqual(loaded.options, opts) {
  55. if loaded.llm != nil {
  56. loaded.llm.Close()
  57. loaded.llm = nil
  58. loaded.digest = ""
  59. }
  60. llm, err := llama.New(model.ModelPath, opts)
  61. if err != nil {
  62. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  63. return
  64. }
  65. loaded.llm = llm
  66. loaded.digest = model.Digest
  67. loaded.options = opts
  68. }
  69. sessionDuration := 5 * time.Minute
  70. loaded.expireAt = time.Now().Add(sessionDuration)
  71. if loaded.expireTimer == nil {
  72. loaded.expireTimer = time.AfterFunc(sessionDuration, func() {
  73. loaded.mu.Lock()
  74. defer loaded.mu.Unlock()
  75. if time.Now().Before(loaded.expireAt) {
  76. return
  77. }
  78. if loaded.llm == nil {
  79. return
  80. }
  81. loaded.llm.Close()
  82. loaded.llm = nil
  83. loaded.digest = ""
  84. })
  85. }
  86. loaded.expireTimer.Reset(sessionDuration)
  87. checkpointLoaded := time.Now()
  88. prompt, err := model.Prompt(req)
  89. if err != nil {
  90. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  91. return
  92. }
  93. ch := make(chan any)
  94. go func() {
  95. defer close(ch)
  96. fn := func(r api.GenerateResponse) {
  97. loaded.expireAt = time.Now().Add(sessionDuration)
  98. loaded.expireTimer.Reset(sessionDuration)
  99. r.Model = req.Model
  100. r.CreatedAt = time.Now().UTC()
  101. if r.Done {
  102. r.TotalDuration = time.Since(checkpointStart)
  103. r.LoadDuration = checkpointLoaded.Sub(checkpointStart)
  104. }
  105. ch <- r
  106. }
  107. if err := loaded.llm.Predict(req.Context, prompt, fn); err != nil {
  108. ch <- gin.H{"error": err.Error()}
  109. }
  110. }()
  111. streamResponse(c, ch)
  112. }
  113. func PullModelHandler(c *gin.Context) {
  114. var req api.PullRequest
  115. if err := c.ShouldBindJSON(&req); err != nil {
  116. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  117. return
  118. }
  119. ch := make(chan any)
  120. go func() {
  121. defer close(ch)
  122. fn := func(r api.ProgressResponse) {
  123. ch <- r
  124. }
  125. regOpts := &RegistryOptions{
  126. Insecure: req.Insecure,
  127. Username: req.Username,
  128. Password: req.Password,
  129. }
  130. if err := PullModel(req.Name, regOpts, fn); err != nil {
  131. ch <- gin.H{"error": err.Error()}
  132. }
  133. }()
  134. streamResponse(c, ch)
  135. }
  136. func PushModelHandler(c *gin.Context) {
  137. var req api.PushRequest
  138. if err := c.ShouldBindJSON(&req); err != nil {
  139. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  140. return
  141. }
  142. ch := make(chan any)
  143. go func() {
  144. defer close(ch)
  145. fn := func(r api.ProgressResponse) {
  146. ch <- r
  147. }
  148. regOpts := &RegistryOptions{
  149. Insecure: req.Insecure,
  150. Username: req.Username,
  151. Password: req.Password,
  152. }
  153. if err := PushModel(req.Name, regOpts, fn); err != nil {
  154. ch <- gin.H{"error": err.Error()}
  155. }
  156. }()
  157. streamResponse(c, ch)
  158. }
  159. func CreateModelHandler(c *gin.Context) {
  160. var req api.CreateRequest
  161. if err := c.ShouldBindJSON(&req); err != nil {
  162. c.JSON(http.StatusBadRequest, gin.H{"message": err.Error()})
  163. return
  164. }
  165. ch := make(chan any)
  166. go func() {
  167. defer close(ch)
  168. fn := func(resp api.ProgressResponse) {
  169. ch <- resp
  170. }
  171. if err := CreateModel(req.Name, req.Path, fn); err != nil {
  172. ch <- gin.H{"error": err.Error()}
  173. }
  174. }()
  175. streamResponse(c, ch)
  176. }
  177. func DeleteModelHandler(c *gin.Context) {
  178. var req api.DeleteRequest
  179. if err := c.ShouldBindJSON(&req); err != nil {
  180. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  181. return
  182. }
  183. if err := DeleteModel(req.Name); err != nil {
  184. if os.IsNotExist(err) {
  185. c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Name)})
  186. } else {
  187. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  188. }
  189. return
  190. }
  191. }
  192. func ListModelsHandler(c *gin.Context) {
  193. var models []api.ListResponseModel
  194. fp, err := GetManifestPath()
  195. if err != nil {
  196. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  197. return
  198. }
  199. err = filepath.Walk(fp, func(path string, info os.FileInfo, err error) error {
  200. if err != nil {
  201. if errors.Is(err, os.ErrNotExist) {
  202. log.Printf("manifest file does not exist: %s", fp)
  203. return nil
  204. }
  205. return err
  206. }
  207. if !info.IsDir() {
  208. fi, err := os.Stat(path)
  209. if err != nil {
  210. log.Printf("skipping file: %s", fp)
  211. return nil
  212. }
  213. path := path[len(fp)+1:]
  214. slashIndex := strings.LastIndex(path, "/")
  215. if slashIndex == -1 {
  216. return nil
  217. }
  218. tag := path[:slashIndex] + ":" + path[slashIndex+1:]
  219. mp := ParseModelPath(tag)
  220. manifest, err := GetManifest(mp)
  221. if err != nil {
  222. log.Printf("skipping file: %s", fp)
  223. return nil
  224. }
  225. model := api.ListResponseModel{
  226. Name: mp.GetShortTagname(),
  227. Size: manifest.GetTotalSize(),
  228. ModifiedAt: fi.ModTime(),
  229. }
  230. models = append(models, model)
  231. }
  232. return nil
  233. })
  234. if err != nil {
  235. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  236. return
  237. }
  238. c.JSON(http.StatusOK, api.ListResponse{Models: models})
  239. }
  240. func CopyModelHandler(c *gin.Context) {
  241. var req api.CopyRequest
  242. if err := c.ShouldBindJSON(&req); err != nil {
  243. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  244. return
  245. }
  246. if err := CopyModel(req.Source, req.Destination); err != nil {
  247. if os.IsNotExist(err) {
  248. c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Source)})
  249. } else {
  250. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  251. }
  252. return
  253. }
  254. }
  255. func Serve(ln net.Listener) error {
  256. config := cors.DefaultConfig()
  257. config.AllowWildcard = true
  258. // only allow http/https from localhost
  259. config.AllowOrigins = []string{
  260. "http://localhost",
  261. "http://localhost:*",
  262. "https://localhost",
  263. "https://localhost:*",
  264. "http://127.0.0.1",
  265. "http://127.0.0.1:*",
  266. "https://127.0.0.1",
  267. "https://127.0.0.1:*",
  268. }
  269. r := gin.Default()
  270. r.Use(cors.New(config))
  271. r.GET("/", func(c *gin.Context) {
  272. c.String(http.StatusOK, "Ollama is running")
  273. })
  274. r.HEAD("/", func(c *gin.Context) {
  275. c.Status(http.StatusOK)
  276. })
  277. r.POST("/api/pull", PullModelHandler)
  278. r.POST("/api/generate", GenerateHandler)
  279. r.POST("/api/create", CreateModelHandler)
  280. r.POST("/api/push", PushModelHandler)
  281. r.POST("/api/copy", CopyModelHandler)
  282. r.GET("/api/tags", ListModelsHandler)
  283. r.DELETE("/api/delete", DeleteModelHandler)
  284. log.Printf("Listening on %s", ln.Addr())
  285. s := &http.Server{
  286. Handler: r,
  287. }
  288. return s.Serve(ln)
  289. }
  290. func streamResponse(c *gin.Context, ch chan any) {
  291. c.Stream(func(w io.Writer) bool {
  292. val, ok := <-ch
  293. if !ok {
  294. return false
  295. }
  296. bts, err := json.Marshal(val)
  297. if err != nil {
  298. log.Printf("streamResponse: json.Marshal failed with %s", err)
  299. return false
  300. }
  301. bts = append(bts, '\n')
  302. if _, err := w.Write(bts); err != nil {
  303. log.Printf("streamResponse: w.Write failed with %s", err)
  304. return false
  305. }
  306. return true
  307. })
  308. }