routes.go 7.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362
  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. "dario.cat/mergo"
  17. "github.com/gin-contrib/cors"
  18. "github.com/gin-gonic/gin"
  19. "github.com/jmorganca/ollama/api"
  20. "github.com/jmorganca/ollama/llama"
  21. )
  22. var loaded struct {
  23. mu sync.Mutex
  24. llm *llama.LLM
  25. expireAt time.Time
  26. expireTimer *time.Timer
  27. digest string
  28. options api.Options
  29. }
  30. func GenerateHandler(c *gin.Context) {
  31. loaded.mu.Lock()
  32. defer loaded.mu.Unlock()
  33. checkpointStart := time.Now()
  34. var req api.GenerateRequest
  35. if err := c.ShouldBindJSON(&req); err != nil {
  36. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  37. return
  38. }
  39. model, err := GetModel(req.Model)
  40. if err != nil {
  41. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  42. return
  43. }
  44. if model.Digest != loaded.digest || !reflect.DeepEqual(loaded.options, req.Options) {
  45. if loaded.llm != nil {
  46. loaded.llm.Close()
  47. loaded.llm = nil
  48. }
  49. opts := api.DefaultOptions()
  50. if err := mergo.Merge(&opts, model.Options, mergo.WithOverride); err != nil {
  51. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  52. return
  53. }
  54. if err := mergo.Merge(&opts, req.Options, mergo.WithOverride); err != nil {
  55. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  56. return
  57. }
  58. llm, err := llama.New(model.ModelPath, opts)
  59. if err != nil {
  60. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  61. return
  62. }
  63. loaded.llm = llm
  64. loaded.digest = model.Digest
  65. }
  66. sessionDuration := 5 * time.Minute
  67. loaded.expireAt = time.Now().Add(sessionDuration)
  68. if loaded.expireTimer == nil {
  69. loaded.expireTimer = time.AfterFunc(sessionDuration, func() {
  70. loaded.mu.Lock()
  71. defer loaded.mu.Unlock()
  72. if time.Now().Before(loaded.expireAt) {
  73. return
  74. }
  75. if loaded.llm == nil {
  76. return
  77. }
  78. loaded.llm.Close()
  79. loaded.llm = nil
  80. })
  81. }
  82. loaded.expireTimer.Reset(sessionDuration)
  83. checkpointLoaded := time.Now()
  84. prompt, err := model.Prompt(req)
  85. if err != nil {
  86. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  87. return
  88. }
  89. ch := make(chan any)
  90. go func() {
  91. defer close(ch)
  92. fn := func(r api.GenerateResponse) {
  93. loaded.expireAt = time.Now().Add(sessionDuration)
  94. loaded.expireTimer.Reset(sessionDuration)
  95. r.Model = req.Model
  96. r.CreatedAt = time.Now().UTC()
  97. if r.Done {
  98. r.TotalDuration = time.Since(checkpointStart)
  99. r.LoadDuration = checkpointLoaded.Sub(checkpointStart)
  100. }
  101. ch <- r
  102. }
  103. if err := loaded.llm.Predict(req.Context, prompt, fn); err != nil {
  104. ch <- gin.H{"error": err.Error()}
  105. }
  106. }()
  107. streamResponse(c, ch)
  108. }
  109. func PullModelHandler(c *gin.Context) {
  110. var req api.PullRequest
  111. if err := c.ShouldBindJSON(&req); err != nil {
  112. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  113. return
  114. }
  115. ch := make(chan any)
  116. go func() {
  117. defer close(ch)
  118. fn := func(r api.ProgressResponse) {
  119. ch <- r
  120. }
  121. regOpts := &RegistryOptions{
  122. Insecure: req.Insecure,
  123. Username: req.Username,
  124. Password: req.Password,
  125. }
  126. if err := PullModel(req.Name, regOpts, fn); err != nil {
  127. ch <- gin.H{"error": err.Error()}
  128. }
  129. }()
  130. streamResponse(c, ch)
  131. }
  132. func PushModelHandler(c *gin.Context) {
  133. var req api.PushRequest
  134. if err := c.ShouldBindJSON(&req); err != nil {
  135. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  136. return
  137. }
  138. ch := make(chan any)
  139. go func() {
  140. defer close(ch)
  141. fn := func(r api.ProgressResponse) {
  142. ch <- r
  143. }
  144. regOpts := &RegistryOptions{
  145. Insecure: req.Insecure,
  146. Username: req.Username,
  147. Password: req.Password,
  148. }
  149. if err := PushModel(req.Name, regOpts, fn); err != nil {
  150. ch <- gin.H{"error": err.Error()}
  151. }
  152. }()
  153. streamResponse(c, ch)
  154. }
  155. func CreateModelHandler(c *gin.Context) {
  156. var req api.CreateRequest
  157. if err := c.ShouldBindJSON(&req); err != nil {
  158. c.JSON(http.StatusBadRequest, gin.H{"message": err.Error()})
  159. return
  160. }
  161. ch := make(chan any)
  162. go func() {
  163. defer close(ch)
  164. fn := func(resp api.ProgressResponse) {
  165. ch <- resp
  166. }
  167. if err := CreateModel(req.Name, req.Path, fn); err != nil {
  168. ch <- gin.H{"error": err.Error()}
  169. }
  170. }()
  171. streamResponse(c, ch)
  172. }
  173. func DeleteModelHandler(c *gin.Context) {
  174. var req api.DeleteRequest
  175. if err := c.ShouldBindJSON(&req); err != nil {
  176. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  177. return
  178. }
  179. if err := DeleteModel(req.Name); err != nil {
  180. if os.IsNotExist(err) {
  181. c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Name)})
  182. } else {
  183. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  184. }
  185. return
  186. }
  187. }
  188. func ListModelsHandler(c *gin.Context) {
  189. var models []api.ListResponseModel
  190. fp, err := GetManifestPath()
  191. if err != nil {
  192. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  193. return
  194. }
  195. err = filepath.Walk(fp, func(path string, info os.FileInfo, err error) error {
  196. if err != nil {
  197. if errors.Is(err, os.ErrNotExist) {
  198. log.Printf("manifest file does not exist: %s", fp)
  199. return nil
  200. }
  201. return err
  202. }
  203. if !info.IsDir() {
  204. fi, err := os.Stat(path)
  205. if err != nil {
  206. log.Printf("skipping file: %s", fp)
  207. return nil
  208. }
  209. path := path[len(fp)+1:]
  210. slashIndex := strings.LastIndex(path, "/")
  211. if slashIndex == -1 {
  212. return nil
  213. }
  214. tag := path[:slashIndex] + ":" + path[slashIndex+1:]
  215. mp := ParseModelPath(tag)
  216. manifest, err := GetManifest(mp)
  217. if err != nil {
  218. log.Printf("skipping file: %s", fp)
  219. return nil
  220. }
  221. model := api.ListResponseModel{
  222. Name: mp.GetShortTagname(),
  223. Size: manifest.GetTotalSize(),
  224. ModifiedAt: fi.ModTime(),
  225. }
  226. models = append(models, model)
  227. }
  228. return nil
  229. })
  230. if err != nil {
  231. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  232. return
  233. }
  234. c.JSON(http.StatusOK, api.ListResponse{Models: models})
  235. }
  236. func CopyModelHandler(c *gin.Context) {
  237. var req api.CopyRequest
  238. if err := c.ShouldBindJSON(&req); err != nil {
  239. c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
  240. return
  241. }
  242. if err := CopyModel(req.Source, req.Destination); err != nil {
  243. if os.IsNotExist(err) {
  244. c.JSON(http.StatusNotFound, gin.H{"error": fmt.Sprintf("model '%s' not found", req.Source)})
  245. } else {
  246. c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
  247. }
  248. return
  249. }
  250. }
  251. func Serve(ln net.Listener) error {
  252. config := cors.DefaultConfig()
  253. config.AllowWildcard = true
  254. // only allow http/https from localhost
  255. config.AllowOrigins = []string{
  256. "http://localhost",
  257. "http://localhost:*",
  258. "https://localhost",
  259. "https://localhost:*",
  260. "http://127.0.0.1",
  261. "http://127.0.0.1:*",
  262. "https://127.0.0.1",
  263. "https://127.0.0.1:*",
  264. }
  265. r := gin.Default()
  266. r.Use(cors.New(config))
  267. r.GET("/", func(c *gin.Context) {
  268. c.String(http.StatusOK, "Ollama is running")
  269. })
  270. r.HEAD("/", func(c *gin.Context) {
  271. c.Status(http.StatusOK)
  272. })
  273. r.POST("/api/pull", PullModelHandler)
  274. r.POST("/api/generate", GenerateHandler)
  275. r.POST("/api/create", CreateModelHandler)
  276. r.POST("/api/push", PushModelHandler)
  277. r.POST("/api/copy", CopyModelHandler)
  278. r.GET("/api/tags", ListModelsHandler)
  279. r.DELETE("/api/delete", DeleteModelHandler)
  280. log.Printf("Listening on %s", ln.Addr())
  281. s := &http.Server{
  282. Handler: r,
  283. }
  284. return s.Serve(ln)
  285. }
  286. func streamResponse(c *gin.Context, ch chan any) {
  287. c.Stream(func(w io.Writer) bool {
  288. val, ok := <-ch
  289. if !ok {
  290. return false
  291. }
  292. bts, err := json.Marshal(val)
  293. if err != nil {
  294. log.Printf("streamResponse: json.Marshal failed with %s", err)
  295. return false
  296. }
  297. bts = append(bts, '\n')
  298. if _, err := w.Write(bts); err != nil {
  299. log.Printf("streamResponse: w.Write failed with %s", err)
  300. return false
  301. }
  302. return true
  303. })
  304. }