routes: update generate handler to use runner with harmony

server/routes.go

@@ -196,13 +196,6 @@ func (s *Server) GenerateHandler(c *gin.Context) {
 	}
 
 	useHarmony := harmony.ShouldUseHarmony(m.Config.ModelFamily, m.Template) && !req.Raw
-	var harmonyMessageHandler *harmony.HarmonyMessageHandler
-	var harmonyToolParser *harmony.HarmonyToolCallAccumulator
-	if useHarmony {
-		harmonyMessageHandler = harmony.NewHarmonyMessageHandler()
-		harmonyMessageHandler.HarmonyParser.AddImplicitStart()
-		harmonyToolParser = harmonyMessageHandler.CreateToolParser()
-	}
 
 	// Validate Think value: string values currently only allowed for gptoss models
 	if req.Think != nil && req.Think.IsString() && !useHarmony {
@@ -363,12 +356,7 @@ func (s *Server) GenerateHandler(c *gin.Context) {
 				},
 			}
 
-			if useHarmony {
-				content, thinking, toolContent := harmonyMessageHandler.AddContent(cr.Content, harmonyToolParser)
-				res.Response = content
-				res.Thinking = thinking
-				harmonyToolParser.Add(toolContent)
-			} else if thinkingState != nil {
+			if !useHarmony && thinkingState != nil {
 				thinking, content := thinkingState.AddContent(cr.Content)
 				res.Thinking = thinking
 				res.Response = content
@@ -379,26 +367,6 @@ func (s *Server) GenerateHandler(c *gin.Context) {
 			}
 
 			if cr.Done {
-				if useHarmony {
-					toolName, toolContent := harmonyToolParser.Drain()
-					if toolName != nil {
-						*toolName = strings.TrimPrefix(*toolName, "functions.")
-						var args api.ToolCallFunctionArguments
-						if err := json.Unmarshal([]byte(toolContent), &args); err != nil {
-							errStr := fmt.Sprintf("error parsing tool call: raw='%s', err=%s", toolContent, err.Error())
-							ch <- gin.H{"error": errStr}
-							return
-						}
-
-						res.ToolCalls = append(res.ToolCalls, api.ToolCall{
-							Function: api.ToolCallFunction{
-								Name:      *toolName,
-								Arguments: args,
-							},
-						})
-					}
-				}
-
 				res.DoneReason = cr.DoneReason.String()
 				res.TotalDuration = time.Since(checkpointStart)
 				res.LoadDuration = checkpointLoaded.Sub(checkpointStart)