feat: add claude-3.7-sonnet 128k output support

docker-compose.yml

@@ -25,12 +25,12 @@ services:
       - db
 
   redis:
-    image: redis:latest
+    image: redis:7.4
     container_name: redis
     restart: always
 
   db:
-    image: mysql:8.2.0
+    image: mysql:8.2
     container_name: mysql
     restart: always
     volumes:

relay/channel/aws/adaptor.go

@@ -48,7 +48,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, info *relaycommon.RelayInfo, re
 
 	var claudeReq *claude.ClaudeRequest
 	var err error
-	claudeReq, err = claude.RequestOpenAI2ClaudeMessage(*request)
+	claudeReq, err = claude.RequestOpenAI2ClaudeMessage(c, *request)
 
 	c.Set("request_model", request.Model)
 	c.Set("converted_request", claudeReq)

relay/channel/claude/adaptor.go

@@ -70,9 +70,9 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, info *relaycommon.RelayInfo, re
 		return nil, errors.New("request is nil")
 	}
 	if a.RequestMode == RequestModeCompletion {
-		return RequestOpenAI2ClaudeComplete(*request), nil
+		return RequestOpenAI2ClaudeComplete(c, *request), nil
 	} else {
-		return RequestOpenAI2ClaudeMessage(*request)
+		return RequestOpenAI2ClaudeMessage(c, *request)
 	}
 }
 

relay/channel/claude/relay-claude.go

@@ -28,7 +28,7 @@ func stopReasonClaude2OpenAI(reason string) string {
 	}
 }
 
-func RequestOpenAI2ClaudeComplete(textRequest dto.GeneralOpenAIRequest) *ClaudeRequest {
+func RequestOpenAI2ClaudeComplete(c *gin.Context, textRequest dto.GeneralOpenAIRequest) *ClaudeRequest {
 
 	claudeRequest := ClaudeRequest{
 		Model:         textRequest.Model,
@@ -59,16 +59,18 @@ func RequestOpenAI2ClaudeComplete(textRequest dto.GeneralOpenAIRequest) *ClaudeR
 	return &claudeRequest
 }
 
-func RequestOpenAI2ClaudeMessage(textRequest dto.GeneralOpenAIRequest) (*ClaudeRequest, error) {
+func RequestOpenAI2ClaudeMessage(c *gin.Context, textRequest dto.GeneralOpenAIRequest) (*ClaudeRequest, error) {
 	claudeTools := make([]Tool, 0, len(textRequest.Tools))
 
 	if strings.HasSuffix(textRequest.Model, "-thinking") {
 		textRequest.Model = strings.TrimSuffix(textRequest.Model, "-thinking")
 
 		if textRequest.MaxTokens == 0 {
-			textRequest.MaxTokens = 4096
+			textRequest.MaxTokens = 8192
-		} else if textRequest.MaxTokens < 1280 {
+		} else if textRequest.MaxTokens < 2048 {
-			textRequest.MaxTokens = 1280
+			textRequest.MaxTokens = 2048
+		} else if textRequest.MaxTokens > 64000 {
+			c.Request.Header.Set("anthropic-beta", "output-128k-2025-02-19")
 		}
 
 		textRequest.TopP = 0
@@ -76,7 +78,7 @@ func RequestOpenAI2ClaudeMessage(textRequest dto.GeneralOpenAIRequest) (*ClaudeR
 		textRequest.Temperature = 0
 		textRequest.Thinking = &dto.Thinking{
 			Type:         "enabled",
-			BudgetTokens: int(float64(textRequest.MaxTokens) * 0.8),
+			BudgetTokens: int(float64(textRequest.MaxTokens) * 0.5),
 		}
 	}
 

relay/channel/vertex/adaptor.go

@@ -122,7 +122,7 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, info *relaycommon.RelayInfo, re
 		return nil, errors.New("request is nil")
 	}
 	if a.RequestMode == RequestModeClaude {
-		claudeReq, err := claude.RequestOpenAI2ClaudeMessage(*request)
+		claudeReq, err := claude.RequestOpenAI2ClaudeMessage(c, *request)
 		if err != nil {
 			return nil, err
 		}