Merge remote-tracking branch 'upstream/master'

fix: fallback to max_completion_tokens for unsupported models

pkg/ai/openai_client.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"strings"
 	"time"
 )
 
@@ -23,12 +24,13 @@ type ChatMessage struct {
 }
 
 type ChatCompletionRequest struct {
-	Model       string        `json:"model"`
+	Model               string        `json:"model"`
-	Messages    []ChatMessage `json:"messages"`
+	Messages            []ChatMessage `json:"messages"`
-	Temperature float64       `json:"temperature,omitempty"`
+	Temperature         float64       `json:"temperature,omitempty"`
-	MaxTokens   int           `json:"max_tokens,omitempty"`
+	MaxTokens           *int          `json:"max_tokens,omitempty"`
-	TopP        float64       `json:"top_p,omitempty"`
+	MaxCompletionTokens *int          `json:"max_completion_tokens,omitempty"`
-	Stream      bool          `json:"stream,omitempty"`
+	TopP                float64       `json:"top_p,omitempty"`
+	Stream              bool          `json:"stream,omitempty"`
 }
 
 type ChatCompletionResponse struct {
@@ -89,6 +91,24 @@ func (c *OpenAIClient) ChatCompletion(messages []ChatMessage, options ...func(*C
 }
 
 func (c *OpenAIClient) sendChatRequest(req *ChatCompletionRequest) (*ChatCompletionResponse, error) {
+	resp, err := c.doChatRequest(req)
+	if err == nil {
+		return resp, nil
+	}
+
+	if shouldRetryWithMaxCompletionTokens(err, req) {
+		tokens := *req.MaxTokens
+		retryReq := *req
+		retryReq.MaxTokens = nil
+		retryReq.MaxCompletionTokens = &tokens
+		fmt.Printf("OpenAI: retrying with max_completion_tokens=%d\n", tokens)
+		return c.doChatRequest(&retryReq)
+	}
+
+	return nil, err
+}
+
+func (c *OpenAIClient) doChatRequest(req *ChatCompletionRequest) (*ChatCompletionResponse, error) {
 	jsonData, err := json.Marshal(req)
 	if err != nil {
 		fmt.Printf("OpenAI: Failed to marshal request: %v\n", err)
@@ -192,7 +212,7 @@ func WithTemperature(temp float64) func(*ChatCompletionRequest) {
 
 func WithMaxTokens(tokens int) func(*ChatCompletionRequest) {
 	return func(req *ChatCompletionRequest) {
-		req.MaxTokens = tokens
+		req.MaxTokens = &tokens
 	}
 }
 
@@ -247,3 +267,21 @@ func (c *OpenAIClient) TestConnection() error {
 	}
 	return err
 }
+
+func shouldRetryWithMaxCompletionTokens(err error, req *ChatCompletionRequest) bool {
+	if err == nil || req == nil || req.MaxTokens == nil || req.MaxCompletionTokens != nil {
+		return false
+	}
+
+	msg := err.Error()
+	if strings.Contains(msg, "Unsupported parameter: 'max_tokens'") {
+		return true
+	}
+	if strings.Contains(msg, "max_tokens is not supported") {
+		return true
+	}
+	if strings.Contains(msg, "max_completion_tokens") {
+		return true
+	}
+	return false
+}