openai: finish streaming tool calls as tool_calls

openai/openai.go

@@ -266,6 +266,9 @@ func toChunk(id string, r api.ChatResponse) ChatCompletionChunk {
 			Index: 0,
 			Delta: Message{Role: "assistant", Content: r.Message.Content, ToolCalls: toolCalls},
 			FinishReason: func(reason string) *string {
+				if len(toolCalls) > 0 {
+					reason = "tool_calls"
+				}
 				if len(reason) > 0 {
 					return &reason
 				}
@@ -571,6 +574,7 @@ type BaseWriter struct {
 
 type ChatWriter struct {
 	stream   bool
+	finished bool
 	id       string
 	BaseWriter
 }
@@ -620,7 +624,10 @@ func (w *ChatWriter) writeResponse(data []byte) (int, error) {
 
 	// chat chunk
 	if w.stream {
-		d, err := json.Marshal(toChunk(w.id, chatResponse))
+		// If we've already finished, don't send any more chunks with choices.
+		if !w.finished {
+			chunk := toChunk(w.id, chatResponse)
+			d, err := json.Marshal(chunk)
 			if err != nil {
 				return 0, err
 			}
@@ -630,6 +637,7 @@ func (w *ChatWriter) writeResponse(data []byte) (int, error) {
 			if err != nil {
 				return 0, err
 			}
+		}
 
 		if chatResponse.Done {
 			_, err = w.ResponseWriter.Write([]byte("data: [DONE]\n\n"))