Fixes usage calculation in streaming mode

Corrects the usage calculation for streaming responses by passing the correct argument to the base function.

It ensures accurate token counting when `n` > 1 is requested, preventing potential discrepancies in billing or rate limiting.

Signed-off-by: Xinyuan Tong <[email protected]>

Author: JustinTong0323

python/sglang/srt/entrypoints/openai/serving_chat.py

@@ -578,7 +578,7 @@ async def generate_stream_resp():
                 # Final chunk with usage
                 if request.stream_options and request.stream_options.include_usage:
                     usage = self._calculate_streaming_usage_base(
-                        prompt_tokens, completion_tokens, cached_tokens, request
+                        prompt_tokens, completion_tokens, cached_tokens, request.n
                     )
                 else:
                     usage = None

python/sglang/srt/entrypoints/openai/serving_completions.py

@@ -289,7 +289,7 @@ async def generate_stream_resp():
                 # Handle final usage chunk
                 if request.stream_options and request.stream_options.include_usage:
                     usage = self._calculate_streaming_usage_base(
-                        prompt_tokens, completion_tokens, cached_tokens, request
+                        prompt_tokens, completion_tokens, cached_tokens, request.n
                     )
                     final_usage_chunk = CompletionStreamResponse(
                         id=content["meta_info"]["id"],