From 30db3e9d8f2ab7c7cb18b515d0eed08162577b73 Mon Sep 17 00:00:00 2001 From: Zhang Yulong <35552275+ZhangYulongg@users.noreply.github.com> Date: Mon, 20 Apr 2026 19:40:17 +0800 Subject: [PATCH] [benchmark] update tools (#7512) --- benchmarks/backend_request_func.py | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/benchmarks/backend_request_func.py b/benchmarks/backend_request_func.py index 079df8ba9f..4b19a88dfc 100644 --- a/benchmarks/backend_request_func.py +++ b/benchmarks/backend_request_func.py @@ -428,6 +428,15 @@ async def async_request_eb_openai_chat_completions( # output.generated_text = generated_text # 在流式结束时，记录最后一个 chunk 收到的时间戳 output.end_timestamp = most_recent_timestamp + # 截断case也记录usage + usage = data.get("usage", {}) + if usage: + output.output_tokens = usage.get("completion_tokens", 0) + output.prompt_tokens = usage.get("prompt_tokens", 0) + if output.prompt_len == 0: + prompt_details = usage.get("prompt_tokens_details", {}) + if prompt_details: + output.prompt_len = prompt_details.get("cached_tokens", 0) if tool_call_buffer: for _, tc in tool_call_buffer.items():