[BugFix] fix tool call parser#7369
Conversation
|
Thanks for your contribution! |
There was a problem hiding this comment.
Pull request overview
该 PR 聚焦修复 OpenAI 入口下 ErnieX1 工具调用解析（tool call parser）在流式场景中的边界问题，尤其是 arguments={}（空字典）以及以数字/布尔值/嵌套 } 结尾时的收尾解析与增量输出。
Changes:
- 调整
ErnieX1ToolParser.extract_tool_calls()的tools_called判定逻辑，避免在未提取到工具调用时仍返回tools_called=True。 - 修复流式解析 close 分支与参数分支中对
arguments的判定：由“真假值判断”改为“None 判断”，从而正确处理空字典等 falsy 参数。 - 增加多组回归/集成测试，覆盖
arguments={}、数字/布尔结尾、嵌套对象结尾等流式边界切分。
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.
| File | Description |
|---|---|
| fastdeploy/entrypoints/openai/tool_parsers/ernie_x1_tool_parser.py | 修复流式参数分支与 close 分支对空参数的错误判定，并调整 tools_called 返回逻辑 |
| tests/entrypoints/openai/tool_parsers/test_ernie_x1_tool_parser.py | 新增回归/集成测试覆盖空参数与多种 close 结尾场景 |
| if self.tool_call_start_token not in model_output: | ||
| return ExtractedToolCallInformation(tools_called=False, tool_calls=[], content=model_output) | ||
| else: | ||
| try: | ||
| tool_call_json_list = self.tool_call_regex.findall(model_output) | ||
| tool_calls = [] | ||
| for tool_call_json in tool_call_json_list: | ||
| tool_call_dict = json.loads(tool_call_json) | ||
| args_str = json.dumps(tool_call_dict.get("arguments", {}), ensure_ascii=False) | ||
| tool_calls.append( | ||
| ToolCall( | ||
| type="function", | ||
| id=random_tool_call_id(), | ||
| function=FunctionCall( | ||
| name=tool_call_dict.get("name", ""), | ||
| arguments=args_str, | ||
| ), | ||
| ) | ||
| ) | ||
| return ExtractedToolCallInformation( | ||
| tools_called=True, | ||
| tools_called=len(tool_calls) > 0, | ||
| tool_calls=tool_calls, | ||
| ) |
| diff = self.prev_tool_call_arr[self.current_tool_id].get("arguments") | ||
| if diff: | ||
| if '"}' not in delta_text: | ||
| if diff is not None: | ||
| if "}" not in delta_text: | ||
| return None | ||
| end_loc = delta_text.rindex("}") | ||
| diff = delta_text[:end_loc] | ||
| if not diff: | ||
| return None |
| def test_streaming_empty_arguments_full_flow(self): | ||
| """Integration: streaming tool call with arguments={} must not lose arguments. | ||
|
|
||
| Simulates a complete streaming flow where the tool call has empty | ||
| arguments. Verifies the name is sent and arguments are streamed. | ||
| """ |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #7369 +/- ##
==========================================
Coverage ? 73.57%
==========================================
Files ? 383
Lines ? 53626
Branches ? 8415
==========================================
Hits ? 39456
Misses ? 11483
Partials ? 2687
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
PaddlePaddle-bot
left a comment
PaddlePaddle-bot
left a comment
There was a problem hiding this comment.
🤖 AI Code Review | 2025-04-14
📋 Review 摘要
PR 概述：修复 ErnieX1ToolParser 中空参数判断错误、tools_called 误判、以及流式输出结尾处理过于严格的问题
变更范围：entrypoints/openai/tool_parsers/
影响面 Tag：[APIServer]
📝 PR 规范检查
问题：PR 标题包含 [BugFix] Tag 且符合格式要求，描述有 Motivation 和 Modifications，但 Checklist 未勾选任何项。
建议：
- 标题格式符合要求，无需修改
- 建议更新 Checklist，至少勾选格式化和单元测试相关项
问题
| 级别 | 文件 | 概述 |
|---|---|---|
| 🟡 建议 | ernie_x1_tool_parser.py:188 |
rindex('}') 在 delta_text 不包含 } 时会抛出 ValueError |
总体评价
PR 修复逻辑正确，测试覆盖全面，但存在一个边界情况未处理。
| diff = self.prev_tool_call_arr[self.current_tool_id].get("arguments") | ||
| if diff: | ||
| if '"}' not in delta_text: | ||
| if diff is not None: |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
There was a problem hiding this comment.
review不对，并不会出现这个问题
* fix tool call parser * add unit test * fix unit test * add unit test
|
✅ Cherry-pick successful! Created PR: #7417 |
* fix tool call parser * add unit test * fix unit test * add unit test
|
ℹ️ Cherry-pick PR already exists: #7417 |
* fix tool call parser * add unit test * fix unit test * add unit test
|
✅ Cherry-pick successful! Created PR: #7418 |
* fix tool call parser * add unit test * fix unit test * add unit test
|
ℹ️ Cherry-pick PR already exists: #7417 |
* fix tool call parser * add unit test * fix unit test * add unit test
|
ℹ️ Cherry-pick PR already exists: #7418 |
* fix tool call parser * add unit test * fix unit test * add unit test
|
✅ Cherry-pick successful! Created PR: #7419 |
Motivation
extract_tool_calls中tools_called在tool_calls为空时仍返回True的问题{}被误判为无参数的问题'"}'检查过于严格导致数字/布尔值结尾的参数无法正确流式传输的问题Modifications
tools_called=True改为tools_called=len(tool_calls) > 0if diff:改为if diff is not None:，正确处理空字典if '"}' not in delta_text:改为if "}" not in delta_text:，并使用rindex("}")获取正确位置if not cur_arguments改为if cur_arguments is None，避免空字典误判Usage or Command
Accuracy Tests
Checklist
[FDConfig],[APIServer],[Engine],[Scheduler],[PD Disaggregation],[Executor],[Graph Optimization],[Speculative Decoding],[RL],[Models],[Quantization],[Loader],[OP],[KVCache],[DataProcessor],[BugFix],[Docs],[CI],[Optimization],[Feature],[Benchmark],[Others],[XPU],[HPU],[GCU],[DCU],[Iluvatar],[Metax]]pre-commitbefore commit.releasebranch, make sure the PR has been submitted to thedevelopbranch, then cherry-pick it to thereleasebranch with the[Cherry-Pick]PR tag.