apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Release/2.1 (#3361)

Browse Source 
* [BugFix] v1/completions add finish_reason

* update TestOpenAIServingCompletion for merge
This commit is contained in:
memoryCoderC
2025-08-12 23:06:51 +08:00
committed by GitHub
parent b4bb54b56b
commit fe2094609f
2 changed files with 128 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
fastdeploy/entrypoints/openai/serving_completion.py
View File

@@ -234,6 +234,15 @@ class OpenAIServingCompletion:
dealer.close()
self.engine_client.semaphore.release()
def calc_finish_reason(self, max_tokens, token_num, output):
if max_tokens is None or token_num != max_tokens:
if self.engine_client.reasoning_parser == "ernie_x1" and output.get("finish_reason", "") == "tool_calls":
return "tool_calls"
else:
return "stop"
else:
return "length"
async def completion_stream_generator(
self,
request: CompletionRequest,
@@ -334,19 +343,13 @@ class OpenAIServingCompletion:
logprobs=logprobs_res,
)
)
if res["finished"]:
if request.max_tokens is None or output_tokens[idx] + 1 != request.max_tokens:
chunk.choices[0].finish_reason = "stop"
if (
self.engine_client.reasoning_parser == "ernie_x1"
and output.get("finish_reason", "") == "tool_calls"
):
chunk.choices[0].finish_reason = "tool_calls"
else:
chunk.choices[0].finish_reason = "length"
output_tokens[idx] += 1
if res["finished"]:
choices[-1].finish_reason = self.calc_finish_reason(
request.max_tokens, output_tokens[idx], output
)
if len(choices) == max_streaming_response_tokens or res["finished"]:
chunk = CompletionStreamResponse(
id=request_id,
@@ -432,6 +435,8 @@ class OpenAIServingCompletion:
token_ids = output["token_ids"]
output_text = output["text"]
finish_reason = self.calc_finish_reason(request.max_tokens, final_res["output_token_ids"], output)
choice_data = CompletionResponseChoice(
token_ids=token_ids,
index=len(choices),
@@ -441,7 +446,7 @@ class OpenAIServingCompletion:
reasoning_content=output.get("reasoning_content"),
tool_calls=output.get("tool_call_content"),
logprobs=aggregated_logprobs,
finish_reason=None,
finish_reason=finish_reason,
)
choices.append(choice_data)