apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Merge branch 'feature/online/vs_think_20250813' into release/2.1

Browse Source
This commit is contained in:
Jiang-Jia-Jun
2025-08-14 17:13:36 +08:00
committed by GitHub
parent 03347626a6 81092c0fe3
commit 02596fc537
23 changed files with 1056 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
fastdeploy/entrypoints/openai/serving_completion.py
View File

@@ -240,9 +240,9 @@ class OpenAIServingCompletion:
dealer.close()
self.engine_client.semaphore.release()
def calc_finish_reason(self, max_tokens, token_num, output):
def calc_finish_reason(self, max_tokens, token_num, output, tool_called):
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":
if tool_called or output.get("tool_call"):
return "tool_calls"
else:
return "stop"
@@ -271,6 +271,7 @@ class OpenAIServingCompletion:
output_tokens = [0] * num_choices
inference_start_time = [0] * num_choices
first_iteration = [True] * num_choices
tool_called = False
max_streaming_response_tokens = (
request.max_streaming_response_tokens
if request.max_streaming_response_tokens is not None
@@ -342,24 +343,41 @@ class OpenAIServingCompletion:
if request.logprobs and output_top_logprobs is not None:
logprobs_res = self._create_completion_logprobs(output_top_logprobs, request.logprobs, 0)
choices.append(
CompletionResponseStreamChoice(
output_tokens[idx] += 1
if self.engine_client.data_processor.tool_parser_obj and not res["finished"]:
tool_delta_message = output["tool_delta_message"]
if tool_delta_message is None:
continue
delta_message = CompletionResponseStreamChoice(
index=idx,
text=output["text"],
prompt_token_ids=None,
completion_token_ids=output.get("token_ids") if request.return_token_ids else None,
raw_prediction=output.get("raw_prediction") if request.return_token_ids else None,
tool_calls=output.get("tool_call_content"),
tool_calls=tool_delta_message.tool_calls,
reasoning_content=output.get("reasoning_content"),
arrival_time=arrival_time,
logprobs=logprobs_res,
)
)
if tool_delta_message.tool_calls:
tool_called = True
else:
delta_message = CompletionResponseStreamChoice(
index=idx,
text=output["text"],
prompt_token_ids=None,
completion_token_ids=output.get("token_ids") if request.return_token_ids else None,
tool_calls=None,
raw_prediction=output.get("raw_prediction") if request.return_token_ids else None,
reasoning_content=output.get("reasoning_content"),
arrival_time=arrival_time,
logprobs=logprobs_res,
)
choices.append(delta_message)
output_tokens[idx] += 1
if res["finished"]:
choices[-1].finish_reason = self.calc_finish_reason(
request.max_tokens, output_tokens[idx], output
request.max_tokens, output_tokens[idx], output, tool_called
)
send_idx = output.get("send_idx")
# 只有当 send_idx 明确为 0 时才记录日志
@@ -458,7 +476,7 @@ 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)
finish_reason = self.calc_finish_reason(request.max_tokens, final_res["output_token_ids"], output, False)
choice_data = CompletionResponseChoice(
token_ids=token_ids,
@@ -469,7 +487,7 @@ class OpenAIServingCompletion:
raw_prediction=output.get("raw_prediction") if request.return_token_ids else None,
text_after_process=text_after_process_list[idx] if request.return_token_ids else None,
reasoning_content=output.get("reasoning_content"),
tool_calls=output.get("tool_call_content"),
tool_calls=output.get("tool_call"),
logprobs=aggregated_logprobs,
finish_reason=finish_reason,
)