apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-07 01:22:59 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Feature/online/vs think 20250813 (#3440)

Browse Source 
* add stream

* fix ernie_vl_reasoning_parsers

* fix bug
This commit is contained in:
luukunn
2025-08-15 18:33:58 +08:00
committed by GitHub
parent 33abfddd9b
commit edf1ca07af
8 changed files with 206 additions and 194 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
fastdeploy/input/ernie_processor.py
View File

@@ -50,8 +50,9 @@ class ErnieProcessor(BaseDataProcessor):
self._init_config()
self.decode_status = dict()
self.tool_parsers = dict()
self.tool_parser_dict = dict()
self.thinking_parser_dict = dict()
self.reasoning_end_dict = dict()
self._load_tokenizer()
data_processor_logger.info(
f"tokenizer information: bos_token is {self.tokenizer.bos_token} \
@@ -291,10 +292,12 @@ class ErnieProcessor(BaseDataProcessor):
token_ids = token_ids[:-1]
delta_text, previous_token_ids, previous_texts = self.ids2tokens(token_ids, req_id)
response_dict["outputs"]["raw_prediction"] = delta_text
if self.reasoning_parser and (
enable_thinking or self.reasoning_parser.__class__.__name__ == "ErnieX1ReasoningParser"
if (
self.reasoning_parser
and req_id not in self.reasoning_end_dict
and (enable_thinking or self.reasoning_parser.__class__.__name__ == "ErnieX1ReasoningParser")
):
reasoning_content, text = self.reasoning_parser.extract_reasoning_content_streaming(
reasoning_delta_message = self.reasoning_parser.extract_reasoning_content_streaming(
previous_texts,
previous_texts + delta_text,
delta_text,
@@ -302,14 +305,13 @@ class ErnieProcessor(BaseDataProcessor):
previous_token_ids + token_ids,
token_ids,
)
response_dict["outputs"]["text"] = text
response_dict["outputs"]["reasoning_content"] = reasoning_content
else:
response_dict["outputs"]["text"] = delta_text
if self.tool_parser_obj:
if req_id not in self.tool_parsers:
self.tool_parsers[req_id] = self.tool_parser_obj(self.tokenizer)
tool_parser = self.tool_parsers[req_id]
response_dict["outputs"]["reasoning_delta_message"] = reasoning_delta_message
if self.reasoning_parser.is_reasoning_end(previous_token_ids + token_ids):
self.reasoning_end_dict[req_id] = True
if self.tool_parser_obj and req_id in self.reasoning_end_dict:
if req_id not in self.tool_parser_dict:
self.tool_parser_dict[req_id] = self.tool_parser_obj(self.tokenizer)
tool_parser = self.tool_parser_dict[req_id]
tool_call = tool_parser.extract_tool_calls_streaming(
previous_texts,
previous_texts + delta_text,
@@ -320,11 +322,14 @@ class ErnieProcessor(BaseDataProcessor):
response_dict,
)
response_dict["outputs"]["tool_delta_message"] = tool_call
response_dict["outputs"]["text"] = delta_text
if is_end:
data_processor_logger.info(f"req_id:{req_id}, decode_status: {self.decode_status[req_id]}")
del self.decode_status[req_id]
if req_id in self.tool_parsers:
del self.tool_parsers[req_id]
if req_id in self.tool_parser_dict:
del self.tool_parser_dict[req_id]
if req_id in self.reasoning_end_dict:
del self.reasoning_end_dict[req_id]
return response_dict
def messages2ids(self, request_or_messages):