iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Refactor][Parser] Simplify parse_delta (#39728)

Browse Source 
Signed-off-by: sfeng33 <4florafeng@gmail.com>
This commit is contained in:
Flora Feng
2026-04-13 21:02:13 +00:00
committed by GitHub
parent 8213e8f880
commit 14cb86c187
1 changed files with 35 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/parser/abstract_parser.py
+35 -41
View File
@@ -568,6 +568,20 @@ class DelegatingParser(Parser):
return input_ids
return self._reasoning_parser.extract_content_ids(input_ids)
def _in_reasoning_phase(self, state: StreamState) -> bool:
if self._reasoning_parser is None:
return False
if self._tool_parser is None:
return True
return not state.reasoning_ended
def _in_tool_call_phase(self, state: StreamState) -> bool:
if self._tool_parser is None:
return False
if self._reasoning_parser is None:
return True
return state.reasoning_ended
def parse_delta(
self,
delta_text: str,
@@ -584,47 +598,10 @@ class DelegatingParser(Parser):
current_text = state.previous_text + delta_text
current_token_ids = state.previous_token_ids + delta_token_ids
delta_message: DeltaMessage | None = None
if self._reasoning_parser and self._tool_parser:
if not state.reasoning_ended:
delta_message = self.extract_reasoning_streaming(
previous_text=state.previous_text,
current_text=current_text,
delta_text=delta_text,
previous_token_ids=state.previous_token_ids,
current_token_ids=current_token_ids,
delta_token_ids=delta_token_ids,
)
if self.is_reasoning_end(delta_token_ids):
state.reasoning_ended = True
current_token_ids = self.extract_content_ids(delta_token_ids)
if delta_message and delta_message.content:
current_text = delta_message.content
delta_message.content = None
else:
current_text = ""
if state.reasoning_ended:
if not state.tool_call_text_started:
state.tool_call_text_started = True
state.previous_text = ""
state.previous_token_ids = []
delta_text = current_text
delta_token_ids = current_token_ids
delta_message = self.extract_tool_calls_streaming(
previous_text=state.previous_text,
current_text=current_text,
delta_text=delta_text,
previous_token_ids=state.previous_token_ids,
current_token_ids=current_token_ids,
delta_token_ids=delta_token_ids,
request=request, # type: ignore[arg-type]
)
elif self._reasoning_parser:
# Reasoning extraction
if self._in_reasoning_phase(state):
delta_message = self.extract_reasoning_streaming(
previous_text=state.previous_text,
current_text=current_text,
@@ -633,8 +610,24 @@ class DelegatingParser(Parser):
current_token_ids=current_token_ids,
delta_token_ids=delta_token_ids,
)
# Hand off remaining content to tool parser
if self._tool_parser and self.is_reasoning_end(delta_token_ids):
state.reasoning_ended = True
current_token_ids = self.extract_content_ids(delta_token_ids)
if delta_message and delta_message.content:
current_text = delta_message.content
delta_message.content = None
else:
current_text = ""
elif self._tool_parser:
# Tool call extraction
if self._in_tool_call_phase(state):
if not state.tool_call_text_started:
state.tool_call_text_started = True
state.previous_text = ""
state.previous_token_ids = []
delta_text = current_text
delta_token_ids = current_token_ids
delta_message = self.extract_tool_calls_streaming(
previous_text=state.previous_text,
current_text=current_text,
@@ -645,7 +638,8 @@ class DelegatingParser(Parser):
request=request, # type: ignore[arg-type]
)
else:
# No parsers: pass through as content
if self._reasoning_parser is None and self._tool_parser is None:
delta_message = DeltaMessage(content=delta_text)
state.previous_text = current_text