[Frontend] Fix Kimi K2 tool call IDs for required tool choice (#46344)

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
This commit is contained in:
Chauncey
2026-06-24 19:59:40 +00:00
committed by GitHub
parent 49f2104c53
commit 84c2f9f0fb
11 changed files with 406 additions and 178 deletions
+11 -2
View File
@@ -13,7 +13,7 @@ from typing import TYPE_CHECKING
import regex as re
from vllm.entrypoints.chat_utils import make_tool_call_id
from vllm.entrypoints.chat_utils import get_tool_call_id_type, make_tool_call_id
from vllm.entrypoints.openai.engine.protocol import (
DeltaFunctionCall,
DeltaMessage,
@@ -89,11 +89,18 @@ class ParserEngine(Parser):
tools: list[Tool] | None = None,
*,
parser_engine_config: ParserEngineConfig,
model_config=None,
**kwargs,
) -> None:
self.model_tokenizer = tokenizer
self._tools = tools
self._stream_state = StreamState()
self._stream_state = StreamState(
tool_call_id_type=(
get_tool_call_id_type(model_config)
if model_config is not None
else "random"
),
)
self._reasoning_parser = None
self._tool_parser = None
self.parser_engine_config = parser_engine_config
@@ -419,6 +426,7 @@ class ParserEngine(Parser):
*,
finished: bool,
) -> DeltaMessage | None:
self._initialize_history_tool_call_cnt(request)
if not self._prompt_streaming_prepared and prompt_token_ids is not None:
# NOTE: call the hook BEFORE setting the flag, because the hook
# may invoke ``_reset`` (e.g. via ``initialize_streaming``) which
@@ -658,6 +666,7 @@ class ParserEngine(Parser):
enable_auto_tools: bool = False,
model_output_token_ids: Sequence[int] = (),
) -> tuple[str | None, str | None, list[FunctionCall] | None]:
self._initialize_history_tool_call_cnt(request)
self._check_skip_tool_parsing(request)
reasoning, content, tool_call_info = self._single_pass_parse(
model_output,