Fix flaky tool calls in the Kimi K2.5 model (#17914)

Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
This commit is contained in:
Xinyuan Tong
2026-01-28 20:58:16 -08:00
committed by GitHub
parent 3c34d2c3eb
commit 9409c43593
+1 -7
View File
@@ -27,7 +27,7 @@ import tempfile
from typing import Any, Callable, Dict, List, Literal, Optional, Union from typing import Any, Callable, Dict, List, Literal, Optional, Union
from sglang.srt.connector import ConnectorType from sglang.srt.connector import ConnectorType
from sglang.srt.environ import ToolStrictLevel, envs from sglang.srt.environ import envs
from sglang.srt.function_call.function_call_parser import FunctionCallParser from sglang.srt.function_call.function_call_parser import FunctionCallParser
from sglang.srt.layers.attention.fla.chunk_delta_h import CHUNK_SIZE as FLA_CHUNK_SIZE from sglang.srt.layers.attention.fla.chunk_delta_h import CHUNK_SIZE as FLA_CHUNK_SIZE
from sglang.srt.lora.lora_registry import LoRARef from sglang.srt.lora.lora_registry import LoRARef
@@ -2485,12 +2485,6 @@ class ServerArgs:
envs.SGLANG_ENABLE_DETERMINISTIC_INFERENCE.set( envs.SGLANG_ENABLE_DETERMINISTIC_INFERENCE.set(
"1" if self.enable_deterministic_inference else "0" "1" if self.enable_deterministic_inference else "0"
) )
# Set the highest strict level for Kimi K2 tool calls
if (
self.tool_call_parser == "kimi_k2"
and not envs.SGLANG_TOOL_STRICT_LEVEL.is_set()
):
envs.SGLANG_TOOL_STRICT_LEVEL.set(ToolStrictLevel.PARAMETER)
def _handle_cache_compatibility(self): def _handle_cache_compatibility(self):
if self.enable_hierarchical_cache and self.disable_radix_cache: if self.enable_hierarchical_cache and self.disable_radix_cache: