[Fronten] Clean up stop_token_ids override for Harmony (#44009)
Signed-off-by: Yifan Zong <yzong@redhat.com>
This commit is contained in:
@@ -46,7 +46,6 @@ from vllm.entrypoints.openai.engine.serving import (
|
||||
from vllm.entrypoints.openai.models.serving import OpenAIServingModels
|
||||
from vllm.entrypoints.openai.parser.harmony_utils import (
|
||||
get_developer_message,
|
||||
get_stop_tokens_for_assistant_actions,
|
||||
get_system_message,
|
||||
get_user_message,
|
||||
has_custom_tools,
|
||||
@@ -222,13 +221,6 @@ class OpenAIServingResponses(OpenAIServing):
|
||||
"For gpt-oss, we ignore --enable-auto-tool-choice "
|
||||
"and always enable tool use."
|
||||
)
|
||||
# OpenAI models have two EOS-like tokens: <|return|> and <|call|>.
|
||||
# We need to add them to the stop token ids.
|
||||
if "stop_token_ids" not in self.default_sampling_params:
|
||||
self.default_sampling_params["stop_token_ids"] = []
|
||||
self.default_sampling_params["stop_token_ids"].extend(
|
||||
get_stop_tokens_for_assistant_actions()
|
||||
)
|
||||
|
||||
self.tool_call_id_type = get_tool_call_id_type(self.model_config)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user