Skip to content

[serve.llm][Fix] retry with model_config arg #52991

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
May 15, 2025
Merged

[serve.llm][Fix] retry with model_config arg #52991

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9fd1cb0
retry with model_config arg
lk-chen May 14, 2025
b68a36b
use helper func
lk-chen May 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 30 additions & 4 deletions python/ray/llm/_internal/serve/deployments/llm/vllm/vllm_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -234,7 +234,9 @@ async def start(self):

If the engine is already running, do nothing.
"""
from vllm.entrypoints.chat_utils import resolve_chat_template_content_format
from vllm.entrypoints.chat_utils import (
resolve_chat_template_content_format as _resolve_chat_template_content_format,
)

if self.running:
# The engine is already running!
Expand All @@ -246,7 +248,21 @@ async def start(self):
self.model_config = await self.engine.get_model_config()

self._tokenizer = await self.engine.get_tokenizer()

def resolve_chat_template_content_format(model_config, **kwargs):
try:
return _resolve_chat_template_content_format(
model_config=model_config, **kwargs
)
except TypeError:
# Legacy API before vLLM 0.9.0.
# TODO(#52975): Remove this try-except once vLLM <0.9.0 is no longer supported.
return _resolve_chat_template_content_format(
trust_remote_code=model_config.trust_remote_code, **kwargs
)

self._resolved_content_format = resolve_chat_template_content_format(
model_config=self.model_config,
# Use HF to get the chat template so set it to None here.
chat_template=None,
# Default to None, change when it's needed.
Expand All @@ -255,7 +271,6 @@ async def start(self):
# Let vLLM decide the content format.
given_format="auto",
tokenizer=self._tokenizer,
trust_remote_code=self.model_config.trust_remote_code,
)

logger.info("Started vLLM engine.")
Expand Down Expand Up @@ -488,7 +503,7 @@ async def prepare_request(
) -> GenerationRequest:
from vllm.entrypoints.chat_utils import (
parse_chat_messages_futures,
apply_hf_chat_template,
apply_hf_chat_template as _apply_hf_chat_template,
)

model_config = self.model_config
Expand All @@ -504,14 +519,25 @@ async def prepare_request(
)
mm_data = await mm_futures

def apply_hf_chat_template(model_config, **kwargs):
try:
return _apply_hf_chat_template(model_config=model_config, **kwargs)
except TypeError:
# Legacy API before vLLM 0.9.0.
# TODO(#52975): Remove above once vLLM <0.9.0 is no longer supported.
return _apply_hf_chat_template(
trust_remote_code=model_config.trust_remote_code, **kwargs
)

prompt_text = apply_hf_chat_template(
model_config=model_config,
tokenizer=self._tokenizer,
conversation=conversation,
chat_template=None,
tools=None,
trust_remote_code=model_config.trust_remote_code,
tokenize=False,
# **kwargs for tokenizer.apply_chat_template
trust_remote_code=model_config.trust_remote_code,
add_generation_prompt=True,
continue_final_message=False,
)
Expand Down