biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core][Bugfix][Perf] Introduce MQLLMEngine to avoid asyncio OH (#8157)

Browse Source 
Co-authored-by: Nick Hill <nickhill@us.ibm.com>
Co-authored-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-neuralmagic@users.noreply.github.com>
Co-authored-by: Simon Mo <simon.mo@hey.com>
This commit is contained in:
Alexander Matveev
2024-09-18 09:56:58 -04:00
committed by GitHub
parent 9d104b5beb
commit 7c7714d856
36 changed files with 1464 additions and 1169 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
vllm/entrypoints/openai/serving_completion.py
View File

@@ -8,7 +8,7 @@ from typing import Tuple, Union, cast
from fastapi import Request
from vllm.config import ModelConfig
from vllm.engine.protocol import AsyncEngineClient
from vllm.engine.protocol import EngineClient
from vllm.entrypoints.logger import RequestLogger
# yapf conflicts with isort for this block
# yapf: disable
@@ -43,7 +43,7 @@ class OpenAIServingCompletion(OpenAIServing):
def __init__(
self,
async_engine_client: AsyncEngineClient,
engine_client: EngineClient,
model_config: ModelConfig,
served_model_names: List[str],
*,
@@ -52,7 +52,7 @@ class OpenAIServingCompletion(OpenAIServing):
request_logger: Optional[RequestLogger],
return_tokens_as_token_ids: bool = False,
):
super().__init__(async_engine_client=async_engine_client,
super().__init__(engine_client=engine_client,
model_config=model_config,
served_model_names=served_model_names,
lora_modules=lora_modules,
@@ -78,6 +78,12 @@ class OpenAIServingCompletion(OpenAIServing):
if error_check_ret is not None:
return error_check_ret
# If the engine is dead, raise the engine's DEAD_ERROR.
# This is required for the streaming case, where we return a
# success status before we actually start generating text :).
if self.engine_client.errored:
raise self.engine_client.dead_error
# Return error for unsupported features.
if request.suffix is not None:
return self.create_error_response(
@@ -95,8 +101,7 @@ class OpenAIServingCompletion(OpenAIServing):
prompt_adapter_request,
) = self._maybe_get_adapters(request)
tokenizer = await self.async_engine_client.get_tokenizer(
lora_request)
tokenizer = await self.engine_client.get_tokenizer(lora_request)
guided_decode_logits_processor = (
await self._guided_decode_logits_processor(request, tokenizer))
@@ -124,8 +129,8 @@ class OpenAIServingCompletion(OpenAIServing):
lora_request=lora_request,
prompt_adapter_request=prompt_adapter_request)
is_tracing_enabled = (
await self.async_engine_client.is_tracing_enabled())
is_tracing_enabled = (await
self.engine_client.is_tracing_enabled())
trace_headers = None
if is_tracing_enabled:
trace_headers = extract_trace_headers(raw_request.headers)
@@ -133,7 +138,7 @@ class OpenAIServingCompletion(OpenAIServing):
raw_request.headers):
log_tracing_disabled_warning()
generator = self.async_engine_client.generate(
generator = self.engine_client.generate(
{"prompt_token_ids": prompt_inputs["prompt_token_ids"]},
sampling_params,
request_id_item,