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

[ Frontend ] Multiprocessing for OpenAI Server with zeromq (#6883)

Browse Source 
Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>
Co-authored-by: Joe Runde <Joseph.Runde@ibm.com>
Co-authored-by: Joe Runde <joe@joerun.de>
Co-authored-by: Nick Hill <nickhill@us.ibm.com>
Co-authored-by: Simon Mo <simon.mo@hey.com>
This commit is contained in:
Robert Shaw
2024-08-02 21:27:28 -04:00
committed by GitHub
parent 708989341e
commit ed812a73fa
20 changed files with 1567 additions and 101 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/entrypoints/openai/serving_tokenization.py
View File

@@ -1,9 +1,9 @@
from typing import List, Optional, Union
from vllm.config import ModelConfig
from vllm.engine.async_llm_engine import AsyncLLMEngine
# yapf conflicts with isort for this block
# yapf: disable
from vllm.engine.protocol import AsyncEngineClient
from vllm.entrypoints.chat_utils import (ConversationMessage,
load_chat_template,
parse_chat_message_content)
@@ -24,7 +24,7 @@ class OpenAIServingTokenization(OpenAIServing):
def __init__(
self,
engine: AsyncLLMEngine,
async_engine_client: AsyncEngineClient,
model_config: ModelConfig,
served_model_names: List[str],
*,
@@ -32,7 +32,7 @@ class OpenAIServingTokenization(OpenAIServing):
request_logger: Optional[RequestLogger],
chat_template: Optional[str],
):
super().__init__(engine=engine,
super().__init__(async_engine_client=async_engine_client,
model_config=model_config,
served_model_names=served_model_names,
lora_modules=lora_modules,
@@ -57,7 +57,7 @@ class OpenAIServingTokenization(OpenAIServing):
prompt_adapter_request,
) = self._maybe_get_adapters(request)
tokenizer = await self.engine.get_tokenizer(lora_request)
tokenizer = await self.async_engine_client.get_tokenizer(lora_request)
if isinstance(request, TokenizeChatRequest):
model_config = self.model_config
@@ -113,7 +113,7 @@ class OpenAIServingTokenization(OpenAIServing):
prompt_adapter_request,
) = self._maybe_get_adapters(request)
tokenizer = await self.engine.get_tokenizer(lora_request)
tokenizer = await self.async_engine_client.get_tokenizer(lora_request)
self._log_inputs(request_id,
request.tokens,