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

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 18:27:28 -07:00
committed by GitHub
co-authored by Joe Runde Joe Runde Nick Hill Simon Mo
parent 708989341e
commit ed812a73fa
20 changed files with 1567 additions and 101 deletions
+7 -6
View File
@@ -6,7 +6,7 @@ import numpy as np
from fastapi import Request
from vllm.config import ModelConfig
from vllm.engine.async_llm_engine import AsyncLLMEngine
from vllm.engine.protocol import AsyncEngineClient
from vllm.entrypoints.logger import RequestLogger
from vllm.entrypoints.openai.protocol import (EmbeddingRequest,
EmbeddingResponse,
@@ -56,13 +56,13 @@ class OpenAIServingEmbedding(OpenAIServing):
def __init__(
self,
engine: AsyncLLMEngine,
async_engine_client: AsyncEngineClient,
model_config: ModelConfig,
served_model_names: List[str],
*,
request_logger: Optional[RequestLogger],
):
super().__init__(engine=engine,
super().__init__(async_engine_client=async_engine_client,
model_config=model_config,
served_model_names=served_model_names,
lora_modules=None,
@@ -99,7 +99,8 @@ class OpenAIServingEmbedding(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)
pooling_params = request.to_pooling_params()
@@ -124,7 +125,7 @@ class OpenAIServingEmbedding(OpenAIServing):
"Prompt adapter is not supported "
"for embedding models")
generator = self.engine.encode(
generator = self.async_engine_client.encode(
{"prompt_token_ids": prompt_inputs["prompt_token_ids"]},
pooling_params,
request_id_item,
@@ -146,7 +147,7 @@ class OpenAIServingEmbedding(OpenAIServing):
async for i, res in result_generator:
if await raw_request.is_disconnected():
# Abort the request if the client disconnects.
await self.engine.abort(f"{request_id}-{i}")
await self.async_engine_client.abort(f"{request_id}-{i}")
return self.create_error_response("Client disconnected")
final_res_batch[i] = res