Re-enable the 80 char line width limit (#3305)
This commit is contained in:
@@ -1,7 +1,9 @@
|
||||
"""
|
||||
NOTE: This API server is used only for demonstrating usage of AsyncEngine and simple performance benchmarks.
|
||||
It is not intended for production use. For production use, we recommend using our OpenAI compatible server.
|
||||
We are also not going to accept PRs modifying this file, please change `vllm/entrypoints/openai/api_server.py` instead.
|
||||
NOTE: This API server is used only for demonstrating usage of AsyncEngine
|
||||
and simple performance benchmarks. It is not intended for production use.
|
||||
For production use, we recommend using our OpenAI compatible server.
|
||||
We are also not going to accept PRs modifying this file, please
|
||||
change `vllm/entrypoints/openai/api_server.py` instead.
|
||||
"""
|
||||
|
||||
import argparse
|
||||
|
||||
@@ -18,7 +18,9 @@ from fastapi.responses import JSONResponse, StreamingResponse, Response
|
||||
import vllm
|
||||
from vllm.engine.arg_utils import AsyncEngineArgs
|
||||
from vllm.engine.async_llm_engine import AsyncLLMEngine
|
||||
from vllm.entrypoints.openai.protocol import CompletionRequest, ChatCompletionRequest, ErrorResponse
|
||||
from vllm.entrypoints.openai.protocol import (CompletionRequest,
|
||||
ChatCompletionRequest,
|
||||
ErrorResponse)
|
||||
from vllm.logger import init_logger
|
||||
from vllm.entrypoints.openai.serving_chat import OpenAIServingChat
|
||||
from vllm.entrypoints.openai.serving_completion import OpenAIServingCompletion
|
||||
@@ -84,13 +86,11 @@ def parse_args():
|
||||
type=json.loads,
|
||||
default=["*"],
|
||||
help="allowed headers")
|
||||
parser.add_argument(
|
||||
"--api-key",
|
||||
type=str,
|
||||
default=None,
|
||||
help=
|
||||
"If provided, the server will require this key to be presented in the header."
|
||||
)
|
||||
parser.add_argument("--api-key",
|
||||
type=str,
|
||||
default=None,
|
||||
help="If provided, the server will require this key "
|
||||
"to be presented in the header.")
|
||||
parser.add_argument("--served-model-name",
|
||||
type=str,
|
||||
default=None,
|
||||
@@ -103,9 +103,8 @@ def parse_args():
|
||||
default=None,
|
||||
nargs='+',
|
||||
action=LoRAParserAction,
|
||||
help=
|
||||
"LoRA module configurations in the format name=path. Multiple modules can be specified."
|
||||
)
|
||||
help="LoRA module configurations in the format name=path. "
|
||||
"Multiple modules can be specified.")
|
||||
parser.add_argument("--chat-template",
|
||||
type=str,
|
||||
default=None,
|
||||
@@ -138,9 +137,10 @@ def parse_args():
|
||||
help="Additional ASGI middleware to apply to the app. "
|
||||
"We accept multiple --middleware arguments. "
|
||||
"The value should be an import path. "
|
||||
"If a function is provided, vLLM will add it to the server using @app.middleware('http'). "
|
||||
"If a class is provided, vLLM will add it to the server using app.add_middleware(). "
|
||||
)
|
||||
"If a function is provided, vLLM will add it to the server "
|
||||
"using @app.middleware('http'). "
|
||||
"If a class is provided, vLLM will add it to the server "
|
||||
"using app.add_middleware(). ")
|
||||
|
||||
parser = AsyncEngineArgs.add_cli_args(parser)
|
||||
return parser.parse_args()
|
||||
@@ -235,9 +235,8 @@ if __name__ == "__main__":
|
||||
elif inspect.iscoroutinefunction(imported):
|
||||
app.middleware("http")(imported)
|
||||
else:
|
||||
raise ValueError(
|
||||
f"Invalid middleware {middleware}. Must be a function or a class."
|
||||
)
|
||||
raise ValueError(f"Invalid middleware {middleware}. "
|
||||
f"Must be a function or a class.")
|
||||
|
||||
logger.info(f"vLLM API server version {vllm.__version__}")
|
||||
logger.info(f"args: {args}")
|
||||
|
||||
@@ -12,7 +12,8 @@ from vllm.entrypoints.openai.protocol import (
|
||||
UsageInfo)
|
||||
from vllm.outputs import RequestOutput
|
||||
from vllm.entrypoints.openai.serving_engine import OpenAIServing, LoRA
|
||||
from vllm.model_executor.guided_decoding import get_guided_decoding_logits_processor
|
||||
from vllm.model_executor.guided_decoding import (
|
||||
get_guided_decoding_logits_processor)
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
@@ -37,8 +38,9 @@ class OpenAIServingChat(OpenAIServing):
|
||||
ChatCompletionResponse]:
|
||||
"""Completion API similar to OpenAI's API.
|
||||
|
||||
See https://platform.openai.com/docs/api-reference/chat/create
|
||||
for the API specification. This API mimics the OpenAI ChatCompletion API.
|
||||
See https://platform.openai.com/docs/api-reference/chat/create
|
||||
for the API specification. This API mimics the OpenAI
|
||||
ChatCompletion API.
|
||||
|
||||
NOTE: Currently we do not support the following feature:
|
||||
- function_call (Users should implement this by themselves)
|
||||
@@ -116,7 +118,8 @@ class OpenAIServingChat(OpenAIServing):
|
||||
# the result_generator, it needs to be sent as the FIRST
|
||||
# response (by the try...catch).
|
||||
if first_iteration:
|
||||
# Send first response for each request.n (index) with the role
|
||||
# Send first response for each request.n (index) with
|
||||
# the role
|
||||
role = self.get_chat_request_role(request)
|
||||
for i in range(request.n):
|
||||
choice_data = ChatCompletionResponseStreamChoice(
|
||||
@@ -133,7 +136,8 @@ class OpenAIServingChat(OpenAIServing):
|
||||
data = chunk.model_dump_json(exclude_unset=True)
|
||||
yield f"data: {data}\n\n"
|
||||
|
||||
# Send response to echo the input portion of the last message
|
||||
# Send response to echo the input portion of the
|
||||
# last message
|
||||
if request.echo:
|
||||
last_msg_content = ""
|
||||
if request.messages and isinstance(
|
||||
@@ -145,11 +149,12 @@ class OpenAIServingChat(OpenAIServing):
|
||||
|
||||
if last_msg_content:
|
||||
for i in range(request.n):
|
||||
choice_data = ChatCompletionResponseStreamChoice(
|
||||
index=i,
|
||||
delta=DeltaMessage(
|
||||
content=last_msg_content),
|
||||
finish_reason=None)
|
||||
choice_data = (
|
||||
ChatCompletionResponseStreamChoice(
|
||||
index=i,
|
||||
delta=DeltaMessage(
|
||||
content=last_msg_content),
|
||||
finish_reason=None))
|
||||
chunk = ChatCompletionStreamResponse(
|
||||
id=request_id,
|
||||
object=chunk_object_type,
|
||||
|
||||
@@ -1,7 +1,8 @@
|
||||
import asyncio
|
||||
import time
|
||||
from fastapi import Request
|
||||
from typing import AsyncGenerator, AsyncIterator, Callable, List, Optional, Dict, Tuple
|
||||
from typing import (AsyncGenerator, AsyncIterator, Callable, List, Optional,
|
||||
Dict, Tuple)
|
||||
from vllm.logger import init_logger
|
||||
from vllm.utils import random_uuid
|
||||
from vllm.engine.async_llm_engine import AsyncLLMEngine
|
||||
@@ -16,7 +17,8 @@ from vllm.entrypoints.openai.protocol import (
|
||||
)
|
||||
from vllm.outputs import RequestOutput
|
||||
from vllm.entrypoints.openai.serving_engine import OpenAIServing, LoRA
|
||||
from vllm.model_executor.guided_decoding import get_guided_decoding_logits_processor
|
||||
from vllm.model_executor.guided_decoding import (
|
||||
get_guided_decoding_logits_processor)
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
@@ -44,9 +46,8 @@ def parse_prompt_format(prompt) -> Tuple[bool, list]:
|
||||
prompt_is_tokens = True
|
||||
prompts = prompt # case 4: array of token arrays
|
||||
else:
|
||||
raise ValueError(
|
||||
"prompt must be a string, array of strings, array of tokens, or array of token arrays"
|
||||
)
|
||||
raise ValueError("prompt must be a string, array of strings, "
|
||||
"array of tokens, or array of token arrays")
|
||||
return prompt_is_tokens, prompts
|
||||
|
||||
|
||||
@@ -156,7 +157,8 @@ class OpenAIServingCompletion(OpenAIServing):
|
||||
int, RequestOutput]] = merge_async_iterators(*generators)
|
||||
|
||||
# Similar to the OpenAI API, when n != best_of, we do not stream the
|
||||
# results. In addition, we do not stream the results when use beam search.
|
||||
# results. In addition, we do not stream the results when use
|
||||
# beam search.
|
||||
stream = (request.stream
|
||||
and (request.best_of is None or request.n == request.best_of)
|
||||
and not request.use_beam_search)
|
||||
@@ -223,7 +225,8 @@ class OpenAIServingCompletion(OpenAIServing):
|
||||
|
||||
for output in res.outputs:
|
||||
i = output.index + prompt_idx * request.n
|
||||
# TODO(simon): optimize the performance by avoiding full text O(n^2) sending.
|
||||
# TODO(simon): optimize the performance by avoiding full
|
||||
# text O(n^2) sending.
|
||||
|
||||
if request.echo and request.max_tokens == 0:
|
||||
# only return the prompt
|
||||
@@ -231,11 +234,12 @@ class OpenAIServingCompletion(OpenAIServing):
|
||||
delta_token_ids = res.prompt_token_ids
|
||||
top_logprobs = res.prompt_logprobs
|
||||
has_echoed[i] = True
|
||||
elif request.echo and request.max_tokens > 0 and not has_echoed[
|
||||
i]:
|
||||
elif (request.echo and request.max_tokens > 0
|
||||
and not has_echoed[i]):
|
||||
# echo the prompt and first token
|
||||
delta_text = res.prompt + output.text
|
||||
delta_token_ids = res.prompt_token_ids + output.token_ids
|
||||
delta_token_ids = (res.prompt_token_ids +
|
||||
output.token_ids)
|
||||
top_logprobs = res.prompt_logprobs + (output.logprobs
|
||||
or [])
|
||||
has_echoed[i] = True
|
||||
@@ -248,7 +252,9 @@ class OpenAIServingCompletion(OpenAIServing):
|
||||
i]:] if output.logprobs else None
|
||||
|
||||
if request.logprobs is not None:
|
||||
assert top_logprobs is not None, "top_logprobs must be provided when logprobs is requested"
|
||||
assert top_logprobs is not None, (
|
||||
"top_logprobs must be provided when logprobs "
|
||||
"is requested")
|
||||
logprobs = self._create_logprobs(
|
||||
token_ids=delta_token_ids,
|
||||
top_logprobs=top_logprobs,
|
||||
|
||||
@@ -50,10 +50,12 @@ class OpenAIServing:
|
||||
except RuntimeError:
|
||||
event_loop = None
|
||||
|
||||
if event_loop is not None and event_loop.is_running(
|
||||
): # If the current is instanced by Ray Serve, there is already a running event loop
|
||||
if event_loop is not None and event_loop.is_running():
|
||||
# If the current is instanced by Ray Serve,
|
||||
# there is already a running event loop
|
||||
event_loop.create_task(self._post_init())
|
||||
else: # When using single vLLM without engine_use_ray
|
||||
else:
|
||||
# When using single vLLM without engine_use_ray
|
||||
asyncio.run(self._post_init())
|
||||
|
||||
async def _post_init(self):
|
||||
@@ -178,8 +180,9 @@ class OpenAIServing:
|
||||
|
||||
if token_num + request.max_tokens > self.max_model_len:
|
||||
raise ValueError(
|
||||
f"This model's maximum context length is {self.max_model_len} tokens. "
|
||||
f"However, you requested {request.max_tokens + token_num} tokens "
|
||||
f"This model's maximum context length is "
|
||||
f"{self.max_model_len} tokens. However, you requested "
|
||||
f"{request.max_tokens + token_num} tokens "
|
||||
f"({token_num} in the messages, "
|
||||
f"{request.max_tokens} in the completion). "
|
||||
f"Please reduce the length of the messages or completion.", )
|
||||
|
||||
Reference in New Issue
Block a user