[mypy] Enable following imports for entrypoints (#7248)
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu> Co-authored-by: Fei <dfdfcai4@gmail.com>
This commit is contained in:
co-authored by
Woosuk Kwon
Fei
parent
4506641212
commit
baaedfdb2d
@@ -6,7 +6,6 @@ from typing import (AsyncGenerator, Callable, Dict, Iterable, List, Mapping,
|
||||
Optional, Set, Tuple, Type, Union)
|
||||
|
||||
import torch
|
||||
from transformers import PreTrainedTokenizer
|
||||
from typing_extensions import assert_never
|
||||
|
||||
import vllm.envs as envs
|
||||
@@ -31,6 +30,7 @@ from vllm.prompt_adapter.request import PromptAdapterRequest
|
||||
from vllm.sampling_params import SamplingParams
|
||||
from vllm.sequence import (ExecuteModelRequest, SamplerOutput,
|
||||
SequenceGroupMetadata)
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
from vllm.usage.usage_lib import UsageContext
|
||||
from vllm.utils import print_warning_once
|
||||
|
||||
@@ -427,8 +427,8 @@ class _AsyncLLMEngine(LLMEngine):
|
||||
lora_request: Optional[LoRARequest],
|
||||
) -> List[int]:
|
||||
"""Async version of :meth:`_tokenize_prompt`."""
|
||||
tokenizer = self.get_tokenizer_group("prompts must be None if "
|
||||
"skip_tokenizer_init is True")
|
||||
tokenizer = self.get_tokenizer_group(
|
||||
missing_msg="prompts must be None if skip_tokenizer_init is True")
|
||||
|
||||
return await tokenizer.encode_async(request_id=request_id,
|
||||
prompt=prompt,
|
||||
@@ -771,7 +771,7 @@ class AsyncLLMEngine:
|
||||
async def get_tokenizer(
|
||||
self,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
) -> "PreTrainedTokenizer":
|
||||
) -> AnyTokenizer:
|
||||
if self.engine_use_ray:
|
||||
return await self.engine.get_tokenizer.remote( # type: ignore
|
||||
lora_request)
|
||||
|
||||
+21
-10
@@ -3,9 +3,9 @@ from contextlib import contextmanager
|
||||
from typing import (TYPE_CHECKING, Any, ClassVar, Dict, Iterable, List,
|
||||
Mapping, Optional)
|
||||
from typing import Sequence as GenericSequence
|
||||
from typing import Set, Tuple, Type, TypeVar, Union
|
||||
from typing import Set, Tuple, Type, Union
|
||||
|
||||
from typing_extensions import assert_never
|
||||
from typing_extensions import TypeVar, assert_never
|
||||
|
||||
import vllm.envs as envs
|
||||
from vllm.config import (CacheConfig, DecodingConfig, DeviceConfig,
|
||||
@@ -43,8 +43,9 @@ from vllm.tracing import (SpanAttributes, SpanKind, extract_trace_context,
|
||||
init_tracer)
|
||||
from vllm.transformers_utils.config import try_get_generation_config
|
||||
from vllm.transformers_utils.detokenizer import Detokenizer
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
from vllm.transformers_utils.tokenizer_group import (
|
||||
AnyTokenizer, BaseTokenizerGroup, init_tokenizer_from_configs)
|
||||
BaseTokenizerGroup, init_tokenizer_from_configs)
|
||||
from vllm.usage.usage_lib import (UsageContext, is_usage_stats_enabled,
|
||||
usage_message)
|
||||
from vllm.utils import Counter, Device
|
||||
@@ -67,6 +68,7 @@ def _load_generation_config_dict(model_config: ModelConfig) -> Dict[str, Any]:
|
||||
return config.to_diff_dict()
|
||||
|
||||
|
||||
_G = TypeVar("_G", bound=BaseTokenizerGroup, default=BaseTokenizerGroup)
|
||||
_O = TypeVar("_O", RequestOutput, EmbeddingRequestOutput)
|
||||
|
||||
PromptComponents = Tuple[Optional[str], List[int],
|
||||
@@ -493,12 +495,21 @@ class LLMEngine:
|
||||
"skip_tokenizer_init is True")
|
||||
|
||||
def get_tokenizer_group(
|
||||
self,
|
||||
fail_msg: str = MISSING_TOKENIZER_GROUP_MSG) -> BaseTokenizerGroup:
|
||||
if self.tokenizer is None:
|
||||
raise ValueError(fail_msg)
|
||||
self,
|
||||
group_type: Type[_G] = BaseTokenizerGroup,
|
||||
*,
|
||||
missing_msg: str = MISSING_TOKENIZER_GROUP_MSG,
|
||||
) -> _G:
|
||||
tokenizer_group = self.tokenizer
|
||||
|
||||
return self.tokenizer
|
||||
if tokenizer_group is None:
|
||||
raise ValueError(missing_msg)
|
||||
if not isinstance(tokenizer_group, group_type):
|
||||
raise TypeError("Invalid type of tokenizer group. "
|
||||
f"Expected type: {group_type}, but "
|
||||
f"found type: {type(tokenizer_group)}")
|
||||
|
||||
return tokenizer_group
|
||||
|
||||
def get_tokenizer(
|
||||
self,
|
||||
@@ -693,8 +704,8 @@ class LLMEngine:
|
||||
* prompt token ids
|
||||
'''
|
||||
|
||||
tokenizer = self.get_tokenizer_group("prompts must be None if "
|
||||
"skip_tokenizer_init is True")
|
||||
tokenizer = self.get_tokenizer_group(
|
||||
missing_msg="prompts must be None if skip_tokenizer_init is True")
|
||||
|
||||
return tokenizer.encode(request_id=request_id,
|
||||
prompt=prompt,
|
||||
|
||||
@@ -1,13 +1,12 @@
|
||||
from abc import ABC, abstractmethod
|
||||
from typing import Callable, List
|
||||
|
||||
from transformers import PreTrainedTokenizer
|
||||
|
||||
from vllm.config import SchedulerConfig
|
||||
from vllm.core.scheduler import Scheduler
|
||||
from vllm.engine.output_processor.stop_checker import StopChecker
|
||||
from vllm.sequence import Sequence, SequenceGroup, SequenceGroupOutput
|
||||
from vllm.transformers_utils.detokenizer import Detokenizer
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
from vllm.utils import Counter
|
||||
|
||||
|
||||
@@ -29,7 +28,7 @@ class SequenceGroupOutputProcessor(ABC):
|
||||
detokenizer: Detokenizer,
|
||||
scheduler: List[Scheduler],
|
||||
seq_counter: Counter,
|
||||
get_tokenizer_for_seq: Callable[[Sequence], PreTrainedTokenizer],
|
||||
get_tokenizer_for_seq: Callable[[Sequence], AnyTokenizer],
|
||||
stop_checker: "StopChecker",
|
||||
):
|
||||
"""Create an output processor.
|
||||
|
||||
@@ -1,8 +1,6 @@
|
||||
import functools
|
||||
from typing import Callable, List
|
||||
|
||||
from transformers import PreTrainedTokenizer
|
||||
|
||||
from vllm.core.scheduler import Scheduler
|
||||
from vllm.engine.output_processor.interfaces import (
|
||||
SequenceGroupOutputProcessor)
|
||||
@@ -12,6 +10,7 @@ from vllm.sampling_params import SamplingParams
|
||||
from vllm.sequence import (Sequence, SequenceGroup, SequenceGroupOutput,
|
||||
SequenceOutput, SequenceStatus)
|
||||
from vllm.transformers_utils.detokenizer import Detokenizer
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
from vllm.utils import Counter
|
||||
|
||||
logger = init_logger(__name__)
|
||||
@@ -36,7 +35,7 @@ class MultiStepOutputProcessor(SequenceGroupOutputProcessor):
|
||||
detokenizer: Detokenizer,
|
||||
scheduler: List[Scheduler],
|
||||
seq_counter: Counter,
|
||||
get_tokenizer_for_seq: Callable[[Sequence], PreTrainedTokenizer],
|
||||
get_tokenizer_for_seq: Callable[[Sequence], AnyTokenizer],
|
||||
stop_checker: StopChecker,
|
||||
):
|
||||
self.detokenizer = detokenizer
|
||||
|
||||
@@ -1,10 +1,9 @@
|
||||
from typing import Callable, Optional
|
||||
|
||||
from transformers import PreTrainedTokenizer
|
||||
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.sampling_params import SamplingParams
|
||||
from vllm.sequence import Sequence, SequenceStatus
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
|
||||
|
||||
class StopChecker:
|
||||
@@ -15,8 +14,7 @@ class StopChecker:
|
||||
"""
|
||||
|
||||
def __init__(self, max_model_len: int,
|
||||
get_tokenizer_for_seq: Callable[[Sequence],
|
||||
PreTrainedTokenizer]):
|
||||
get_tokenizer_for_seq: Callable[[Sequence], AnyTokenizer]):
|
||||
# Do not use it directly, but use `self._get_max_model_len`.
|
||||
self._max_model_len = max_model_len
|
||||
self.get_tokenizer_for_seq = get_tokenizer_for_seq
|
||||
|
||||
+11
-6
@@ -1,8 +1,6 @@
|
||||
from typing import (AsyncGenerator, List, Mapping, Optional, Protocol,
|
||||
runtime_checkable)
|
||||
|
||||
from transformers import PreTrainedTokenizer
|
||||
|
||||
from vllm.config import DecodingConfig, ModelConfig
|
||||
from vllm.core.scheduler import SchedulerOutputs
|
||||
from vllm.inputs.data import PromptInputs
|
||||
@@ -12,6 +10,7 @@ from vllm.pooling_params import PoolingParams
|
||||
from vllm.prompt_adapter.request import PromptAdapterRequest
|
||||
from vllm.sampling_params import SamplingParams
|
||||
from vllm.sequence import SamplerOutput
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
|
||||
|
||||
@runtime_checkable
|
||||
@@ -40,6 +39,7 @@ class AsyncEngineClient(Protocol):
|
||||
prompt_adapter_request: Optional[PromptAdapterRequest] = None
|
||||
) -> AsyncGenerator[RequestOutput, None]:
|
||||
"""Generates outputs for a request"""
|
||||
...
|
||||
|
||||
def encode(
|
||||
self,
|
||||
@@ -50,6 +50,7 @@ class AsyncEngineClient(Protocol):
|
||||
trace_headers: Optional[Mapping[str, str]] = None,
|
||||
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
|
||||
"""Generate outputs for a request from an embedding model."""
|
||||
...
|
||||
|
||||
async def abort(self, request_id: str) -> None:
|
||||
"""Abort a request.
|
||||
@@ -60,25 +61,29 @@ class AsyncEngineClient(Protocol):
|
||||
|
||||
async def get_model_config(self) -> ModelConfig:
|
||||
"""Get the model configuration of the vLLM engine."""
|
||||
...
|
||||
|
||||
async def get_decoding_config(self) -> DecodingConfig:
|
||||
...
|
||||
"""Get the decoding configuration of the vLLM engine."""
|
||||
|
||||
async def get_tokenizer(
|
||||
self,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
) -> PreTrainedTokenizer:
|
||||
"""Get the appropriate Tokenizer for the request"""
|
||||
) -> AnyTokenizer:
|
||||
"""Get the appropriate tokenizer for the request"""
|
||||
...
|
||||
|
||||
async def is_tracing_enabled(self) -> bool:
|
||||
pass
|
||||
...
|
||||
|
||||
async def do_log_stats(
|
||||
self,
|
||||
scheduler_outputs: Optional[SchedulerOutputs] = None,
|
||||
model_output: Optional[List[SamplerOutput]] = None,
|
||||
) -> None:
|
||||
pass
|
||||
...
|
||||
|
||||
async def check_health(self) -> None:
|
||||
"""Raise if unhealthy"""
|
||||
...
|
||||
|
||||
Reference in New Issue
Block a user