[mypy] Misc. typing improvements (#7417)
This commit is contained in:
@@ -1,10 +1,9 @@
|
||||
from typing import Dict, List, Optional, Tuple, Union
|
||||
|
||||
from transformers import PreTrainedTokenizer, PreTrainedTokenizerFast
|
||||
from typing import Dict, List, Optional, Tuple
|
||||
|
||||
from vllm.sequence import Logprob, SamplingParams, Sequence, SequenceGroup
|
||||
from vllm.transformers_utils.tokenizer_group.base_tokenizer_group import (
|
||||
BaseTokenizerGroup)
|
||||
|
||||
from .tokenizer import AnyTokenizer
|
||||
from .tokenizer_group import BaseTokenizerGroup
|
||||
|
||||
# Used eg. for marking rejected tokens in spec decoding.
|
||||
INVALID_TOKEN_ID = -1
|
||||
@@ -16,8 +15,7 @@ class Detokenizer:
|
||||
def __init__(self, tokenizer_group: BaseTokenizerGroup):
|
||||
self.tokenizer_group = tokenizer_group
|
||||
|
||||
def get_tokenizer_for_seq(self,
|
||||
sequence: Sequence) -> "PreTrainedTokenizer":
|
||||
def get_tokenizer_for_seq(self, sequence: Sequence) -> AnyTokenizer:
|
||||
"""Returns the HF tokenizer to use for a given sequence."""
|
||||
return self.tokenizer_group.get_lora_tokenizer(sequence.lora_request)
|
||||
|
||||
@@ -174,7 +172,7 @@ def _replace_none_with_empty(tokens: List[Optional[str]]):
|
||||
|
||||
|
||||
def _convert_tokens_to_string_with_added_encoders(
|
||||
tokenizer: Union[PreTrainedTokenizer, PreTrainedTokenizerFast],
|
||||
tokenizer: AnyTokenizer,
|
||||
output_tokens: List[str],
|
||||
skip_special_tokens: bool,
|
||||
spaces_between_special_tokens: bool,
|
||||
@@ -213,7 +211,7 @@ INITIAL_INCREMENTAL_DETOKENIZATION_OFFSET = 5
|
||||
|
||||
|
||||
def convert_prompt_ids_to_tokens(
|
||||
tokenizer: Union[PreTrainedTokenizer, PreTrainedTokenizerFast],
|
||||
tokenizer: AnyTokenizer,
|
||||
prompt_ids: List[int],
|
||||
skip_special_tokens: bool = False,
|
||||
) -> Tuple[List[str], int, int]:
|
||||
@@ -240,7 +238,7 @@ def convert_prompt_ids_to_tokens(
|
||||
# https://github.com/huggingface/text-generation-inference/blob/v0.9.4/server/text_generation_server/models/model.py#L62C9-L62C15
|
||||
# under Apache 2.0 license
|
||||
def detokenize_incrementally(
|
||||
tokenizer: Union[PreTrainedTokenizer, PreTrainedTokenizerFast],
|
||||
tokenizer: AnyTokenizer,
|
||||
all_input_ids: List[int],
|
||||
prev_tokens: Optional[List[str]],
|
||||
prefix_offset: int,
|
||||
|
||||
@@ -12,10 +12,10 @@ from vllm.lora.request import LoRARequest
|
||||
from vllm.transformers_utils.tokenizers import BaichuanTokenizer
|
||||
from vllm.utils import make_async
|
||||
|
||||
from .tokenizer_group import AnyTokenizer
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
AnyTokenizer = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
|
||||
|
||||
|
||||
def get_cached_tokenizer(tokenizer: AnyTokenizer) -> AnyTokenizer:
|
||||
"""Get tokenizer with cached properties.
|
||||
@@ -141,7 +141,7 @@ def get_tokenizer(
|
||||
|
||||
|
||||
def get_lora_tokenizer(lora_request: LoRARequest, *args,
|
||||
**kwargs) -> Optional[PreTrainedTokenizer]:
|
||||
**kwargs) -> Optional[AnyTokenizer]:
|
||||
if lora_request is None:
|
||||
return None
|
||||
try:
|
||||
|
||||
@@ -8,8 +8,7 @@ from .base_tokenizer_group import AnyTokenizer, BaseTokenizerGroup
|
||||
from .tokenizer_group import TokenizerGroup
|
||||
|
||||
if ray:
|
||||
from vllm.transformers_utils.tokenizer_group.ray_tokenizer_group import (
|
||||
RayTokenizerGroupPool)
|
||||
from .ray_tokenizer_group import RayTokenizerGroupPool
|
||||
else:
|
||||
RayTokenizerGroupPool = None # type: ignore
|
||||
|
||||
|
||||
@@ -1,12 +1,9 @@
|
||||
from abc import ABC, abstractmethod
|
||||
from typing import List, Optional, Union
|
||||
|
||||
from transformers import PreTrainedTokenizer, PreTrainedTokenizerFast
|
||||
from typing import List, Optional
|
||||
|
||||
from vllm.config import TokenizerPoolConfig
|
||||
from vllm.lora.request import LoRARequest
|
||||
|
||||
AnyTokenizer = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
|
||||
|
||||
class BaseTokenizerGroup(ABC):
|
||||
@@ -24,9 +21,10 @@ class BaseTokenizerGroup(ABC):
|
||||
pass
|
||||
|
||||
@abstractmethod
|
||||
def get_max_input_len(self,
|
||||
lora_request: Optional[LoRARequest] = None
|
||||
) -> Optional[int]:
|
||||
def get_max_input_len(
|
||||
self,
|
||||
lora_request: Optional[LoRARequest] = None,
|
||||
) -> Optional[int]:
|
||||
"""Get the maximum input length for the LoRA request."""
|
||||
pass
|
||||
|
||||
|
||||
@@ -13,8 +13,9 @@ from vllm.config import TokenizerPoolConfig
|
||||
from vllm.executor.ray_utils import ray
|
||||
from vllm.logger import init_logger
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.transformers_utils.tokenizer import AnyTokenizer
|
||||
|
||||
from .base_tokenizer_group import AnyTokenizer, BaseTokenizerGroup
|
||||
from .base_tokenizer_group import BaseTokenizerGroup
|
||||
from .tokenizer_group import TokenizerGroup
|
||||
|
||||
logger = init_logger(__name__)
|
||||
|
||||
@@ -2,12 +2,13 @@ from typing import List, Optional
|
||||
|
||||
from vllm.config import TokenizerPoolConfig
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.transformers_utils.tokenizer import (get_lora_tokenizer,
|
||||
from vllm.transformers_utils.tokenizer import (AnyTokenizer,
|
||||
get_lora_tokenizer,
|
||||
get_lora_tokenizer_async,
|
||||
get_tokenizer)
|
||||
from vllm.utils import LRUCache
|
||||
|
||||
from .base_tokenizer_group import AnyTokenizer, BaseTokenizerGroup
|
||||
from .base_tokenizer_group import BaseTokenizerGroup
|
||||
|
||||
|
||||
class TokenizerGroup(BaseTokenizerGroup):
|
||||
|
||||
Reference in New Issue
Block a user