[mypy] Enable following imports for entrypoints (#7248)

Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Co-authored-by: Fei <dfdfcai4@gmail.com>

vllm/engine/output_processor/interfaces.py

@@ -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.

vllm/engine/output_processor/multi_step.py

@@ -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

vllm/engine/output_processor/stop_checker.py

@@ -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