@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 3257d449fa
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 16, 05:45 UTC

[Misc] Remove deprecated names (#10817)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

Cyrus Leung committed Dec 3, 2024 at 14:52 UTC 3257d449fa0fd3e05aa20cc8c5fff79ad101984f
10 files changed +31 -102
vllm/engine/async_llm_engine.py
+6 -2
@@ -6,6 +6,8 @@ from typing import (Any, AsyncGenerator, Callable, Coroutine, Dict, Iterable,
6 List, Mapping, Optional, Set, Tuple, Type, Union, overload)
7 from weakref import ReferenceType
8
9 +from typing_extensions import deprecated
10 +
11 import vllm.envs as envs
12 from vllm.config import (DecodingConfig, LoRAConfig, ModelConfig,
13 ParallelConfig, SchedulerConfig, VllmConfig)
@@ -422,7 +424,8 @@ class _AsyncLLMEngine(LLMEngine):
424 return await (
425 self.get_tokenizer_group().get_lora_tokenizer_async(lora_request))
426
425 - @overload # DEPRECATED
427 + @overload
428 + @deprecated("'inputs' will be renamed to 'prompt")
429 async def add_request_async(
430 self,
431 request_id: str,
@@ -894,7 +897,8 @@ class AsyncLLMEngine(EngineClient):
897
898 # This method does not need to be async, but kept that way
899 # for backwards compatibility.
897 - @overload # DEPRECATED
900 + @overload
901 + @deprecated("'inputs' will be renamed to 'prompt")
902 def add_request(
903 self,
904 request_id: str,
vllm/engine/llm_engine.py
+3 -2
@@ -10,7 +10,7 @@ from typing import Sequence as GenericSequence
10 from typing import Set, Type, Union, cast, overload
11
12 import torch
13 -from typing_extensions import TypeVar
13 +from typing_extensions import TypeVar, deprecated
14
15 import vllm.envs as envs
16 from vllm.config import (DecodingConfig, LoRAConfig, ModelConfig,
@@ -719,7 +719,8 @@ class LLMEngine:
719 def stop_remote_worker_execution_loop(self) -> None:
720 self.model_executor.stop_remote_worker_execution_loop()
721
722 - @overload # DEPRECATED
722 + @overload
723 + @deprecated("'inputs' will be renamed to 'prompt")
724 def add_request(
725 self,
726 request_id: str,
vllm/engine/multiprocessing/__init__.py
+4 -1
@@ -2,6 +2,8 @@ from dataclasses import dataclass
2 from enum import Enum
3 from typing import List, Mapping, Optional, Union, overload
4
5 +from typing_extensions import deprecated
6 +
7 from vllm import PoolingParams
8 from vllm.inputs import PromptType
9 from vllm.lora.request import LoRARequest
@@ -32,7 +34,8 @@ class RPCProcessRequest:
34 prompt_adapter_request: Optional[PromptAdapterRequest] = None
35 priority: int = 0
36
35 - @overload # DEPRECATED
37 + @overload
38 + @deprecated("'inputs' will be renamed to 'prompt")
39 def __init__(
40 self,
41 *,
vllm/engine/multiprocessing/client.py
+5 -2
@@ -9,6 +9,7 @@ import cloudpickle
9 import psutil
10 import zmq
11 import zmq.asyncio
12 +from typing_extensions import deprecated
13 from zmq import Frame # type: ignore[attr-defined]
14 from zmq.asyncio import Socket
15
@@ -414,7 +415,8 @@ class MQLLMEngineClient(EngineClient):
415 def dead_error(self) -> BaseException:
416 return ENGINE_DEAD_ERROR(self._errored_with)
417
417 - @overload # DEPRECATED
418 + @overload
419 + @deprecated("'inputs' will be renamed to 'prompt")
420 def generate(
421 self,
422 *,
@@ -485,7 +487,8 @@ class MQLLMEngineClient(EngineClient):
487 lora_request, trace_headers,
488 prompt_adapter_request, priority)
489
488 - @overload # DEPRECATED
490 + @overload
491 + @deprecated("'inputs' will be renamed to 'prompt")
492 def encode(
493 self,
494 *,
vllm/entrypoints/llm.py
+11
@@ -6,6 +6,7 @@ from typing import (Any, ClassVar, Dict, List, Optional, Sequence, Tuple, Type,
6 Union, cast, overload)
7
8 from tqdm import tqdm
9 +from typing_extensions import deprecated
10
11 from vllm import envs
12 from vllm.beam_search import (BeamSearchInstance, BeamSearchOutput,
@@ -256,6 +257,7 @@ class LLM:
257 tokenizer_group.tokenizer = get_cached_tokenizer(tokenizer)
258
259 @overload # LEGACY: single (prompt + optional token ids)
260 + @deprecated("'prompt_token_ids' will become part of 'prompts")
261 def generate(
262 self,
263 prompts: str,
@@ -268,6 +270,7 @@ class LLM:
270 ...
271
272 @overload # LEGACY: multi (prompt + optional token ids)
273 + @deprecated("'prompt_token_ids' will become part of 'prompts")
274 def generate(
275 self,
276 prompts: List[str],
@@ -280,6 +283,7 @@ class LLM:
283 ...
284
285 @overload # LEGACY: single (token ids + optional prompt)
286 + @deprecated("'prompt_token_ids' will become part of 'prompts")
287 def generate(
288 self,
289 prompts: Optional[str] = None,
@@ -293,6 +297,7 @@ class LLM:
297 ...
298
299 @overload # LEGACY: multi (token ids + optional prompt)
300 + @deprecated("'prompt_token_ids' will become part of 'prompts")
301 def generate(
302 self,
303 prompts: Optional[List[str]] = None,
@@ -306,6 +311,7 @@ class LLM:
311 ...
312
313 @overload # LEGACY: single or multi token ids [pos-only]
314 + @deprecated("'prompt_token_ids' will become part of 'prompts")
315 def generate(
316 self,
317 prompts: None,
@@ -671,6 +677,7 @@ class LLM:
677 )
678
679 @overload # LEGACY: single (prompt + optional token ids)
680 + @deprecated("'prompt_token_ids' will become part of 'prompts")
681 def encode(
682 self,
683 prompts: str,
@@ -683,6 +690,7 @@ class LLM:
690 ...
691
692 @overload # LEGACY: multi (prompt + optional token ids)
693 + @deprecated("'prompt_token_ids' will become part of 'prompts")
694 def encode(
695 self,
696 prompts: List[str],
@@ -695,6 +703,7 @@ class LLM:
703 ...
704
705 @overload # LEGACY: single (token ids + optional prompt)
706 + @deprecated("'prompt_token_ids' will become part of 'prompts")
707 def encode(
708 self,
709 prompts: Optional[str] = None,
@@ -708,6 +717,7 @@ class LLM:
717 ...
718
719 @overload # LEGACY: multi (token ids + optional prompt)
720 + @deprecated("'prompt_token_ids' will become part of 'prompts")
721 def encode(
722 self,
723 prompts: Optional[List[str]] = None,
@@ -721,6 +731,7 @@ class LLM:
731 ...
732
733 @overload # LEGACY: single or multi token ids [pos-only]
734 + @deprecated("'prompt_token_ids' will become part of 'prompts")
735 def encode(
736 self,
737 prompts: None,
vllm/inputs/__init__.py
-31
@@ -38,34 +38,3 @@ __all__ = [
38 "InputProcessingContext",
39 "InputRegistry",
40 ]
41 -
42 -
43 -def __getattr__(name: str):
44 - import warnings
45 -
46 - if name == "PromptInput":
47 - msg = ("PromptInput has been renamed to PromptType. "
48 - "The original name will be removed in an upcoming version.")
49 -
50 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
51 -
52 - return PromptType
53 -
54 - if name == "LLMInputs":
55 - msg = ("LLMInputs has been renamed to DecoderOnlyInputs. "
56 - "The original name will be removed in an upcoming version.")
57 -
58 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
59 -
60 - return DecoderOnlyInputs
61 -
62 - if name == "EncoderDecoderLLMInputs":
63 - msg = (
64 - "EncoderDecoderLLMInputs has been renamed to EncoderDecoderInputs. "
65 - "The original name will be removed in an upcoming version.")
66 -
67 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
68 -
69 - return EncoderDecoderInputs
70 -
71 - raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
vllm/inputs/data.py
-31
@@ -358,34 +358,3 @@ def to_enc_dec_tuple_list(
358 return [(enc_dec_prompt["encoder_prompt"],
359 enc_dec_prompt["decoder_prompt"])
360 for enc_dec_prompt in enc_dec_prompts]
361 -
362 -
363 -def __getattr__(name: str):
364 - import warnings
365 -
366 - if name == "PromptInput":
367 - msg = ("PromptInput has been renamed to PromptType. "
368 - "The original name will be removed in an upcoming version.")
369 -
370 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
371 -
372 - return PromptType
373 -
374 - if name == "LLMInputs":
375 - msg = ("LLMInputs has been renamed to DecoderOnlyInputs. "
376 - "The original name will be removed in an upcoming version.")
377 -
378 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
379 -
380 - return DecoderOnlyInputs
381 -
382 - if name == "EncoderDecoderLLMInputs":
383 - msg = (
384 - "EncoderDecoderLLMInputs has been renamed to EncoderDecoderInputs. "
385 - "The original name will be removed in an upcoming version.")
386 -
387 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
388 -
389 - return EncoderDecoderInputs
390 -
391 - raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
vllm/model_executor/models/aria.py
+2 -3
@@ -32,9 +32,8 @@ from vllm.model_executor.models.utils import (AutoWeightsLoader, WeightsMapper,
32 maybe_prefix,
33 merge_multimodal_embeddings)
34 from vllm.multimodal import MULTIMODAL_REGISTRY
35 -from vllm.multimodal.base import MultiModalInputs
35 from vllm.multimodal.image import cached_get_image_processor
37 -from vllm.multimodal.inputs import NestedTensors
36 +from vllm.multimodal.inputs import MultiModalKwargs, NestedTensors
37 from vllm.multimodal.utils import (cached_get_tokenizer,
38 repeat_and_pad_placeholder_tokens)
39 from vllm.sequence import IntermediateTensors
@@ -451,7 +450,7 @@ def get_max_multimodal_tokens(ctx):
450
451
452 def input_mapper_for_aria(ctx, data):
454 - return MultiModalInputs(data)
453 + return MultiModalKwargs(data)
454
455
456 def input_processor(ctx, llm_inputs):
vllm/multimodal/__init__.py
-15
@@ -27,18 +27,3 @@ __all__ = [
27 "MULTIMODAL_REGISTRY",
28 "MultiModalRegistry",
29 ]
30 -
31 -
32 -def __getattr__(name: str):
33 - import warnings
34 -
35 - if name == "MultiModalInputs":
36 - msg = ("MultiModalInputs has been renamed to MultiModalKwargs. "
37 - "The original name will take another meaning in an upcoming "
38 - "version.")
39 -
40 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
41 -
42 - return MultiModalKwargs
43 -
44 - raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
vllm/multimodal/base.py
-15
@@ -433,18 +433,3 @@ class MultiModalPlaceholderMap:
433
434 return MultiModalPlaceholderMap.IndexMap(src=src_indices,
435 dest=dest_indices)
436 -
437 -
438 -def __getattr__(name: str):
439 - import warnings
440 -
441 - if name == "MultiModalInputs":
442 - msg = ("MultiModalInputs has been renamed to MultiModalKwargs. "
443 - "The original name will take another meaning in an upcoming "
444 - "version.")
445 -
446 - warnings.warn(DeprecationWarning(msg), stacklevel=2)
447 -
448 - return MultiModalKwargs
449 -
450 - raise AttributeError(f"module {__name__!r} has no attribute {name!r}")