Convert formatting to use ruff instead of yapf + isort (#26247)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-10-05 07:06:22 -07:00
committed by GitHub
parent 17edd8a807
commit d6953beb91
1508 changed files with 115205 additions and 94107 deletions
+20 -6
View File
@@ -1,12 +1,26 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from .data import (DataPrompt, DecoderOnlyInputs, EmbedsInputs, EmbedsPrompt,
EncoderDecoderInputs, ExplicitEncoderDecoderPrompt,
ProcessorInputs, PromptType, SingletonInputs,
SingletonPrompt, TextPrompt, TokenInputs, TokensPrompt,
build_explicit_enc_dec_prompt, embeds_inputs,
to_enc_dec_tuple_list, token_inputs, zip_enc_dec_prompts)
from .data import (
DataPrompt,
DecoderOnlyInputs,
EmbedsInputs,
EmbedsPrompt,
EncoderDecoderInputs,
ExplicitEncoderDecoderPrompt,
ProcessorInputs,
PromptType,
SingletonInputs,
SingletonPrompt,
TextPrompt,
TokenInputs,
TokensPrompt,
build_explicit_enc_dec_prompt,
embeds_inputs,
to_enc_dec_tuple_list,
token_inputs,
zip_enc_dec_prompts,
)
__all__ = [
"DataPrompt",
+34 -25
View File
@@ -7,8 +7,11 @@ import torch
from typing_extensions import NotRequired, TypedDict, TypeIs, TypeVar
if TYPE_CHECKING:
from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalInputs,
MultiModalUUIDDict)
from vllm.multimodal.inputs import (
MultiModalDataDict,
MultiModalInputs,
MultiModalUUIDDict,
)
class TextPrompt(TypedDict):
@@ -134,23 +137,27 @@ more than one prompt, i.e.
def is_tokens_prompt(prompt: SingletonPrompt) -> TypeIs[TokensPrompt]:
return (isinstance(prompt, dict) and "prompt_token_ids" in prompt
and "prompt_embeds" not in prompt)
return (
isinstance(prompt, dict)
and "prompt_token_ids" in prompt
and "prompt_embeds" not in prompt
)
def is_embeds_prompt(prompt: SingletonPrompt) -> TypeIs[EmbedsPrompt]:
return (isinstance(prompt, dict) and "prompt_token_ids" not in prompt
and "prompt_embeds" in prompt)
return (
isinstance(prompt, dict)
and "prompt_token_ids" not in prompt
and "prompt_embeds" in prompt
)
_T1_co = TypeVar("_T1_co",
bound=SingletonPrompt,
default=SingletonPrompt,
covariant=True)
_T2_co = TypeVar("_T2_co",
bound=SingletonPrompt,
default=SingletonPrompt,
covariant=True)
_T1_co = TypeVar(
"_T1_co", bound=SingletonPrompt, default=SingletonPrompt, covariant=True
)
_T2_co = TypeVar(
"_T2_co", bound=SingletonPrompt, default=SingletonPrompt, covariant=True
)
# TODO: Make fields ReadOnly once mypy supports it
@@ -309,8 +316,9 @@ def build_explicit_enc_dec_prompt(
def zip_enc_dec_prompts(
enc_prompts: Iterable[_T1],
dec_prompts: Iterable[Optional[_T2]],
mm_processor_kwargs: Optional[Union[Iterable[dict[str, Any]],
dict[str, Any]]] = None,
mm_processor_kwargs: Optional[
Union[Iterable[dict[str, Any]], dict[str, Any]]
] = None,
) -> list[ExplicitEncoderDecoderPrompt[_T1, _T2]]:
"""
Zip encoder and decoder prompts together into a list of
@@ -329,20 +337,21 @@ def zip_enc_dec_prompts(
encoder_prompt,
decoder_prompt,
cast(dict[str, Any], mm_processor_kwargs),
) for (encoder_prompt,
decoder_prompt) in zip(enc_prompts, dec_prompts)
)
for (encoder_prompt, decoder_prompt) in zip(enc_prompts, dec_prompts)
]
return [
build_explicit_enc_dec_prompt(encoder_prompt, decoder_prompt,
mm_proc_kwargs)
for (encoder_prompt, decoder_prompt, mm_proc_kwargs
) in zip(enc_prompts, dec_prompts, mm_processor_kwargs)
build_explicit_enc_dec_prompt(encoder_prompt, decoder_prompt, mm_proc_kwargs)
for (encoder_prompt, decoder_prompt, mm_proc_kwargs) in zip(
enc_prompts, dec_prompts, mm_processor_kwargs
)
]
def to_enc_dec_tuple_list(
enc_dec_prompts: Iterable[ExplicitEncoderDecoderPrompt[_T1, _T2]],
) -> list[tuple[_T1, Optional[_T2]]]:
return [(enc_dec_prompt["encoder_prompt"],
enc_dec_prompt["decoder_prompt"])
for enc_dec_prompt in enc_dec_prompts]
return [
(enc_dec_prompt["encoder_prompt"], enc_dec_prompt["decoder_prompt"])
for enc_dec_prompt in enc_dec_prompts
]
+25 -16
View File
@@ -1,16 +1,22 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from collections.abc import Sequence
from typing import (TYPE_CHECKING, Literal, NamedTuple, Optional, TypedDict,
Union, cast)
from typing import TYPE_CHECKING, Literal, NamedTuple, Optional, TypedDict, Union, cast
from typing_extensions import TypeIs
from vllm.utils import is_list_of
from .data import (EmbedsPrompt, ExplicitEncoderDecoderPrompt, ProcessorInputs,
PromptType, SingletonInputs, SingletonPrompt, TextPrompt,
TokensPrompt)
from .data import (
EmbedsPrompt,
ExplicitEncoderDecoderPrompt,
ProcessorInputs,
PromptType,
SingletonInputs,
SingletonPrompt,
TextPrompt,
TokensPrompt,
)
if TYPE_CHECKING:
import torch
@@ -44,8 +50,10 @@ def parse_raw_prompts(
# case 4: array of token arrays
return [TokensPrompt(prompt_token_ids=elem) for elem in prompt]
raise TypeError("prompt must be a string, array of strings, "
"array of tokens, or array of token arrays")
raise TypeError(
"prompt must be a string, array of strings, "
"array of tokens, or array of token arrays"
)
class ParsedStrPrompt(TypedDict):
@@ -68,8 +76,9 @@ class ParsedEmbedsPrompt(TypedDict):
content: EmbedsPrompt
ParsedSingletonPrompt = Union[ParsedStrPrompt, ParsedTextPrompt,
ParsedTokensPrompt, ParsedEmbedsPrompt]
ParsedSingletonPrompt = Union[
ParsedStrPrompt, ParsedTextPrompt, ParsedTokensPrompt, ParsedEmbedsPrompt
]
def parse_singleton_prompt(prompt: SingletonPrompt) -> ParsedSingletonPrompt:
@@ -79,19 +88,19 @@ def parse_singleton_prompt(prompt: SingletonPrompt) -> ParsedSingletonPrompt:
# Type ignores are because mypy does not correctly infer the TypedDicts
# Pyright does succeed.
if "prompt_embeds" in prompt:
return ParsedEmbedsPrompt(
type="embeds", content=prompt) # type: ignore[typeddict-item]
return ParsedEmbedsPrompt(type="embeds", content=prompt) # type: ignore[typeddict-item]
elif "prompt_token_ids" in prompt:
return ParsedTokensPrompt(
type="tokens", content=prompt) # type: ignore[typeddict-item]
return ParsedTokensPrompt(type="tokens", content=prompt) # type: ignore[typeddict-item]
elif "prompt" in prompt:
return ParsedTextPrompt(type="text", content=prompt)
raise TypeError(
"inputs must be a string, TextPrompt, TokensPrompt, or EmbedsPrompt")
"inputs must be a string, TextPrompt, TokensPrompt, or EmbedsPrompt"
)
def is_explicit_encoder_decoder_prompt(
prompt: PromptType, ) -> TypeIs[ExplicitEncoderDecoderPrompt]:
prompt: PromptType,
) -> TypeIs[ExplicitEncoderDecoderPrompt]:
return isinstance(prompt, dict) and "encoder_prompt" in prompt
@@ -118,7 +127,7 @@ def get_prompt_components(prompt: PromptType) -> PromptComponents:
if isinstance(prompt, str):
return PromptComponents(text=prompt)
if (encoder_prompt := prompt.get("encoder_prompt")):
if encoder_prompt := prompt.get("encoder_prompt"):
return get_prompt_components(encoder_prompt) # type: ignore[arg-type]
return PromptComponents(
+106 -68
View File
@@ -10,23 +10,37 @@ from vllm.config import ModelConfig
from vllm.logger import init_logger
from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
from vllm.multimodal.cache import BaseMultiModalProcessorCache
from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalEncDecInputs,
MultiModalInputs, MultiModalUUIDDict)
from vllm.multimodal.inputs import (
MultiModalDataDict,
MultiModalEncDecInputs,
MultiModalInputs,
MultiModalUUIDDict,
)
from vllm.multimodal.processing import BaseMultiModalProcessor
from vllm.transformers_utils.tokenizer import AnyTokenizer
from .data import (DecoderOnlyInputs, EmbedsInputs, EmbedsPrompt,
EncoderDecoderInputs, ExplicitEncoderDecoderPrompt,
ProcessorInputs, PromptType, SingletonInputs,
SingletonPrompt, TextPrompt, TokenInputs, TokensPrompt,
embeds_inputs, token_inputs)
from .data import (
DecoderOnlyInputs,
EmbedsInputs,
EmbedsPrompt,
EncoderDecoderInputs,
ExplicitEncoderDecoderPrompt,
ProcessorInputs,
PromptType,
SingletonInputs,
SingletonPrompt,
TextPrompt,
TokenInputs,
TokensPrompt,
embeds_inputs,
token_inputs,
)
from .parse import is_explicit_encoder_decoder_prompt, parse_singleton_prompt
logger = init_logger(__name__)
class InputPreprocessor:
def __init__(
self,
model_config: ModelConfig,
@@ -43,23 +57,26 @@ class InputPreprocessor:
def get_tokenizer(self) -> AnyTokenizer:
if self.tokenizer is None:
raise ValueError("You cannot pass text prompts when "
"`skip_tokenizer_init` is True")
raise ValueError(
"You cannot pass text prompts when `skip_tokenizer_init` is True"
)
return self.tokenizer
def get_bos_token_id(self) -> Optional[int]:
if self.tokenizer is None:
logger.warning("Using None for BOS token id because tokenizer "
"is not initialized")
logger.warning(
"Using None for BOS token id because tokenizer is not initialized"
)
return None
return self.tokenizer.bos_token_id
def get_eos_token_id(self) -> Optional[int]:
if self.tokenizer is None:
logger.warning("Using None for EOS token id because tokenizer "
"is not initialized")
logger.warning(
"Using None for EOS token id because tokenizer is not initialized"
)
return None
return self.tokenizer.eos_token_id
@@ -74,22 +91,26 @@ class InputPreprocessor:
if not self.model_config.is_encoder_decoder:
logger.warning_once(
"Using None for decoder start token id because "
"this is not an encoder/decoder model.")
"this is not an encoder/decoder model."
)
return None
if self.model_config is None or self.model_config.hf_config is None:
logger.warning_once(
"Using None for decoder start token id because "
"model config is not available.")
"model config is not available."
)
return None
dec_start_token_id = getattr(self.model_config.hf_config,
"decoder_start_token_id", None)
dec_start_token_id = getattr(
self.model_config.hf_config, "decoder_start_token_id", None
)
if dec_start_token_id is None:
logger.warning_once(
"Falling back on <BOS> for decoder start token "
"id because decoder start token id is not "
"available.")
"available."
)
dec_start_token_id = self.get_bos_token_id()
return dec_start_token_id
@@ -159,8 +180,10 @@ class InputPreprocessor:
# use decoder_start_token_id as decoder_input_ids
decoder_input_ids = self._get_default_enc_dec_decoder_prompt()
if (len(decoder_input_ids) == 0
or decoder_input_ids[0] != decoder_start_token_id):
if (
len(decoder_input_ids) == 0
or decoder_input_ids[0] != decoder_start_token_id
):
decoder_input_ids = [decoder_start_token_id] + decoder_input_ids
return decoder_input_ids
@@ -254,7 +277,8 @@ class InputPreprocessor:
raise ValueError(
f"mm_hashes must contain only strings, got: {mm_hashes}. "
"This is likely due to an incorrect custom implementation of "
"MultiModalProcessor.apply method.")
"MultiModalProcessor.apply method."
)
return mm_input
@@ -263,8 +287,9 @@ class InputPreprocessor:
parsed_content: EmbedsPrompt,
) -> EmbedsInputs:
if not self.model_config.enable_prompt_embeds:
raise ValueError("You must set `--enable-prompt-embeds` to input "
"`prompt_embeds`.")
raise ValueError(
"You must set `--enable-prompt-embeds` to input `prompt_embeds`."
)
prompt_embeds = parsed_content["prompt_embeds"]
@@ -276,24 +301,25 @@ class InputPreprocessor:
prompt_embeds = prompt_embeds.squeeze(dim=0)
if prompt_embeds.ndim != 2:
raise ValueError(
"prompt_embeds must be of shape (seq_len, hidden_size).")
raise ValueError("prompt_embeds must be of shape (seq_len, hidden_size).")
# Tensors must be on CPU for serialization between processes
# in the MsgpackEncoder. Casting to CPU here ensures that there is no
# hidden device transfer in the critical path of generation.
prompt_embeds = prompt_embeds.cpu()
return embeds_inputs(prompt_embeds=prompt_embeds,
cache_salt=parsed_content.get("cache_salt"))
return embeds_inputs(
prompt_embeds=prompt_embeds, cache_salt=parsed_content.get("cache_salt")
)
def _truncate_inputs(
self,
inputs: list[int],
tokenization_kwargs: Optional[dict[str, Any]] = None) -> list[int]:
if not tokenization_kwargs or "truncation" not in \
tokenization_kwargs or self.tokenizer is None:
self, inputs: list[int], tokenization_kwargs: Optional[dict[str, Any]] = None
) -> list[int]:
if (
not tokenization_kwargs
or "truncation" not in tokenization_kwargs
or self.tokenizer is None
):
return inputs
max_length = tokenization_kwargs["max_length"]
@@ -311,7 +337,8 @@ class InputPreprocessor:
mm_uuids: Optional[MultiModalUUIDDict] = None,
) -> Union[TokenInputs, MultiModalInputs]:
prompt_token_ids = self._truncate_inputs(
parsed_content["prompt_token_ids"], tokenization_kwargs)
parsed_content["prompt_token_ids"], tokenization_kwargs
)
inputs: Union[TokenInputs, MultiModalInputs]
if self.model_config.is_multimodal_model:
@@ -324,8 +351,7 @@ class InputPreprocessor:
)
else:
if parsed_content.get("multi_modal_data"):
raise ValueError(
"This model does not support multimodal inputs")
raise ValueError("This model does not support multimodal inputs")
inputs = token_inputs(prompt_token_ids)
@@ -354,8 +380,7 @@ class InputPreprocessor:
)
else:
if parsed_content.get("multi_modal_data"):
raise ValueError(
"This model does not support multimodal inputs")
raise ValueError("This model does not support multimodal inputs")
prompt_token_ids = self._tokenize_prompt(
prompt_text,
@@ -415,16 +440,20 @@ class InputPreprocessor:
encoder_inputs: SingletonInputs,
decoder_inputs: Optional[SingletonInputs],
) -> EncoderDecoderInputs:
if (encoder_inputs["type"] == "embeds"
or decoder_inputs and decoder_inputs["type"] == "embeds"):
raise ValueError("Embedding inputs are not supported for encoder-"
"decoder models")
if (
encoder_inputs["type"] == "embeds"
or decoder_inputs
and decoder_inputs["type"] == "embeds"
):
raise ValueError(
"Embedding inputs are not supported for encoder-decoder models"
)
# Needed for mypy
encoder_inputs = cast(Union[TokenInputs, MultiModalInputs],
encoder_inputs)
decoder_inputs = cast(Optional[Union[TokenInputs, MultiModalInputs]],
decoder_inputs)
encoder_inputs = cast(Union[TokenInputs, MultiModalInputs], encoder_inputs)
decoder_inputs = cast(
Optional[Union[TokenInputs, MultiModalInputs]], decoder_inputs
)
if decoder_inputs is None:
if self.model_config.hf_config.model_type == "whisper":
@@ -434,16 +463,18 @@ class InputPreprocessor:
# overridden by the audio features.
dec_token_ids = encoder_inputs["prompt_token_ids"].copy()
else:
dec_token_ids = self._prepare_decoder_input_ids_for_generation(
None)
dec_token_ids = self._prepare_decoder_input_ids_for_generation(None)
decoder_inputs = token_inputs(dec_token_ids)
else:
if "multi_modal_data" in decoder_inputs:
raise ValueError("Multi-modal decoder inputs of encoder-"
"decoder models are not supported yet")
raise ValueError(
"Multi-modal decoder inputs of encoder-"
"decoder models are not supported yet"
)
dec_token_ids = self._prepare_decoder_input_ids_for_generation(
decoder_inputs["prompt_token_ids"])
decoder_inputs["prompt_token_ids"]
)
decoder_inputs["prompt_token_ids"] = dec_token_ids
return EncoderDecoderInputs(
@@ -460,10 +491,14 @@ class InputPreprocessor:
For encoder/decoder models only:
Separate Encoder/Decoder inputs from a MultiModalEncDecInputs
"""
if (inputs["type"] == "embeds" or decoder_inputs_to_override
and decoder_inputs_to_override["type"] == "embeds"):
raise ValueError("Embedding inputs are not supported for encoder-"
"decoder models")
if (
inputs["type"] == "embeds"
or decoder_inputs_to_override
and decoder_inputs_to_override["type"] == "embeds"
):
raise ValueError(
"Embedding inputs are not supported for encoder-decoder models"
)
# Needed for mypy
inputs = cast(
@@ -480,9 +515,11 @@ class InputPreprocessor:
if inputs["type"] == "multimodal": # Multimodal data inputs
if "encoder_prompt_token_ids" not in inputs:
raise RuntimeError("You should register an encoder-decoder "
"multi-modal processor for encoder-decoder "
"models.")
raise RuntimeError(
"You should register an encoder-decoder "
"multi-modal processor for encoder-decoder "
"models."
)
inputs = cast(MultiModalEncDecInputs, inputs)
encoder_inputs = token_inputs(inputs["encoder_prompt_token_ids"])
@@ -564,9 +601,9 @@ class InputPreprocessor:
# For multimodal model, override decoder prompt from processor
# with explicit decoder prompt.
if self.model_config.is_multimodal_model:
encoder_inputs, decoder_inputs = (
self._split_enc_dec_mm_inputs(encoder_inputs,
decoder_inputs))
encoder_inputs, decoder_inputs = self._split_enc_dec_mm_inputs(
encoder_inputs, decoder_inputs
)
else:
# `cast` is needed for mypy, but not pyright
inputs = self._prompt_to_llm_inputs(
@@ -576,8 +613,7 @@ class InputPreprocessor:
)
if self.model_config.is_multimodal_model:
# Encoder-Decoder Multimodal model
encoder_inputs, decoder_inputs = (
self._split_enc_dec_mm_inputs(inputs))
encoder_inputs, decoder_inputs = self._split_enc_dec_mm_inputs(inputs)
else:
encoder_inputs = inputs
decoder_inputs = None
@@ -589,8 +625,9 @@ class InputPreprocessor:
prompt_inputs: DecoderOnlyInputs,
) -> DecoderOnlyInputs:
if "prompt_token_ids" in prompt_inputs:
prompt_inputs = cast(Union[TokenInputs, MultiModalInputs],
prompt_inputs) # Needed for mypy
prompt_inputs = cast(
Union[TokenInputs, MultiModalInputs], prompt_inputs
) # Needed for mypy
return prompt_inputs
@@ -641,8 +678,9 @@ class InputPreprocessor:
)
if is_explicit_encoder_decoder_prompt(prompt):
raise ValueError("Cannot pass encoder-decoder prompt "
"to decoder-only models")
raise ValueError(
"Cannot pass encoder-decoder prompt to decoder-only models"
)
# Decoder-only operation
# `cast` is needed for mypy, but not pyright