biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Deprecation] Remove prompt_token_ids arg fallback in LLM.generate and LLM.embed (#18800)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-22 10:56:57 +08:00
committed by GitHub
parent 19fe1a0510
commit 8896eb72eb
24 changed files with 116 additions and 467 deletions
Download Patch File Download Diff File Expand all files Collapse all files

309
vllm/entrypoints/llm.py
View File

@@ -3,15 +3,13 @@
import itertools
from collections.abc import Sequence
from contextlib import contextmanager
from typing import (TYPE_CHECKING, Any, Callable, ClassVar, Optional, Union,
cast, overload)
from typing import TYPE_CHECKING, Any, Callable, Optional, Union, cast
import cloudpickle
import torch.nn as nn
from pydantic import ValidationError
from tqdm.auto import tqdm
from typing_extensions import TypeVar, deprecated
from typing_extensions import TypeVar
import vllm.envs as envs
from vllm.beam_search import (BeamSearchInstance, BeamSearchOutput,
@@ -40,7 +38,6 @@ from vllm.entrypoints.score_utils import (ScoreContentPartParam,
from vllm.entrypoints.utils import (_validate_truncation_size,
log_non_default_args)
from vllm.inputs import PromptType, SingletonPrompt, TextPrompt, TokensPrompt
from vllm.inputs.parse import parse_and_batch_prompt
from vllm.logger import init_logger
from vllm.lora.request import LoRARequest
from vllm.model_executor.layers.quantization import QuantizationMethods
@@ -54,7 +51,7 @@ from vllm.tasks import PoolingTask
from vllm.transformers_utils.tokenizer import (AnyTokenizer, MistralTokenizer,
get_cached_tokenizer)
from vllm.usage.usage_lib import UsageContext
from vllm.utils import Counter, Device, deprecate_kwargs, is_list_of
from vllm.utils import Counter, Device, is_list_of
from vllm.v1.sample.logits_processor import LogitsProcessor
if TYPE_CHECKING:
@@ -157,18 +154,6 @@ class LLM:
serving, use the [AsyncLLMEngine][vllm.AsyncLLMEngine] class instead.
"""
DEPRECATE_LEGACY: ClassVar[bool] = True
"""A flag to toggle whether to deprecate the legacy generate/encode API."""
@classmethod
@contextmanager
def deprecate_legacy_api(cls):
cls.DEPRECATE_LEGACY = True
yield
cls.DEPRECATE_LEGACY = False
def __init__(
self,
model: str,
@@ -325,99 +310,14 @@ class LLM:
return SamplingParams.from_optional(**self.default_sampling_params)
return SamplingParams()
@overload
def generate(
self,
prompts: Union[PromptType, Sequence[PromptType]],
/,
sampling_params: Optional[Union[SamplingParams,
Sequence[SamplingParams]]] = None,
*,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
) -> list[RequestOutput]:
...
@overload # LEGACY: single (prompt + optional token ids)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def generate(
self,
prompts: str,
sampling_params: Optional[Union[SamplingParams,
list[SamplingParams]]] = None,
prompt_token_ids: Optional[list[int]] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
) -> list[RequestOutput]:
...
@overload # LEGACY: multi (prompt + optional token ids)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def generate(
self,
prompts: list[str],
sampling_params: Optional[Union[SamplingParams,
list[SamplingParams]]] = None,
prompt_token_ids: Optional[list[list[int]]] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
) -> list[RequestOutput]:
...
@overload # LEGACY: single (token ids + optional prompt)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def generate(
self,
prompts: Optional[str] = None,
sampling_params: Optional[Union[SamplingParams,
list[SamplingParams]]] = None,
*,
prompt_token_ids: list[int],
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
) -> list[RequestOutput]:
...
@overload # LEGACY: multi (token ids + optional prompt)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def generate(
self,
prompts: Optional[list[str]] = None,
sampling_params: Optional[Union[SamplingParams,
list[SamplingParams]]] = None,
*,
prompt_token_ids: list[list[int]],
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
) -> list[RequestOutput]:
...
@overload # LEGACY: single or multi token ids [pos-only]
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def generate(
self,
prompts: None,
sampling_params: None,
prompt_token_ids: Union[list[int], list[list[int]]],
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
) -> list[RequestOutput]:
...
@deprecate_kwargs(
"prompt_token_ids",
is_deprecated=lambda: LLM.DEPRECATE_LEGACY,
additional_message="Please use the 'prompts' parameter instead.",
)
def generate(
self,
prompts: Union[Union[PromptType, Sequence[PromptType]],
Optional[Union[str, list[str]]]] = None,
sampling_params: Optional[Union[SamplingParams,
Sequence[SamplingParams]]] = None,
prompt_token_ids: Optional[Union[list[int], list[list[int]]]] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
priority: Optional[list[int]] = None,
) -> list[RequestOutput]:
"""Generates the completions for the input prompts.
@@ -460,15 +360,6 @@ class LLM:
"Try passing `--runner generate` to use the model as a "
"generative model.")
if prompt_token_ids is not None:
parsed_prompts = self._convert_v1_inputs(
prompts=cast(Optional[Union[str, list[str]]], prompts),
prompt_token_ids=prompt_token_ids,
)
else:
parsed_prompts = cast(Union[PromptType, Sequence[PromptType]],
prompts)
if sampling_params is None:
# Use default sampling params.
sampling_params = self.get_default_sampling_params()
@@ -483,10 +374,10 @@ class LLM:
# Add any modality specific loras to the corresponding prompts
lora_request = self._get_modality_specific_lora_reqs(
parsed_prompts, lora_request)
prompts, lora_request)
self._validate_and_add_requests(
prompts=parsed_prompts,
prompts=prompts,
params=sampling_params,
use_tqdm=use_tqdm,
lora_request=lora_request,
@@ -498,7 +389,7 @@ class LLM:
return self.engine_class.validate_outputs(outputs, RequestOutput)
def _get_modality_specific_lora_reqs(
self, parsed_prompts: Union[PromptType, Sequence[PromptType]],
self, prompts: Union[PromptType, Sequence[PromptType]],
lora_request: Optional[Union[list[LoRARequest], LoRARequest]]):
# Grab the lora config off the vllm config on the engine,
# since this is the same for both v0 & v1.
@@ -511,35 +402,33 @@ class LLM:
or (lora_config and lora_config.default_mm_loras is None)):
return lora_request
if not isinstance(parsed_prompts, Sequence):
parsed_prompts = [parsed_prompts]
if not isinstance(prompts, Sequence):
prompts = [prompts]
optional_loras = ([lora_request] * len(parsed_prompts)
optional_loras = ([lora_request] * len(prompts)
if not isinstance(lora_request, Sequence) else
lora_request)
return [
self._resolve_single_prompt_mm_lora(
parsed_prompt,
prompt,
opt_lora_req,
lora_config.default_mm_loras,
) for parsed_prompt, opt_lora_req in zip(parsed_prompts,
optional_loras)
) for prompt, opt_lora_req in zip(prompts, optional_loras)
]
def _resolve_single_prompt_mm_lora(self, parsed_prompt: PromptType,
def _resolve_single_prompt_mm_lora(self, prompt: PromptType,
lora_request: Optional[LoRARequest],
default_mm_loras: Optional[dict[str,
str]]):
if (not default_mm_loras or not isinstance(parsed_prompt, dict)
or "multi_modal_data" not in parsed_prompt):
if (not default_mm_loras or not isinstance(prompt, dict)
or "multi_modal_data" not in prompt):
return lora_request
parsed_prompt = cast(Union[TextPrompt, TokensPrompt], parsed_prompt)
prompt = cast(Union[TextPrompt, TokensPrompt], prompt)
intersection = set(
parsed_prompt["multi_modal_data"].keys()).intersection(
default_mm_loras.keys())
intersection = set(prompt["multi_modal_data"].keys()) \
.intersection(default_mm_loras.keys())
if not intersection:
return lora_request
if len(intersection) > 1:
@@ -933,11 +822,9 @@ class LLM:
lora_request=lora_request,
)
@overload
def encode(
self,
prompts: Union[PromptType, Sequence[PromptType]],
/,
pooling_params: Optional[Union[PoolingParams,
Sequence[PoolingParams]]] = None,
*,
@@ -946,107 +833,6 @@ class LLM:
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
pooling_task: PoolingTask = "encode",
tokenization_kwargs: Optional[dict[str, Any]] = None,
) -> list[PoolingRequestOutput]:
...
@overload # LEGACY: single (prompt + optional token ids)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def encode(
self,
prompts: str,
pooling_params: Optional[Union[PoolingParams,
Sequence[PoolingParams]]] = None,
prompt_token_ids: Optional[list[int]] = None,
truncate_prompt_tokens: Optional[int] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
pooling_task: PoolingTask = "encode",
tokenization_kwargs: Optional[dict[str, Any]] = None,
) -> list[PoolingRequestOutput]:
...
@overload # LEGACY: multi (prompt + optional token ids)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def encode(
self,
prompts: list[str],
pooling_params: Optional[Union[PoolingParams,
Sequence[PoolingParams]]] = None,
prompt_token_ids: Optional[list[list[int]]] = None,
truncate_prompt_tokens: Optional[int] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
pooling_task: PoolingTask = "encode",
tokenization_kwargs: Optional[dict[str, Any]] = None,
) -> list[PoolingRequestOutput]:
...
@overload # LEGACY: single (token ids + optional prompt)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def encode(
self,
prompts: Optional[str] = None,
pooling_params: Optional[Union[PoolingParams,
Sequence[PoolingParams]]] = None,
*,
prompt_token_ids: list[int],
truncate_prompt_tokens: Optional[int] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
pooling_task: PoolingTask = "encode",
tokenization_kwargs: Optional[dict[str, Any]] = None,
) -> list[PoolingRequestOutput]:
...
@overload # LEGACY: multi (token ids + optional prompt)
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def encode(
self,
prompts: Optional[list[str]] = None,
pooling_params: Optional[Union[PoolingParams,
Sequence[PoolingParams]]] = None,
*,
prompt_token_ids: list[list[int]],
truncate_prompt_tokens: Optional[int] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
pooling_task: PoolingTask = "encode",
tokenization_kwargs: Optional[dict[str, Any]] = None,
) -> list[PoolingRequestOutput]:
...
@overload # LEGACY: single or multi token ids [pos-only]
@deprecated("'prompt_token_ids' will become part of 'prompts'")
def encode(
self,
prompts: None,
pooling_params: None,
prompt_token_ids: Union[list[int], list[list[int]]],
truncate_prompt_tokens: Optional[int] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
pooling_task: PoolingTask = "encode",
tokenization_kwargs: Optional[dict[str, Any]] = None,
) -> list[PoolingRequestOutput]:
...
@deprecate_kwargs(
"prompt_token_ids",
is_deprecated=lambda: LLM.DEPRECATE_LEGACY,
additional_message="Please use the 'prompts' parameter instead.",
)
def encode(
self,
prompts: Union[Union[PromptType, Sequence[PromptType]],
Optional[Union[str, list[str]]]] = None,
pooling_params: Optional[Union[PoolingParams,
Sequence[PoolingParams]]] = None,
prompt_token_ids: Optional[Union[list[int], list[list[int]]]] = None,
truncate_prompt_tokens: Optional[int] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
lora_request: Optional[Union[list[LoRARequest], LoRARequest]] = None,
pooling_task: Optional[PoolingTask] = None,
tokenization_kwargs: Optional[dict[str, Any]] = None,
) -> list[PoolingRequestOutput]:
"""Apply pooling to the hidden states corresponding to the input
prompts.
@@ -1108,15 +894,6 @@ class LLM:
raise ValueError(
f"pooling_task must be one of {self.supported_tasks}.")
if prompt_token_ids is not None:
parsed_prompts = self._convert_v1_inputs(
prompts=cast(Optional[Union[str, list[str]]], prompts),
prompt_token_ids=prompt_token_ids,
)
else:
parsed_prompts = cast(Union[PromptType, Sequence[PromptType]],
prompts)
if pooling_params is None:
# Use default pooling params.
pooling_params = PoolingParams()
@@ -1134,7 +911,7 @@ class LLM:
tokenization_kwargs)
self._validate_and_add_requests(
prompts=parsed_prompts,
prompts=prompts,
params=pooling_params,
use_tqdm=use_tqdm,
lora_request=lora_request,
@@ -1148,7 +925,6 @@ class LLM:
def embed(
self,
prompts: Union[PromptType, Sequence[PromptType]],
/,
*,
truncate_prompt_tokens: Optional[int] = None,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
@@ -1198,7 +974,6 @@ class LLM:
def classify(
self,
prompts: Union[PromptType, Sequence[PromptType]],
/,
*,
use_tqdm: Union[bool, Callable[..., tqdm]] = True,
pooling_params: Optional[Union[PoolingParams,
@@ -1348,7 +1123,7 @@ class LLM:
_validate_truncation_size(model_config.max_model_len,
truncate_prompt_tokens, tokenization_kwargs)
parsed_prompts = []
prompts = list[PromptType]()
input_pairs = [(t1, t2) for t1, t2 in zip(data_1, data_2)]
@@ -1372,10 +1147,10 @@ class LLM:
else:
pooling_params_list.append(pooling_params)
parsed_prompts.append(engine_prompt)
prompts.append(engine_prompt)
self._validate_and_add_requests(
prompts=parsed_prompts,
prompts=prompts,
params=pooling_params_list,
use_tqdm=use_tqdm,
lora_request=lora_request,
@@ -1585,48 +1360,6 @@ class LLM:
assert isinstance(self.llm_engine, V1LLMEngine)
return self.llm_engine.get_metrics()
# LEGACY
def _convert_v1_inputs(
self,
prompts: Optional[Union[str, list[str]]],
prompt_token_ids: Optional[Union[list[int], list[list[int]]]],
):
# skip_tokenizer_init is now checked in engine
if prompts is None and prompt_token_ids is None:
raise ValueError(
"Either prompts or prompt_token_ids must be provided.")
if prompts is not None and prompt_token_ids is not None \
and len(prompts) != len(prompt_token_ids):
raise ValueError(
"The lengths of prompts and prompt_token_ids must be the same."
)
if prompts is not None:
prompts = [p["content"] for p in parse_and_batch_prompt(prompts)]
if prompt_token_ids is not None:
prompt_token_ids = [
p["content"] for p in parse_and_batch_prompt(prompt_token_ids)
]
if prompts is not None:
num_requests = len(prompts)
elif prompt_token_ids is not None:
num_requests = len(prompt_token_ids)
parsed_prompts: list[PromptType] = []
for i in range(num_requests):
item: PromptType
if prompts is not None:
item = TextPrompt(prompt=prompts[i])
elif prompt_token_ids is not None:
item = TokensPrompt(prompt_token_ids=prompt_token_ids[i])
else:
raise AssertionError
parsed_prompts.append(item)
return parsed_prompts
def _validate_and_add_requests(
self,
prompts: Union[PromptType, Sequence[PromptType]],