[Misc] Reorganize inputs (#35182)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
@@ -7,20 +7,18 @@ from typing import Any, Literal
|
||||
|
||||
import vllm.envs as envs
|
||||
from vllm.config import VllmConfig
|
||||
from vllm.inputs.data import (
|
||||
ProcessorInputs,
|
||||
from vllm.inputs import (
|
||||
EngineInput,
|
||||
PromptType,
|
||||
SingletonInputs,
|
||||
SingletonInput,
|
||||
split_enc_dec_input,
|
||||
)
|
||||
from vllm.inputs.parse import split_enc_dec_inputs
|
||||
from vllm.inputs.preprocess import InputPreprocessor
|
||||
from vllm.logger import init_logger
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
|
||||
from vllm.multimodal.encoder_budget import MultiModalBudget
|
||||
from vllm.multimodal.inputs import (
|
||||
MultiModalFeatureSpec,
|
||||
)
|
||||
from vllm.multimodal.inputs import MultiModalFeatureSpec
|
||||
from vllm.multimodal.utils import argsort_mm_positions
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.pooling_params import PoolingParams
|
||||
@@ -197,7 +195,7 @@ class InputProcessor:
|
||||
def process_inputs(
|
||||
self,
|
||||
request_id: str,
|
||||
prompt: PromptType | ProcessorInputs,
|
||||
prompt: PromptType | EngineInput,
|
||||
params: SamplingParams | PoolingParams,
|
||||
supported_tasks: tuple[SupportedTask, ...],
|
||||
arrival_time: float | None = None,
|
||||
@@ -232,7 +230,7 @@ class InputProcessor:
|
||||
if arrival_time is None:
|
||||
arrival_time = prompt.get("arrival_time", time.time()) # type: ignore[assignment]
|
||||
|
||||
processed_inputs: ProcessorInputs = prompt # type: ignore[assignment]
|
||||
processed_inputs: EngineInput = prompt # type: ignore[assignment]
|
||||
else:
|
||||
logger.warning_once(
|
||||
"Passing raw prompts to InputProcessor is deprecated "
|
||||
@@ -250,7 +248,7 @@ class InputProcessor:
|
||||
|
||||
current_platform.validate_request(processed_inputs, params)
|
||||
|
||||
encoder_inputs, decoder_inputs = split_enc_dec_inputs(processed_inputs)
|
||||
encoder_inputs, decoder_inputs = split_enc_dec_input(processed_inputs)
|
||||
self._validate_model_inputs(encoder_inputs, decoder_inputs)
|
||||
|
||||
# Mypy can be conservative for TypedDict unions; normalize access.
|
||||
@@ -385,7 +383,7 @@ class InputProcessor:
|
||||
|
||||
def _validate_model_input(
|
||||
self,
|
||||
prompt_inputs: SingletonInputs,
|
||||
prompt_input: SingletonInput,
|
||||
prompt_type: Literal["encoder", "decoder"],
|
||||
) -> None:
|
||||
model_config = self.model_config
|
||||
@@ -393,20 +391,18 @@ class InputProcessor:
|
||||
|
||||
prompt_ids = (
|
||||
None
|
||||
if prompt_inputs["type"] == "embeds"
|
||||
else prompt_inputs["prompt_token_ids"]
|
||||
if prompt_input["type"] == "embeds"
|
||||
else prompt_input["prompt_token_ids"]
|
||||
)
|
||||
prompt_embeds = (
|
||||
prompt_inputs["prompt_embeds"]
|
||||
if prompt_inputs["type"] == "embeds"
|
||||
else None
|
||||
prompt_input["prompt_embeds"] if prompt_input["type"] == "embeds" else None
|
||||
)
|
||||
|
||||
prompt_len = length_from_prompt_token_ids_or_embeds(prompt_ids, prompt_embeds)
|
||||
self._validate_prompt_len(prompt_len, prompt_type)
|
||||
|
||||
if prompt_inputs["type"] == "multimodal":
|
||||
decoder_mm_positions = prompt_inputs["mm_placeholders"]
|
||||
if prompt_input["type"] == "multimodal":
|
||||
decoder_mm_positions = prompt_input["mm_placeholders"]
|
||||
for modality, mm_positions in decoder_mm_positions.items():
|
||||
for mm_position in mm_positions:
|
||||
embed_length = mm_position.get_num_embeds()
|
||||
@@ -439,10 +435,10 @@ class InputProcessor:
|
||||
|
||||
def _validate_model_inputs(
|
||||
self,
|
||||
encoder_inputs: SingletonInputs | None,
|
||||
decoder_inputs: SingletonInputs,
|
||||
encoder_input: SingletonInput | None,
|
||||
decoder_input: SingletonInput,
|
||||
):
|
||||
if encoder_inputs is not None:
|
||||
self._validate_model_input(encoder_inputs, prompt_type="encoder")
|
||||
if encoder_input is not None:
|
||||
self._validate_model_input(encoder_input, prompt_type="encoder")
|
||||
|
||||
self._validate_model_input(decoder_inputs, prompt_type="decoder")
|
||||
self._validate_model_input(decoder_input, prompt_type="decoder")
|
||||
|
||||
Reference in New Issue
Block a user