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

[V1] Remove legacy input registry (#15673)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-03-28 14:34:34 +08:00
committed by GitHub
parent 8693e47e6a
commit 355f66348c
17 changed files with 132 additions and 153 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/v1/engine/async_llm.py
View File

@@ -14,10 +14,11 @@ from vllm.config import ModelConfig, VllmConfig
from vllm.engine.arg_utils import AsyncEngineArgs
from vllm.engine.protocol import EngineClient
from vllm.envs import VLLM_V1_OUTPUT_PROC_CHUNK_SIZE
from vllm.inputs import INPUT_REGISTRY, InputRegistry, PromptType
from vllm.inputs import PromptType
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.outputs import RequestOutput
from vllm.pooling_params import PoolingParams
from vllm.prompt_adapter.request import PromptAdapterRequest
@@ -48,7 +49,7 @@ class AsyncLLM(EngineClient):
executor_class: type[Executor],
log_stats: bool,
usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
input_registry: InputRegistry = INPUT_REGISTRY,
mm_registry: MultiModalRegistry = MULTIMODAL_REGISTRY,
use_cached_outputs: bool = False,
log_requests: bool = True,
start_engine_loop: bool = True,
@@ -90,7 +91,7 @@ class AsyncLLM(EngineClient):
self.processor = Processor(
vllm_config=vllm_config,
tokenizer=self.tokenizer,
input_registry=input_registry,
mm_registry=mm_registry,
)
# OutputProcessor (converts EngineCoreOutputs --> RequestOutput).

4
vllm/v1/engine/llm_engine.py
View File

@@ -11,7 +11,7 @@ from vllm.config import ParallelConfig, VllmConfig
from vllm.distributed import stateless_destroy_torch_distributed_process_group
from vllm.engine.arg_utils import EngineArgs
from vllm.engine.metrics_types import StatLoggerBase
from vllm.inputs import INPUT_REGISTRY, InputRegistry, PromptType
from vllm.inputs import PromptType
from vllm.logger import init_logger
from vllm.lora.request import LoRARequest
from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
@@ -44,7 +44,6 @@ class LLMEngine:
log_stats: bool,
usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
stat_loggers: Optional[dict[str, StatLoggerBase]] = None,
input_registry: InputRegistry = INPUT_REGISTRY,
mm_registry: MultiModalRegistry = MULTIMODAL_REGISTRY,
use_cached_outputs: bool = False,
multiprocess_mode: bool = False,
@@ -80,7 +79,6 @@ class LLMEngine:
# Processor (convert Inputs --> EngineCoreRequests)
self.processor = Processor(vllm_config=vllm_config,
tokenizer=self.tokenizer,
input_registry=input_registry,
mm_registry=mm_registry)
# OutputProcessor (convert EngineCoreOutputs --> RequestOutput).

22
vllm/v1/engine/processor.py
View File

@@ -5,8 +5,7 @@ from collections.abc import Mapping
from typing import Optional, Union
from vllm.config import VllmConfig
from vllm.inputs import (INPUT_REGISTRY, InputRegistry, ProcessorInputs,
PromptType, SingletonInputsAdapter)
from vllm.inputs import ProcessorInputs, PromptType
from vllm.inputs.parse import split_enc_dec_inputs
from vllm.inputs.preprocess import InputPreprocessor
from vllm.lora.request import LoRARequest
@@ -31,7 +30,6 @@ class Processor:
self,
vllm_config: VllmConfig,
tokenizer: BaseTokenizerGroup,
input_registry: InputRegistry = INPUT_REGISTRY,
mm_registry: MultiModalRegistry = MULTIMODAL_REGISTRY,
):
@@ -210,7 +208,6 @@ class Processor:
self._validate_model_inputs(processed_inputs, lora_request)
encoder_inputs, decoder_inputs = split_enc_dec_inputs(processed_inputs)
decoder_inputs = SingletonInputsAdapter(decoder_inputs)
# TODO: Impl encoder-decoder
if encoder_inputs is not None:
@@ -221,8 +218,9 @@ class Processor:
sampling_params = params.clone()
# If unset max tokens, then generate up to the max_model_len.
if sampling_params.max_tokens is None:
sampling_params.max_tokens = (self.model_config.max_model_len -
len(decoder_inputs.prompt_token_ids))
sampling_params.max_tokens = (
self.model_config.max_model_len -
len(decoder_inputs["prompt_token_ids"]))
sampling_params.update_from_generation_config(
self.generation_config_fields, eos_token_id)
sampling_params.update_from_tokenizer(
@@ -232,8 +230,8 @@ class Processor:
sorted_mm_inputs: Optional[list[MultiModalKwargs]] = None
sorted_mm_positions: Optional[list[PlaceholderRange]] = None
sorted_mm_hashes: Optional[list[str]] = None
if (decoder_mm_inputs := decoder_inputs.multi_modal_data):
assert isinstance(decoder_mm_inputs, MultiModalKwargs)
if decoder_inputs["type"] == "multimodal":
decoder_mm_inputs = decoder_inputs["mm_kwargs"]
# The output of merged multi-modal processor (`decoder_mm_inputs`)
# contains the kwargs for all items from all modalities.
@@ -254,8 +252,8 @@ class Processor:
sorted_mm_positions,
sorted_mm_hashes,
) = merge_and_sort_multimodal_metadata(
decoder_inputs.multi_modal_placeholders,
decoder_inputs.multi_modal_hashes if self.use_hash else None,
decoder_inputs["mm_placeholders"],
decoder_inputs["mm_hashes"] if self.use_hash else None,
)
# NOTE: Sort multimodal inputs/kwargs ONLY IF there are multiple
@@ -281,8 +279,8 @@ class Processor:
return EngineCoreRequest(
request_id=request_id,
prompt=decoder_inputs.prompt,
prompt_token_ids=decoder_inputs.prompt_token_ids,
prompt=decoder_inputs.get("prompt"),
prompt_token_ids=decoder_inputs["prompt_token_ids"],
mm_inputs=sorted_mm_inputs,
mm_hashes=sorted_mm_hashes,
mm_placeholders=sorted_mm_positions,