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

[CLI][Doc] Formalize --mm-encoder-tp-mode (#23190)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-20 23:42:28 +08:00
committed by GitHub
parent b17109beea
commit 5efd6905bc
7 changed files with 104 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
vllm/engine/arg_utils.py
View File

@@ -28,12 +28,12 @@ from vllm.config import (BlockSize, CacheConfig, CacheDType, CompilationConfig,
DeviceConfig, DistributedExecutorBackend,
GuidedDecodingBackend, HfOverrides, KVEventsConfig,
KVTransferConfig, LoadConfig, LogprobsMode,
LoRAConfig, MambaDType, ModelConfig, ModelDType,
ModelImpl, MultiModalConfig, ObservabilityConfig,
ParallelConfig, PoolerConfig, PrefixCachingHashAlgo,
RunnerOption, SchedulerConfig, SchedulerPolicy,
SpeculativeConfig, TaskOption, TokenizerMode,
VllmConfig, get_attr_docs, get_field)
LoRAConfig, MambaDType, MMEncoderTPMode, ModelConfig,
ModelDType, ModelImpl, MultiModalConfig,
ObservabilityConfig, ParallelConfig, PoolerConfig,
PrefixCachingHashAlgo, RunnerOption, SchedulerConfig,
SchedulerPolicy, SpeculativeConfig, TaskOption,
TokenizerMode, VllmConfig, get_attr_docs, get_field)
from vllm.logger import init_logger
from vllm.platforms import CpuArchEnum, current_platform
from vllm.plugins import load_general_plugins
@@ -352,6 +352,7 @@ class EngineArgs:
MultiModalConfig.mm_processor_kwargs
disable_mm_preprocessor_cache: bool = False # DEPRECATED
mm_processor_cache_gb: int = MultiModalConfig.mm_processor_cache_gb
mm_encoder_tp_mode: MMEncoderTPMode = MultiModalConfig.mm_encoder_tp_mode
skip_mm_profiling: bool = MultiModalConfig.skip_mm_profiling
# LoRA fields
enable_lora: bool = False
@@ -434,16 +435,14 @@ class EngineArgs:
use_tqdm_on_load: bool = LoadConfig.use_tqdm_on_load
pt_load_map_location: str = LoadConfig.pt_load_map_location
enable_multimodal_encoder_data_parallel: bool = \
ParallelConfig.enable_multimodal_encoder_data_parallel
# DEPRECATED
enable_multimodal_encoder_data_parallel: bool = False
logits_processors: Optional[list[Union[
str, type[LogitsProcessor]]]] = ModelConfig.logits_processors
"""Custom logitproc types"""
async_scheduling: bool = SchedulerConfig.async_scheduling
# DEPRECATED
enable_prompt_adapter: bool = False
kv_sharing_fast_prefill: bool = \
CacheConfig.kv_sharing_fast_prefill
@@ -685,7 +684,8 @@ class EngineArgs:
**parallel_kwargs["worker_extension_cls"])
parallel_group.add_argument(
"--enable-multimodal-encoder-data-parallel",
**parallel_kwargs["enable_multimodal_encoder_data_parallel"])
action="store_true",
deprecated=True)
# KV cache arguments
cache_kwargs = get_kwargs(CacheConfig)
@@ -735,6 +735,8 @@ class EngineArgs:
multimodal_group.add_argument("--disable-mm-preprocessor-cache",
action="store_true",
deprecated=True)
multimodal_group.add_argument(
"--mm-encoder-tp-mode", **multimodal_kwargs["mm_encoder_tp_mode"])
multimodal_group.add_argument(
"--interleave-mm-strings",
**multimodal_kwargs["interleave_mm_strings"])
@@ -909,6 +911,14 @@ class EngineArgs:
self.mm_processor_cache_gb = envs.VLLM_MM_INPUT_CACHE_GIB
if self.enable_multimodal_encoder_data_parallel:
logger.warning(
"--enable-multimodal-encoder-data-parallel` is deprecated "
"and will be removed in v0.13. "
"Please use `--mm-encoder-tp-mode data` instead.")
self.mm_encoder_tp_mode = "data"
return ModelConfig(
model=self.model,
hf_config_path=self.hf_config_path,
@@ -947,6 +957,7 @@ class EngineArgs:
config_format=self.config_format,
mm_processor_kwargs=self.mm_processor_kwargs,
mm_processor_cache_gb=self.mm_processor_cache_gb,
mm_encoder_tp_mode=self.mm_encoder_tp_mode,
override_neuron_config=self.override_neuron_config,
override_pooler_config=self.override_pooler_config,
logits_processor_pattern=self.logits_processor_pattern,
@@ -1258,8 +1269,6 @@ class EngineArgs:
distributed_executor_backend=self.distributed_executor_backend,
worker_cls=self.worker_cls,
worker_extension_cls=self.worker_extension_cls,
enable_multimodal_encoder_data_parallel=self.
enable_multimodal_encoder_data_parallel,
)
if model_config.is_multimodal_model: