xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-09 15:37:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Doc] Improve documentation for multimodal CLI args (#16960)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung 2025-04-22 16:35:35 +08:00 committed by GitHub
parent e4d6144232
commit 8f7bace7c3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 17 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
vllm/config.py
View File

@ -54,13 +54,15 @@ if TYPE_CHECKING:
from vllm.transformers_utils.tokenizer_group.base_tokenizer_group import ( from vllm.transformers_utils.tokenizer_group.base_tokenizer_group import (
BaseTokenizerGroup) BaseTokenizerGroup)
Config = TypeVar("Config", bound=DataclassInstance) ConfigType = type[DataclassInstance]
else: else:
QuantizationConfig = None QuantizationConfig = None
Config = TypeVar("Config") ConfigType = type
logger = init_logger(__name__) logger = init_logger(__name__)
ConfigT = TypeVar("ConfigT", bound=ConfigType)
# This value is chosen to have a balance between ITL and TTFT. Note it is # This value is chosen to have a balance between ITL and TTFT. Note it is
# not optimized for throughput. # not optimized for throughput.
_DEFAULT_MAX_NUM_BATCHED_TOKENS = 2048 _DEFAULT_MAX_NUM_BATCHED_TOKENS = 2048
@ -162,7 +164,7 @@ def get_attr_docs(cls: type[Any]) -> dict[str, str]:
return out return out
def config(cls: type[Config]) -> type[Config]: def config(cls: ConfigT) -> ConfigT:
""" """
A decorator that ensures all fields in a dataclass have default values A decorator that ensures all fields in a dataclass have default values
and that each field has a docstring. and that each field has a docstring.
@ -181,7 +183,7 @@ def config(cls: type[Config]) -> type[Config]:
return cls return cls
def get_field(cls: type[Config], name: str) -> Field: def get_field(cls: ConfigType, name: str) -> Field:
"""Get the default factory field of a dataclass by name. Used for getting """Get the default factory field of a dataclass by name. Used for getting
default factory fields in `EngineArgs`.""" default factory fields in `EngineArgs`."""
if not is_dataclass(cls): if not is_dataclass(cls):
@ -2749,6 +2751,9 @@ class MultiModalConfig:
The maximum number of input items allowed per prompt for each modality. The maximum number of input items allowed per prompt for each modality.
This should be a JSON string that will be parsed into a dictionary. This should be a JSON string that will be parsed into a dictionary.
Defaults to 1 (V0) or 999 (V1) for each modality. Defaults to 1 (V0) or 999 (V1) for each modality.
For example, to allow up to 16 images and 2 videos per prompt:
``{"images": 16, "videos": 2}``
""" """
def compute_hash(self) -> str: def compute_hash(self) -> str:

14
vllm/engine/arg_utils.py
View File

@ -17,7 +17,7 @@ from typing_extensions import TypeIs
import vllm.envs as envs import vllm.envs as envs
from vllm import version from vllm import version
from vllm.config import (BlockSize, CacheConfig, CacheDType, CompilationConfig, from vllm.config import (BlockSize, CacheConfig, CacheDType, CompilationConfig,
Config, ConfigFormat, DecodingConfig, Device, ConfigFormat, ConfigType, DecodingConfig, Device,
DeviceConfig, DistributedExecutorBackend, HfOverrides, DeviceConfig, DistributedExecutorBackend, HfOverrides,
KVTransferConfig, LoadConfig, LoadFormat, LoRAConfig, KVTransferConfig, LoadConfig, LoadFormat, LoRAConfig,
ModelConfig, ModelImpl, MultiModalConfig, ModelConfig, ModelImpl, MultiModalConfig,
@ -304,7 +304,7 @@ class EngineArgs:
"""Check if the class is a custom type.""" """Check if the class is a custom type."""
return cls.__module__ != "builtins" return cls.__module__ != "builtins"
def get_kwargs(cls: type[Config]) -> dict[str, Any]: def get_kwargs(cls: ConfigType) -> dict[str, Any]:
cls_docs = get_attr_docs(cls) cls_docs = get_attr_docs(cls)
kwargs = {} kwargs = {}
for field in fields(cls): for field in fields(cls):
@ -678,13 +678,15 @@ class EngineArgs:
'--mm-processor-kwargs', '--mm-processor-kwargs',
default=None, default=None,
type=json.loads, type=json.loads,
help=('Overrides for the multimodal input mapping/processing, ' help=('Overrides for the multi-modal processor obtained from '
'e.g., image processor. For example: ``{"num_crops": 4}``.')) '``AutoProcessor.from_pretrained``. The available overrides '
'depend on the model that is being run.'
'For example, for Phi-3-Vision: ``{"num_crops": 4}``.'))
parser.add_argument( parser.add_argument(
'--disable-mm-preprocessor-cache', '--disable-mm-preprocessor-cache',
action='store_true', action='store_true',
help='If true, then disables caching of the multi-modal ' help='If True, disable caching of the processed multi-modal '
'preprocessor/mapper. (not recommended)') 'inputs.')
# LoRA related configs # LoRA related configs
parser.add_argument('--enable-lora', parser.add_argument('--enable-lora',