[Frontend] new online quantization frontend (#38138)

Signed-off-by: Vasiliy Kuznetsov <vasiliy@meta.com>
This commit is contained in:
Vasiliy Kuznetsov
2026-04-03 11:58:39 -04:00
committed by GitHub
parent 97f92c6b47
commit 7b1a7423be
13 changed files with 1205 additions and 0 deletions
+5
View File
@@ -21,6 +21,7 @@ from vllm.config.multimodal import (
MultiModalConfig,
)
from vllm.config.pooler import PoolerConfig
from vllm.config.quantization import OnlineQuantizationConfigArgs
from vllm.config.scheduler import RunnerType
from vllm.config.utils import config, getattr_iter
from vllm.logger import init_logger
@@ -199,6 +200,10 @@ class ModelConfig:
`quantization_config` attribute in the model config file. If that is
`None`, we assume the model weights are not quantized and use `dtype` to
determine the data type of the weights."""
quantization_config: dict[str, Any] | OnlineQuantizationConfigArgs | None = None
"""Arguments for online quantization.
Auto-created when `quantization` equals to one of the string values of
the `OnlineQuantScheme` enum."""
allow_deprecated_quantization: bool = False
"""Whether to allow deprecated quantization methods."""
enforce_eager: bool = False