[Quant] [Bugfix] Fix quantization config matching with hf_to_vllm_mapper (#20046)

This commit is contained in:
Kyle Sayers
2025-07-01 19:20:34 +09:00
committed by GitHub
parent c05596f1a3
commit 9025a9a705
17 changed files with 107 additions and 29 deletions
@@ -2,7 +2,7 @@
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import functools
from typing import Any, Callable, Optional, Union
from typing import TYPE_CHECKING, Any, Callable, Optional, Union
import torch
import torch.nn.functional as F
@@ -39,6 +39,9 @@ from vllm.platforms import current_platform
from vllm.scalar_type import scalar_types
from vllm.utils import has_deep_gemm
if TYPE_CHECKING:
from vllm.model_executor.models.utils import WeightsMapper
ACTIVATION_SCHEMES = ["static", "dynamic"]
logger = init_logger(__name__)
@@ -100,6 +103,11 @@ class Fp8Config(QuantizationConfig):
def get_config_filenames(cls) -> list[str]:
return []
def apply_vllm_mapper(self, hf_to_vllm_mapper: "WeightsMapper"):
if self.ignored_layers is not None:
self.ignored_layers = hf_to_vllm_mapper.apply_list(
self.ignored_layers)
@classmethod
def from_config(cls, config: dict[str, Any]) -> "Fp8Config":
quant_method = cls.get_from_keys(config, ["quant_method"])