[Model][Quantization] Add GGUF support for MiniMax-M2.1 (#36965)

Signed-off-by: kangletian <Letian.Kang@amd.com>
This commit is contained in:
PikaPikachu
2026-03-30 14:24:06 +08:00
committed by GitHub
parent fec5aeca12
commit 63babd17f1
5 changed files with 137 additions and 10 deletions
@@ -3,7 +3,10 @@
from collections.abc import Mapping
from types import MappingProxyType
from typing import Any
from typing import TYPE_CHECKING, Any
if TYPE_CHECKING:
from vllm.model_executor.layers.quantization import QuantizationMethods
import gguf
import torch
@@ -79,6 +82,16 @@ class GGUFConfig(QuantizationConfig):
def from_config(cls, config: dict[str, Any]) -> "GGUFConfig":
return cls()
@classmethod
def override_quantization_method(
cls, hf_quant_cfg: dict[str, Any], user_quant: str | None
) -> "QuantizationMethods | None":
# When user explicitly specifies --quantization gguf, override
# whatever quantization method is in the HF model config (e.g. fp8).
if user_quant == "gguf":
return "gguf"
return None
def get_quant_method(
self, layer: torch.nn.Module, prefix: str
) -> "QuantizeMethodBase | None":