[Quantization] [Refactor] Create special "GptOssMxfp4MoeMethod" (#39604)

Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
This commit is contained in:
Yongye Zhu
2026-04-13 12:53:58 -04:00
committed by GitHub
parent 4d042ed85f
commit 739e5945dc
18 changed files with 129 additions and 34 deletions
@@ -406,7 +406,7 @@ class ModelOptFp8Config(ModelOptQuantConfigBase):
@classmethod
def override_quantization_method(
cls, hf_quant_cfg, user_quant
cls, hf_quant_cfg, user_quant, hf_config=None
) -> QuantizationMethods | None:
algo = cls._extract_modelopt_quant_algo(hf_quant_cfg)
if algo is not None and algo == "FP8":
@@ -1028,7 +1028,7 @@ class ModelOptNvFp4Config(ModelOptQuantConfigBase):
@classmethod
def override_quantization_method(
cls, hf_quant_cfg, user_quant
cls, hf_quant_cfg, user_quant, hf_config=None
) -> QuantizationMethods | None:
algo = cls._extract_modelopt_quant_algo(hf_quant_cfg)
if algo is not None and ("NVFP4" in algo or "FP4" in algo):
@@ -1525,7 +1525,7 @@ class ModelOptMxFp8Config(ModelOptQuantConfigBase):
@classmethod
def override_quantization_method(
cls, hf_quant_cfg, user_quant
cls, hf_quant_cfg, user_quant, hf_config=None
) -> QuantizationMethods | None:
algo = cls._extract_modelopt_quant_algo(hf_quant_cfg)
if algo is not None and "MXFP8" in algo:
@@ -2052,7 +2052,7 @@ class ModelOptMixedPrecisionConfig(ModelOptQuantConfigBase):
@classmethod
def override_quantization_method(
cls, hf_quant_cfg, user_quant
cls, hf_quant_cfg, user_quant, hf_config=None
) -> QuantizationMethods | None:
algo = cls._extract_modelopt_quant_algo(hf_quant_cfg)
if algo is not None and algo == "MIXED_PRECISION":