fix(humming): handle missing quant_method (#31185)
This commit is contained in:
@@ -227,7 +227,7 @@ class HummingConfig(QuantizationConfig):
|
|||||||
|
|
||||||
@classmethod
|
@classmethod
|
||||||
def override_quantization_method(cls, hf_quant_cfg, user_quant) -> str | None:
|
def override_quantization_method(cls, hf_quant_cfg, user_quant) -> str | None:
|
||||||
if hf_quant_cfg["quant_method"] == "mxfp4":
|
if hf_quant_cfg.get("quant_method") == "mxfp4":
|
||||||
# NOTE: gpt-oss has a special weight loading logic, so we don't support it now.
|
# NOTE: gpt-oss has a special weight loading logic, so we don't support it now.
|
||||||
# TODO: integrate humming kernels to mxfp4.py
|
# TODO: integrate humming kernels to mxfp4.py
|
||||||
return None
|
return None
|
||||||
|
|||||||
Reference in New Issue
Block a user