fix(humming): handle missing quant_method (#31185)

2026-07-16 11:15:52 +08:00
parent 22453ca63c
commit 7f9a902cd9
1 changed files with 1 additions and 1 deletions
@@ -227,7 +227,7 @@ class HummingConfig(QuantizationConfig):

    @classmethod
    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.
            # TODO: integrate humming kernels to mxfp4.py
            return None