[Quant] Support modelopt_mixed on Ampere (SM80/SM86) (#45306)

Signed-off-by: Mike G <180722391+mikekg@users.noreply.github.com>
This commit is contained in:
Mike G
2026-06-16 08:43:44 -04:00
committed by GitHub
parent 7e179e4bc0
commit a8c86eeb16
@@ -2299,7 +2299,13 @@ class ModelOptMixedPrecisionConfig(ModelOptQuantConfigBase):
@classmethod
def get_min_capability(cls) -> int:
return 89
# Ampere (SM80/SM86): NVFP4 routed experts run via Marlin W4A16, and FP8
# weight-only dense layers run via MarlinFP8 (W8A16, compute in
# bf16/fp16). FP8 MoE, if present, also routes to Marlin because
# TritonExperts gates its FP8 schemes behind supports_fp8() (cc>=89).
# None of these paths require native FP8 tensor cores, so SM80 is
# sufficient.
return 80
@classmethod
def override_quantization_method(