[Feature] Support batch invariance on ROCm (#52231)

Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
Matt
2026-10-12 10:53:26 -05:00
committed by GitHub
co-authored by Claude Opus 5.5 Andreas Karatzas
parent 50b21a2667
commit e0317b28ce
31 changed files with 1106 additions and 122 deletions
@@ -51,6 +51,10 @@ class AiterMxfp8Experts(Mxfp8TritonExpertsBase):
def _supports_parallel_config(moe_parallel_config) -> bool:
return True
@staticmethod
def _supports_batch_invariance() -> bool:
return False
@staticmethod
def is_supported_config(
cls, moe_config, weight_key, activation_key, activation_format
@@ -218,7 +218,7 @@ class OCP_MXQuantizationEmulationTritonExperts(TritonExperts):
global_num_experts=global_num_experts,
expert_map=expert_map,
a1q_scale=None,
a2_scale=None,
a2_scale=a2_scale,
workspace13=workspace13,
workspace2=workspace2,
expert_tokens_meta=expert_tokens_meta,