[Kernel] Enable fp8 support for pplx and BatchedTritonExperts. (#18864)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2025-07-03 14:55:40 -07:00
committed by GitHub
parent 2f2fcb31b8
commit 78fe77534b
25 changed files with 1277 additions and 663 deletions
@@ -802,10 +802,7 @@ class Fp8MoEMethod(FusedMoEMethodBase):
self.quant_config.weight_block_size, False)
return BatchedTritonOrDeepGemmExperts(
max_num_tokens=max_num_tokens_per_rank,
world_size=prepare_finalize.
world_size, # type: ignore [attr-defined]
dp_size=prepare_finalize.
dp_size, # type: ignore [attr-defined]
num_dispatchers=prepare_finalize.num_dispatchers(),
use_fp8_w8a8=True,
block_shape=self.quant_config.weight_block_size,
per_act_token_quant=False,