[Kernel] Enable fp8 support for pplx and BatchedTritonExperts. (#18864)
Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
@@ -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,
|
||||
|
||||
Reference in New Issue
Block a user