Support async in DeepEP (#4610)
Co-authored-by: Cheng Wan <cwan39@gatech.edu>
This commit is contained in:
@@ -239,6 +239,7 @@ class DeepseekV2MoE(nn.Module):
|
||||
num_local_experts=config.n_routed_experts // self.tp_size,
|
||||
hidden_size=config.hidden_size,
|
||||
params_dtype=config.torch_dtype,
|
||||
async_finish=True, # TODO
|
||||
)
|
||||
|
||||
def forward(
|
||||
|
||||
Reference in New Issue
Block a user