mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-04 16:22:57 +08:00
remove some code in ep.py (#2947)
This commit is contained in:
@@ -361,22 +361,6 @@ class EPDecoderRunner(EPRunner):
|
||||
return recv_hidden_states, recv_expert_count, handle
|
||||
|
||||
def combine(self, ffn_out, topk_idx, topk_weights, handle):
|
||||
# TODO(@wufeisheng): Delete them when deepep in PaddlePaddle is fixed
|
||||
(
|
||||
src_info,
|
||||
layout_range,
|
||||
num_max_dispatch_tokens_per_rank,
|
||||
num_experts,
|
||||
) = handle
|
||||
|
||||
handle = (
|
||||
src_info,
|
||||
layout_range,
|
||||
num_max_dispatch_tokens_per_rank,
|
||||
None,
|
||||
num_experts,
|
||||
)
|
||||
|
||||
combined_hidden_states, combine_hook = self.ep_engine.low_latency_combine(
|
||||
ffn_out, topk_idx, topk_weights, handle
|
||||
)
|
||||
|
Reference in New Issue
Block a user