fix conflict

Author: Wanglongzhi2001

fastdeploy/model_executor/layers/moe/ep.py

@@ -83,7 +83,7 @@ def __init__(
             # prefill engine
             self.prefill_deepep_engine = deep_ep.Buffer(
                 self.group,
-                int(1e9),
+                int(5e8),
                 0,
                 low_latency_mode=False,
                 num_qps_per_rank=1,
@@ -97,7 +97,7 @@ def __init__(
             elif moe_phase.phase == "prefill":
                 self.prefill_deepep_engine = deep_ep.Buffer(
                     self.group,
-                    int(1e9),
+                    int(5e8),
                     0,
                     low_latency_mode=False,
                     num_qps_per_rank=1,
@@ -261,7 +261,6 @@ def __init__(
             num_max_dispatch_tokens_per_rank=num_max_dispatch_tokens_per_rank,
             hidden=hidden,
             num_experts=num_experts + redundant_experts_num,
-            moe_phase=moe_phase,
             ep_size=ep_size,
             ep_rank=ep_rank,
             splitwise_role=splitwise_role,

fastdeploy/model_executor/layers/moe/fused_moe_backend_base.py

@@ -53,6 +53,7 @@ def init_ep(self, layer: nn.Layer) -> None:
                     layer.fd_config.parallel_config.splitwise_role,
                     layer.ep_size,
                     layer.ep_rank,
+                    layer.fd_config.model_config.redundant_experts_num,
                 )
                 self.ep_decoder_runner = EPDecoderRunner(
                     layer.top_k,
@@ -75,6 +76,7 @@ def init_ep(self, layer: nn.Layer) -> None:
                         layer.fd_config.parallel_config.splitwise_role,
                         layer.ep_size,
                         layer.ep_rank,
+                        layer.fd_config.model_config.redundant_experts_num,
                     )
                 else:
                     from .ep import EPDecoderRunner
@@ -87,6 +89,7 @@ def init_ep(self, layer: nn.Layer) -> None:
                         layer.fd_config.parallel_config.splitwise_role,
                         layer.ep_size,
                         layer.ep_rank,
+                        layer.fd_config.model_config.redundant_experts_num,
                     )
 
     def process_loaded_weights(self, layer, weights) -> None: