simplify eplb policy

Author: wanghanqingLYT

vllm_ascend/eplb/core/policy/policy_factory.py

@@ -13,15 +13,10 @@ def generate_policy(policy_type: int, config: DynamicConfig) -> EplbPolicy:
             # Constraint applying Dynamic EPLB policy V2:
             # If there exists redundant expert:
             # only one redundant expert can be placed in one NPU and its physical expert index must be 0
-            
-            # Applying bipartite d2d expert weight update composing
-            0:MockLoadBalance,     # MockLoadBalance
-            1:DynamicEplb,         # Dynamic EPLB policy
-            2:DynamicEplbV2,       # Dynamic EPLB policy V2
 
             # Applying greedy d2d expert weight update composing
-            3:MockLoadBalance,   # MockLoadBalance
-            4:DynamicEplb,       # Dynamic EPLB policy
-            5:DynamicEplbV2,     # Dynamic EPLB policy V2
+            0:MockLoadBalance,     # MockLoadBalance: shuffle last physical expert on NPU 1 and 3
+            1:DynamicEplb,         # Dynamic EPLB policy: overall expert replacement based on current moe load
+            2:DynamicEplbV2,       # Dynamic EPLB policy V2:  expert replacement with constrained number of expert shuffle
         }
         return policy.get(policy_type, MockLoadBalance)(config)

vllm_ascend/eplb/core/worker/eplb_worker.py

@@ -71,8 +71,7 @@ def do_update(self):
         self.update_expert_map(new_expert_maps)
         logger.debug(f"[EPLB Process  new_map differs, performing D2D")
 
-        update_info = self.compose_expert_update_info_bipartite(new_expert_maps, self.old_expert_maps)\
-            if self.policy_type <= 2 else self.compose_expert_update_info_greedy(new_expert_maps, self.old_expert_maps)
+        update_info = self.compose_expert_update_info_greedy(new_expert_maps, self.old_expert_maps)
         self.old_expert_maps = new_expert_maps
         logger.info("EPLB Process compute complete")