diff --git a/scripts/performance/configs/deepseek/deepseek_llm_pretrain.py b/scripts/performance/configs/deepseek/deepseek_llm_pretrain.py
index 293a2f3c06..4c4235c7d2 100644
--- a/scripts/performance/configs/deepseek/deepseek_llm_pretrain.py
+++ b/scripts/performance/configs/deepseek/deepseek_llm_pretrain.py
@@ -203,10 +203,6 @@ def deepseek_v3_pretrain_config_h100(
     )
     precision_config = get_precision_config(precision)
 
-    pp_layout = None
-    if base_cfg.pp_layout:
-        pp_layout = base_cfg.pp_layout
-
     cfg = pretrain_config()
     cfg.mixed_precision = precision_config
 
@@ -214,7 +210,11 @@ def deepseek_v3_pretrain_config_h100(
     cfg.model.pipeline_model_parallel_size = base_cfg.pipeline_model_parallel_size
     cfg.model.virtual_pipeline_model_parallel_size = base_cfg.virtual_pipeline_model_parallel_size
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
-    cfg.model.pipeline_model_parallel_layout = pp_layout or "Et|(tt|)*30mL"
+    if base_cfg.pp_layout:
+        cfg.model.pipeline_model_parallel_layout = base_cfg.pp_layout
+    else:
+        # Recompute layout based on updated PP/VP sizes
+        set_deepseek_v3_pipeline_model_parallel_layout(cfg.model)
 
     set_deepseek_v3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
diff --git a/scripts/performance/configs/qwen/qwen3_llm_pretrain.py b/scripts/performance/configs/qwen/qwen3_llm_pretrain.py
index 86446da203..c113920e56 100644
--- a/scripts/performance/configs/qwen/qwen3_llm_pretrain.py
+++ b/scripts/performance/configs/qwen/qwen3_llm_pretrain.py
@@ -78,6 +78,7 @@ def qwen3_235b_a22b_pretrain_config_gb300(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=True)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "flex"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
@@ -103,6 +104,7 @@ def qwen3_235b_a22b_pretrain_config_gb200(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=True)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "flex"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
@@ -128,6 +130,7 @@ def qwen3_235b_a22b_pretrain_config_b300(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=True)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "alltoall"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
@@ -153,6 +156,7 @@ def qwen3_235b_a22b_pretrain_config_b200(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=True)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "alltoall"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
@@ -178,6 +182,7 @@ def qwen3_235b_a22b_pretrain_config_h100(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=False)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "alltoall"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
@@ -203,6 +208,7 @@ def qwen3_30b_a3b_pretrain_config_gb300(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=True)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "flex"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
@@ -228,6 +234,7 @@ def qwen3_30b_a3b_pretrain_config_gb200(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=True)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "flex"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
@@ -253,6 +260,7 @@ def qwen3_30b_a3b_pretrain_config_b300(
     cfg.mixed_precision = precision_config
     cfg.comm_overlap = CommOverlapConfig(tp_comm_overlap=True)
     cfg.model.moe_flex_dispatcher_backend = base_cfg.moe_flex_dispatcher_backend
+    cfg.model.moe_token_dispatcher_type = "flex"
 
     set_qwen3_common_configs(cfg)
     set_workload_base_configs(cfg, base_cfg)
diff --git a/scripts/performance/configs/qwen/qwen3_workload_base_configs.py b/scripts/performance/configs/qwen/qwen3_workload_base_configs.py
index ec853e3ff8..85671f4904 100644
--- a/scripts/performance/configs/qwen/qwen3_workload_base_configs.py
+++ b/scripts/performance/configs/qwen/qwen3_workload_base_configs.py
@@ -31,6 +31,7 @@
 
 BASE_QWEN3_235B_A22B_CONFIG = WorkloadBaseConfig(
     expert_tensor_parallel_size=1,
+    moe_flex_dispatcher_backend="deepep",
 )
 
 
@@ -38,6 +39,7 @@
     expert_model_parallel_size=8,
     expert_tensor_parallel_size=1,
     global_batch_size=512,
+    moe_flex_dispatcher_backend="deepep",
 )
 
 BASE_QWEN3_NEXT_80B_A3B_CONFIG = WorkloadBaseConfig(