no high_priority_a2a_comm_stream

Signed-off-by: Malay Nagda <malayn@nvidia.com>

Author: malay-nagda

src/megatron/bridge/perf_recipes/deepseek/gb300/deepseek_v3.py

@@ -211,7 +211,7 @@ def deepseek_v3_pretrain_256gpu_gb300_nvfp4_config() -> ConfigContainer:
     _benchmark_common(cfg)
     _enable_deepseek_full_iteration_mxfp8(cfg, fp8_dot_product_attention=False, fp8_output_proj=False)
     # cfg.model.moe_hybridep_num_sms_preprocessing = 108
-    # cfg.model.high_priority_a2a_comm_stream = False
+    cfg.model.high_priority_a2a_comm_stream = False
     cfg.model.mla_down_proj_fusion = True
 
     # Keep process settings next to the recipe so users can see the exact benchmark environment.