diff --git a/scripts/performance/configs/qwen/qwen3_workload_base_configs.py b/scripts/performance/configs/qwen/qwen3_workload_base_configs.py index 85671f4904..2450ba695b 100644 --- a/scripts/performance/configs/qwen/qwen3_workload_base_configs.py +++ b/scripts/performance/configs/qwen/qwen3_workload_base_configs.py @@ -190,22 +190,13 @@ QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_BF16_V1, num_gpus=256, pipeline_model_parallel_size=4, - virtual_pipeline_model_parallel_size=12, - expert_model_parallel_size=16, - global_batch_size=8192, -) - - -QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_CS_V2 = replace( - QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_CS_V1, - num_gpus=256, - pipeline_model_parallel_size=4, expert_model_parallel_size=32, cuda_graph_scope=["attn", "moe_router", "moe_preprocess"], global_batch_size=8192, ) +QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_CS_V2 = QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_BF16_V2 QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_MX_V2 = QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_CS_V2