-
Notifications
You must be signed in to change notification settings - Fork 52
Open
Description
In the Generating customized workload description files part, the example code is
python -m workload_generator.AIOB_simAI_workload_generator
--world_size=32 --global_batch=64 --micro_batch=1
--num_layers=8 --num_attention_heads=176 --hidden_size=5120
--tensor_model_parallel_size=2 --seq_length=4096 --swiglu --ffn_hidden_size=16384
--moe_router_topk=4 --enable_sequence_parallel --expert_model_parallel_size=16
--num_experts=64 --moe_grouped_gemm --moe_enable --num_experts=4
But the workload_generator module has been updated which now contains both inference and training workload_generator option. Therefore, it is better to update the example code or else it won't execute.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels