#Multi-LoRA Tuning

Note: The LoRA configuration folder should be specified by exporting VLLM_TUNED_CONFIG_FOLDER=/path/to/configs. Without this, the shrink/expand kernels will use default configurations.

#Tuning Process

Multi-lora shrink/expand Triton kernel tuning follows a similar methodology from Triton MoE tuning.

1. Define the searching space. Here is an example of searching space:

   block_m_range = [16, 32, 64, 128, 256]
   block_n_range = [32, 64, 128, 256]
   block_k_range = [32, 64, 128, 256]
   num_warps_range = [4, 8]
   num_stage_range = [2, 3, 4, 5]
   num_ctas_range = [1]
   split_k_range = [4, 8, 16, 32, 64]

1. Get all hidden_state sizes and num_slices that the target model uses for a specific TP size.

For example, you can acquire the info by simply checking add_lora_linear:

   print(f"x_shape: {x.view(-1, x.shape[-1]).shape}")
   print(f"num_slices: {len(output_slices)}")
   for i in range(len(output_slices)):
       print(f"a{i} shape: {lora_a_stacked[i].shape}")
       print(f"b{i} shape: {lora_b_stacked[i].shape}")
   print("y_shape", y.shape)

1. Benchmark the shrink/expand kernel runtime with different kernel configurations generated from the pre-defined search space by performing a grid search to find the optimal kernel configuration. vLLM's benchmark_lora.py can be used to search for configurations for different shapes.

#Config Files

#File Naming

The gpu_name can be automatically detected by calling torch.cuda.get_device_name().

#JSON Structure

Optimal kernel configuration files are saved as JSON files with the structure config_data[max_loras][num_slices][m][k][n][i], where i is an optional dimension in the fused_moe_lora configuration, representing the intermediate size of the MoE layer.