About Flash attention and FlashAttention only supports Ampere GPUs or newer

[xiatiandefeng666]

Thank you very much for your work. I would like to know when my server is training the model, it prompts: FlashAttention only supports Ampere GPUs or newer, so can I train the model without changing the server? For example, without using FlashAttention，Thank you very much.

[RunsenXu]

Hi,

You could try using vanilla attention. Just comment this row:

PointLLM/pointllm/train/train_mem.py

Line 8 in e4875ce

replace_llama_attn_with_flash_attn()