fix(internevo): mock flash_attn_2_cuda

DeepLink-org / DeepLinkExt

BSD 3-Clause "New" or "Revised" License

11 stars 0 forks source link

fix(internevo): mock flash_attn_2_cuda #54

Closed lljbash closed 6 months ago

lljbash commented 6 months ago

InternEVO uses FlashAttention 2.2.1, where the CUDA module is renamed from flash_attn_cuda to flash_attn_2_cuda. This commit mocks the correct module name.