(query, key, value, is_causal=False)
| 11 | |
| 12 | |
| 13 | def flash_attention(query, key, value, is_causal=False): |
| 14 | with sdpa_kernel(SDPBackend.FLASH_ATTENTION): |
| 15 | return F.scaled_dot_product_attention(query, key, value, is_causal=is_causal) |
| 16 | |
| 17 | |
| 18 | def cudnn_sdpa(query, key, value, is_causal=False): |
no outgoing calls
no test coverage detected