This commit is contained in:
Woosuk Kwon 2024-04-26 05:27:38 +00:00
parent d16a348477
commit 4ea41d01a9

View File

@ -8,6 +8,7 @@ _DEFAULT_BLOCK_SIZES = {
"block_b": 2, "block_b": 2,
} }
def flash_attn( def flash_attn(
q: jax.Array, # [batch, seq_len, num_heads, head_size] q: jax.Array, # [batch, seq_len, num_heads, head_size]
k: jax.Array, # [batch, seq_len, num_heads, head_size] k: jax.Array, # [batch, seq_len, num_heads, head_size]