Skip to content

add zero_tensors option with default False to flash_attn_varlen_func api #175

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wangye805 wants to merge 1 commit into from
Closed

add zero_tensors option with default False to flash_attn_varlen_func api #175

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 7 additions & 1 deletion flash_attn/flash_attn_interface.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -921,6 +921,7 @@ def forward(
return_softmax,
block_table,
is_grad_enabled,
zero_tensors: bool = False,
):
is_grad = is_grad_enabled and any(
x.requires_grad for x in [q, k, v]
Expand Down Expand Up @@ -949,6 +950,7 @@ def forward(
alibi_slopes=alibi_slopes,
return_softmax=return_softmax and dropout_p > 0,
block_table=block_table,
zero_tensors = zero_tensors,
)
if is_grad:
ctx.save_for_backward(
Expand All @@ -963,6 +965,7 @@ def forward(
ctx.softcap = softcap
ctx.alibi_slopes = alibi_slopes
ctx.deterministic = deterministic
ctx.zero_tensors = zero_tensors

out = out_padded[..., :head_size_og]
return out if not return_softmax else (out, softmax_lse, S_dmask)
Expand Down Expand Up @@ -998,11 +1001,12 @@ def backward(ctx, dout, *args):
ctx.alibi_slopes,
ctx.deterministic,
rng_state=rng_state,
zero_tensors = ctx.zero_tensors
)
dq = dq[..., : dout.shape[-1]] # We could have padded the head dimension
dk = dk[..., : dout.shape[-1]]
dv = dv[..., : dout.shape[-1]]
return dq, dk, dv, None, None, None, None, None, None, None, None, None, None, None, None, None, None
return dq, dk, dv, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None


def flash_attn_qkvpacked_func(
Expand Down Expand Up @@ -1394,6 +1398,7 @@ def flash_attn_varlen_func(
deterministic=False,
return_attn_probs=False,
block_table=None,
zero_tensors = False,
):
"""dropout_p should be set to 0.0 during evaluation
Supports multi-query and grouped-query attention (MQA/GQA) by passing in K, V with fewer heads
Expand Down Expand Up @@ -1468,6 +1473,7 @@ def flash_attn_varlen_func(
return_attn_probs,
block_table,
torch.is_grad_enabled(),
zero_tensors,
)


Expand Down
Loading