Search results

<code>_compile=True</code> blows up GPU memory when seq length / block_mask change, but is best when both are fixed (FlexAttention)

Status: Open.

#159 In pytorch-labs/attention-gym;

· wileewang opened on Jul 22, 2025

[Question] sliding window attention + dynamic input shapes

Status: Open.

#138 In pytorch-labs/attention-gym;

· pfeatherstone opened on May 6, 2025

[Question] Precision of bfloat16 computations

Status: Open.

#131 In pytorch-labs/attention-gym;

· lukasHoel opened on Apr 4, 2025

RuntimeError in flex_attention: CUDA Device-Side Assertion Failure

Status: Open.

#130 In pytorch-labs/attention-gym;

· NuanBaobao opened on Mar 21, 2025

Would FlexAttention be useful over SDPA for masked language modeling?

Status: Open.

#129 In pytorch-labs/attention-gym;

· abdulfatir opened on Mar 20, 2025

Add Block Diffusion Attention Mask to attn_gym/masks

Status: Open.

#126 In pytorch-labs/attention-gym;

· Skylion007 opened on Mar 13, 2025

Hard-coded max_seq_len in document example

Status: Open.

#124 In pytorch-labs/attention-gym;

· adamoyoung opened on Mar 1, 2025

BlockMask Value Error

Status: Open.

#122 In pytorch-labs/attention-gym;

· Leo-T-Zang opened on Feb 17, 2025

Use case against efficient SDPA backend

Status: Open.

#121 In pytorch-labs/attention-gym;

· TParcollet opened on Feb 13, 2025

FlexAttention customizability for softmax

Status: Open.

#117 In pytorch-labs/attention-gym;

· veritas9872 opened on Feb 9, 2025

Sigmoid attention?

Status: Open.

#116 In pytorch-labs/attention-gym;

· lhallee opened on Feb 9, 2025

Error when using flex attention and F.sdpa together.

Status: Open.

#115 In pytorch-labs/attention-gym;

· wuyushuwys opened on Feb 8, 2025