Issues

• Setting a large number of max_num_documents for varlen_attn impacts the performance

  meta-pytorch/attention-gym#644 · Filter by author fegin opened on Oct 27, 2026
  1

• Fuse GDN gate and beta preprocessing in chunk training

  meta-pytorch/attention-gym#457 · Filter by author drisspg opened on Sep 2, 2026
  1

• Optimal way to handle Bert attention mask with Flex Attention

  meta-pytorch/attention-gym#208 · Filter by author conceptofmind opened on May 30, 2026
  10

• Add Exclusive Self-Attention (XSA) - output_mod pattern for FlexAttention

  meta-pytorch/attention-gym#201 · Filter by author ishaannk opened on May 5, 2026
  1

• Does FlexAttention Support torch.vmap?

  meta-pytorch/attention-gym#25 · Filter by author MiladInk opened on Aug 17, 2024
  3

• Writing to a globally scoped tensor from score_mod function

  meta-pytorch/attention-gym#19 · Filter by author jeffwillette opened on Aug 15, 2024
  5