All pull requests

Pull requests

• Add opt-in freeze_attention / freeze_layernorm flags to attribute()

  #123·SpiliosDimakopoulos opened Oct 17, 2026·

  ·

• Chunk feature-node scores so MPS does not hit its INT_MAX limit

  #122·natemacfadden opened Oct 17, 2026·

  ·

• Load transcoder weights on CPU when the device is MPS

  #121·natemacfadden opened Oct 16, 2026·

  ·

• Read lazy decoder rows on CPU when the device is MPS

  #120·natemacfadden opened Oct 16, 2026·

  ·
  1

• Add head loadings: split attribution edges across attention heads

  #114·VivianSobers opened Sep 13, 2026·

  ·

• feat: interp-engine support

  #111·hijohnnylin opened Aug 21, 2026Collaborator·

  ·
  39

• Add long-context memory estimator

  #109·cv-chaitali opened Jul 17, 2026·

  ·
  2

• Auto-restart Colab runtime when the install swaps the preloaded numpy

  #106·KlyneChrysler opened Jul 5, 2026·

  ·
  1

  1

• Don't raise in ReplacementModel.__del__ (guard teardown cleanup)

  #104·iamsharduld opened Jun 23, 2026·

  ·
  1

• Stabilize row L1 graph normalization

  #97·Andrey170170 opened May 13, 2026Contributor·

  ·

• Fail fast on batched feature intervention inputs

  #89·Sohailm25 opened Mar 14, 2026·

  ·

• Fix batched get_activations zero-position masking

  #88·Sohailm25 opened Mar 14, 2026·

  ·

• Add transformers v5.0.0 compatibility

  #79·speediedan opened Jan 31, 2026Contributor·

  ·
  16

• TransformerLens 3.0 migration

  #57·bryce13950 opened Dec 7, 2025·

  ·

• WIP updating to use TLens v3

  #38·chanind opened Sep 3, 2025Contributor·

  ·
  25