All pull requests

Pull requests

• exp369: Analyze default-model conditioning and monomer discrepancies

  agent-generated
  #371·timodonnell opened Oct 21, 2026Member·

  ·
  4

• exp368: benchmark guidance and confirm rollout-weighted contact maps

  agent-generated
  #370·timodonnell opened Oct 21, 2026Member·

  ·

• exp364: redesign-only sweep, completed cooldowns, and checkpoint publication

  agent-generated
  #367·timodonnell opened Oct 13, 2026Member·

  ·

• exp362: train smaller models on the complete default corpus

  agent-generated
  #365·timodonnell opened Oct 13, 2026Member·

  ·

• exp363: train the double-width 5.86B default-model variant

  agent-generated
  #366·timodonnell opened Oct 13, 2026Member·

  ·

• exp341: hooked-readout harness (TransformersBackend.from_model) and harness checks 1-2

  #361·jsilter opened Oct 10, 2026Contributor·

  ·
  6

• contacts_v1: make the pairwise readout and metric helpers public

  #360·jsilter opened Oct 10, 2026Contributor·

  ·
  1

• exp357: Measure training versus validation contact R-precision

  agent-generated
  #359·timodonnell opened Oct 9, 2026Member·

  ·
  5

• Train and publish a MarinFold rollout proofreader on 50,000 PDB structures

  agent-generated
  #358·timodonnell opened Oct 9, 2026Member·

  ·
  2

• exp354: short rollouts reduce precision but improve AUC on eval-val

  agent-generated
  #355·timodonnell opened Oct 9, 2026Member·

  ·
  4

• Report default-model eval-val precision with historical CASP context

  agent-generated
  #353·timodonnell opened Oct 8, 2026Member·

  ·
  2

• exp350: evaluate pair-held-out FoldBench complexes

  agent-generated
  #351·timodonnell opened Oct 5, 2026Member·

  ·
  8

• exp347: fine-tune Qwen3.5 base models for protein contacts

  agent-generated
  #348·timodonnell opened Oct 2, 2026Member·

  ·

• exp345: audit contact-map errors and 19,400 whole-map rollouts

  agent-generated
  #346·timodonnell opened Sep 30, 2026Member·

  ·
  5

• exp180: update contact accuracy slides through September

  agent-generated
  #344·timodonnell opened Sep 30, 2026Member·

  ·

• Analyze MarinFold vs ESMFold2 performance on PDB-deduped monomers

  #342·znichols opened Sep 28, 2026Contributor·

  ·
  1

• exp339: diagnose prediction diversity and accuracy bottlenecks

  agent-generated
  #340·timodonnell opened Sep 26, 2026Member·

  ·

• exp336: quantify structure-aware training-corpus deduplication

  agent-generated
  #338·timodonnell opened Sep 25, 2026Member·

  ·

• exp335: pilot Helico decoy confidence ranking

  agent-generated
  #337·timodonnell opened Sep 25, 2026Member·

  ·
  47

• exp308: online contact novelty penalties for fold search

  agent-generated
  #309·timodonnell opened Sep 22, 2026Member·

  ·
  3

• exp306: contact-block beam search for eval-val and fold switching

  agent-generated
  #307·timodonnell opened Sep 22, 2026Member·

  ·
  5

• exp304: blind inference search for alternate protein folds

  agent-generated
  #305·timodonnell opened Sep 22, 2026Member·

  ·
  4

• exp301: does MarinFold represent both folds of fold-switching proteins?

  #303·timodonnell opened Sep 21, 2026Member·

  ·
  9

• Add exp299 delta-stream contact experiment

  #300·znichols opened Sep 16, 2026Contributor·

  ·
  14

• Two animations: MarinFold writing a contact map, and what the map is worth

  #296·timodonnell opened Sep 16, 2026Member·

  ·
  6