Skip to content

v0.3.0

Choose a tag to compare

@github-actions github-actions released this 17 Jul 02:31

[0.3.0] - 2026-07-17

First release shipping OpenDDE antibody-antigen co-folding (--model opendde / opendde-abag, built on the Protenix-v2 stack plus a structural-token expander), the ESMC fused-RoPE attention kernel (an accuracy-neutral speedup for the embed path), and opt-in diffusion trace replay for Boltz-2, BoltzGen, Protenix-v2, and OpenDDE (lossless, collapses per-step host dispatch). Also lands the standing perf-regression and UX-regression harnesses as release-gate legs, plus two release-gate blocker fixes (OF3 pairformer gates now skip gracefully when their golden key is host-missing; the perf gate compares against the right per-card-type baseline so a P150a run is no longer judged against a P300c number).

OpenDDE's antibody-antigen accuracy is currently weak on the 9dsg target, a confirmed reference-level ceiling rather than a port bug (see docs/opendde-port.md); the device-vs-reference pharma-parity legs for 9dsg and 1ahw are resolved and documented.

Release gate (scripts/release_gate.py, examples/prot.yaml, 200 steps / 5 samples, seed 0, Blackhole P150a):

model CA-RMSD TM floor result
Boltz-2 1.863 Å 0.891 ≤3.0 Å / ≥0.75 PASS
ESMFold2 1.774 Å 0.915 ≤4.0 Å / ≥0.65 PASS
ESMFold2-fast 1.725 Å 0.909 ≤4.5 Å / ≥0.60 PASS
Protenix-v2 1.417 Å 0.936 ≤6.0 Å / ≥0.50 PASS

BoltzGen designability — n=4, examples/binder.yaml: scRMSD median 0.820 Å, 4/4 designs (100%) ≤2 Å (floor ≤2.0 Å / ≥50%) — PASS.

ESMC embedding parity (fused-RoPE shipped path vs reference esm, 76-residue sequence, PCC floor 0.99):

model per-res PCC pooled logits argmax result
esmc-300m 0.99961 0.99993 0.99990 1.0000 PASS
esmc-600m 0.99964 0.99989 0.99996 1.0000 PASS

UX gate (scripts/ux_regression.py, examples/trpcage.yaml): every shipped surface (Boltz-2, ESMFold2, ESMFold2-fast, Protenix-v2, OpenDDE, ESMC-600m embed) cleared live-progress advancement, strict mmCIF/npz parse, and results/manifest shape — PASS.

Perf gate (scripts/perf_regression.py, Blackhole P150a, trpcage 20 aa single-sequence, 1 recycle / 10 steps / 1 sample, warm 2+5, ±15% threshold):

model metric baseline current delta result
boltz2 structures/s 1.186 1.190 +0.3% PASS
esmfold2 structures/s 1.665 1.705 +2.4% PASS
esmfold2-fast structures/s 2.271 2.290 +0.8% PASS
protenix-v2 structures/s 2.406 2.383 -1.0% PASS
opendde structures/s 1.920 1.922 +0.1% PASS
esmc-600m seq/s 21.09 20.92 -0.8% PASS

No perf regression. No OOM observed through the gate targets.

Added

  • OpenDDE antibody-antigen co-folding (opendde / opendde-abag).
  • ESMC fused-RoPE attention kernel for the embed path (accuracy-neutral speedup).
  • Opt-in diffusion trace replay for Boltz-2, BoltzGen, Protenix-v2, and OpenDDE.
  • perf-regression and UX-regression harnesses as standing release-gate legs.

Fixed

  • Release gate now skips OF3 pairformer gates gracefully when their golden key is host-missing.
  • Perf gate compares against the correct per-card-type baseline (P300c vs P150a mismatch no longer reads as a false regression).