yasserrmd's picture
WSP 2.0 sparse weight edit: strict-JSON fence repair (32,768 BF16 values, 16 w2 columns)
1a2bb00 verified
Raw History Blame Contribute Delete
1.9 kB
{
"selected": "c11",
"frozen_before_test": true,
"utc": "2026-11-18T17:24:09Z",
"chosen_patch_sha256": "38b9bb6891edfe0064c76faf057e94f33ed7f77a8e08fb7088db7a53efc41d59",
"reasoning": [
"Only c11 passed all six pre-registered gates on validation.",
"c11: pooled B exact 209/256 vs base 171 (+38); matched random-position control 172 and same-column random-direction control 172.",
"Greedy B 52 vs base 45; array exact 98 vs 73 with 0 lost base-correct array answers; A lost 1 (s2 count formatting), C/S/F lost 0; F_balanced 48 vs 45.",
"Passage PPL -0.034% relative; KL first-token 0.459 (lowest of all candidates), KL passage 0.00154 (highest among candidates but tiny).",
"Ranking is identical whether KL tie-break uses passage KL or first-token KL (only one passing candidate). Note: the KL variant for tie-breaking was fixed after c01-c09 results were seen; it did not affect the outcome.",
"10 of 11 rejected candidates failed G5 (lost >=1 base-correct array answer), several also G3 (A/C losses); c09 failed G1/G2 (fences increased to 117)."
],
"edits": [
{
"param": "model.layers.12.feed_forward.w2.weight",
"cols": [
4731,
7264,
7919
],
"kind": "scale",
"alpha": null,
"scale": 0.5,
"dir": "u"
},
{
"param": "model.layers.13.feed_forward.w2.weight",
"cols": [
488,
5186,
6715
],
"kind": "scale",
"alpha": null,
"scale": 0.5,
"dir": "u"
},
{
"param": "model.layers.14.feed_forward.w2.weight",
"cols": [
3659
],
"kind": "scale",
"alpha": null,
"scale": 0.5,
"dir": "u"
},
{
"param": "model.layers.15.feed_forward.w2.weight",
"cols": [
99,
532,
660,
926,
2017,
2439,
3263,
3682,
5191
],
"kind": "scale",
"alpha": null,
"scale": 0.5,
"dir": "u"
}
],
"changed_scalars": 32768,
"pct_unique": 0.0027998686686602605
}