495 Hessian/YAQA-correctable real targets — of 497 total language-model tensors (2 stay full precision, 1 handled separately, see the tensor taxonomy below)
Estimated completion
Complete
Real output model already saved -- nothing left to estimate.
Total time running8h 46m
Real build composition: 287 tensor(s) reused from a prior build's seeded cache · 208 computed live this run · 495 done total
System memory: 88.4 GB free · swap: total = 5120.00M used = 3702.50M free = 1417.50M (encrypted)
Live process
COMPLETE
real output model saved
Live log
[MTP-YAQA] mlp.gate_proj: {'label': 'mtp.layers.0.mlp.gate_proj', 'effective_rank_in': 1.1500806755441806, 'effective_rank_out': 1.2108201208333162, 'trials': [{'finite': True, 'weighted_err': 0.00012108818918932229, 'naive_weighted_err': 0.1725025177001953, 'beats_weighted': True, 'frob': 11.043532371520996, 'naive_frob': 10.9600830078125, 'within_frob': True, 'max_abs_hatW': 0.3464686870574951, 'max_abs_W': 0.34765625, 'within_magnitude': True, 'safe': True, 'sigma_O': 1.0}], 'path': 'default'}
mtp.layers.0.mlp.up_proj: real effective rank -- H_I=1.3/5120, H_O=1.8/17408
mtp.layers.0.mlp.up_proj/Hin: block_LDL succeeded on attempt 1, final sigma_reg=1.0000
mtp.layers.0.mlp.up_proj/Hout(sigma_O=1.0): block_LDL succeeded on attempt 1, final sigma_reg=1.0000
mtp.layers.0.mlp.up_proj: sigma_O=1e+00 weighted=0.0004x naive frob=1.011x naive SAFE=True
[MTP-YAQA] mlp.up_proj: {'label': 'mtp.layers.0.mlp.up_proj', 'effective_rank_in': 1.2617210386981141, 'effective_rank_out': 1.8151113341541463, 'trials': [{'finite': True, 'weighted_err': 0.00019944790983572602, 'naive_weighted_err': 0.4933461546897888, 'beats_weighted': True, 'frob': 12.054424285888672, 'naive_frob': 11.917745590209961, 'within_frob': True, 'max_abs_hatW': 0.36346638202667236, 'max_abs_W': 0.36328125, 'within_magnitude': True, 'safe': True, 'sigma_O': 1.0}], 'path': 'default'}
mtp.layers.0.mlp.down_proj: real effective rank -- H_I=1.3/17408, H_O=1.5/5120
mtp.layers.0.mlp.down_proj/Hin: block_LDL succeeded on attempt 1, final sigma_reg=1.0000
mtp.layers.0.mlp.down_proj/Hout(sigma_O=1.0): block_LDL succeeded on attempt 1, final sigma_reg=1.0000
mtp.layers.0.mlp.down_proj: sigma_O=1e+00 weighted=0.0003x naive frob=1.014x naive SAFE=True
[MTP-YAQA] mlp.down_proj: {'label': 'mtp.layers.0.mlp.down_proj', 'effective_rank_in': 1.3327311110731677, 'effective_rank_out': 1.5433333162443796, 'trials': [{'finite': True, 'weighted_err': 8.058748790062964e-05, 'naive_weighted_err': 0.26353752613067627, 'beats_weighted': True, 'frob': 11.539915084838867, 'naive_frob': 11.385348320007324, 'within_frob': True, 'max_abs_hatW': 0.881946325302124, 'max_abs_W': 0.8828125, 'within_magnitude': True, 'safe': True, 'sigma_O': 1.0}], 'path': 'default'}
[MTP-YAQA] Wrote real YAQA-corrected sidecar to /Users/hghelab/.mtplx/models/Qwen3.8-27B-heretic-ara-YAQA-GODMODE-v1/optiq/mtp.safetensors (29 tensors, 7 real corrected).
Part 5c PASS: real output model saved to /Users/hghelab/.mtplx/models/Qwen3.8-27B-heretic-ara-YAQA-GODMODE-v1.
[hessian-scores] 208 real tensor(s) scored -- wrote /Users/hghelab/.mtplx/models/Qwen3.8-27B-heretic-ara-YAQA-GODMODE-v1.yaqa_resume/hessian_scores.json
Real completion velocity — tensors per 15-minute window
Real tensor taxonomy — what every number below is actually counted out of: this plan covers 497 real language-model tensors in total (vision/audio/MTP sidecars are reattached separately, not counted here). Of those: 2 (0%) stay at full precision (bits=16, never quantized at all); 495 (100%) are real quantization targets, split into 495 tensors that go through Hessian/YAQA two-sided correction (the number the panel below tracks) and exactly 1 (n/a) handled separately because its own real output-side Hessian would need 246.7GB.
Real quality result, scoped to the 495 Hessian/YAQA-correctable real targets only — how much error YAQA actually removed vs. naive rounding
495/495
of 495 Hessian/YAQA-correctable tensors: reduces real weighted error
0
tensors where it does not (flagged, not hidden)
99.38%
avg error reduction · range 93.24% – 99.99%
472/495
raw Frobenius got worse (expected, see below)
287/287reused from seed, NOT re-measured · avg 99.6400%
208/208genuinely computed live this run · avg 99.0127%
Distribution of real per-tensor error reduction (zoomed into the real 90-100% band -- nearly everything real is in it)
99.2735%4-bit (254 tensors)
99.4955%5-bit (194 tensors)
99.3564%6-bit (29 tensors)
99.5770%8-bit (18 tensors)
Live interpretation — computed fresh from this exact refresh's real data
Right now, 495 of 495 completed tensors (100%) show a real reduction in Hessian-weighted error vs. naive rounding -- a strong, consistent real win. Most of them (495 tensors, 100%) cluster in the 90-100% reduction band. The single best real result so far is language_model.model.layers.7.self_attn.v_proj, at 99.9854% error reduction. Hybrid build breakdown: 287 of these tensors (58%) are reused as-is from the seeded prior build and were NOT re-measured this run (avg 99.6400% reduction, inherited); the other 208 were genuinely computed live during this run (avg 99.0127% reduction, 208/208 real wins -- this is the actual new signal from this build).
err_yaqa / err_naive (Metric A, what's plotted above): the real Hessian‑weighted
reconstruction error — how much the rounding error matters once weighted by real sensitivity
from a real backward pass. Lower is better; this is what YAQA actually optimizes. frob_yaqa / frob_naive (Metric B, "Frobenius got worse" above): plain, unweighted distance
from the original weight. YAQA can legitimately make this worse while making the weighted error
better — it deliberately trades raw magnitude for lower‑impact error, not a bug.
Reveal the real build sequence — final model assembly
✓Loading cached tensors
✓Verifying correction results
✓Packing the final model
✓Computing final bits-per-weight
✓Saving model to disk
✓Reattaching vision/audio sidecars
✓YAQA-correcting MTP sidecar
✓Complete
Trunk bits per weight: 6.000 -- language-model tensors only (vision sidecars stay at original precision, untouched; MTP is corrected separately with no bpw of its own logged)
Real output model saved to: /Users/hghelab/.mtplx/models/Qwen3.8-27B-heretic-ara-YAQA-GODMODE-v1