481 Hessian/YAQA-correctable real targets — of 497 total language-model tensors (15 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 running26h 40m
Real build composition: 402 tensor(s) reused from a prior build's seeded cache · 79 computed live this run · 481 done total
System memory: 90.1 GB free · swap: total = 0.00M used = 0.00M free = 0.00M (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': 1.1500230381418677, 'effective_rank_out': 1.2106892607651596, 'trials': [{'finite': True, 'weighted_err': 0.00012122821499360725, 'naive_weighted_err': 0.1716715395450592, 'beats_weighted': True, 'frob': 11.043546676635742, 'naive_frob': 10.9600830078125, 'within_frob': True, 'max_abs_hatW': 0.3472197949886322, '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': 1.2617075160295539, 'effective_rank_out': 1.8149694803508254, 'trials': [{'finite': True, 'weighted_err': 0.00018970253586303443, 'naive_weighted_err': 0.4978216290473938, 'beats_weighted': True, 'frob': 12.054434776306152, 'naive_frob': 11.917745590209961, 'within_frob': True, 'max_abs_hatW': 0.3634834289550781, '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': 1.3327087778515567, 'effective_rank_out': 1.543025686521053, 'trials': [{'finite': True, 'weighted_err': 7.983627438079566e-05, 'naive_weighted_err': 0.2647320032119751, 'beats_weighted': True, 'frob': 11.540105819702148, 'naive_frob': 11.385348320007324, 'within_frob': True, 'max_abs_hatW': 0.8817949295043945, '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-v2-lmhead6-boundary16/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-v2-lmhead6-boundary16.
[hessian-scores] 82 real tensor(s) scored -- wrote /Users/hghelab/.mtplx/models/Qwen3.8-27B-heretic-ara-YAQA-GODMODE-v2-lmhead6-boundary16.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: 15 (3%) stay at full precision (bits=16, never quantized at all); 482 (97%) are real quantization targets, split into 481 tensors that go through Hessian/YAQA two-sided correction (the number the panel below tracks) and exactly 1 (lm_head (6-bit, native fallback, uncorrected)) handled separately because its own real output-side Hessian would need 246.7GB.
Real quality result, scoped to the 481 Hessian/YAQA-correctable real targets only — how much error YAQA actually removed vs. naive rounding
481/481
of 481 Hessian/YAQA-correctable tensors: reduces real weighted error
0
tensors where it does not (flagged, not hidden)
99.36%
avg error reduction · range 92.83% – 99.99%
457/481
raw Frobenius got worse (expected, see below)
402/402reused from seed, NOT re-measured · avg 99.3921%
79/79genuinely computed live this run · avg 99.2011%
Distribution of real per-tensor error reduction (zoomed into the real 90-100% band -- nearly everything real is in it)
99.1661%4-bit (165 tensors)
99.5094%5-bit (258 tensors)
99.1330%6-bit (43 tensors)
99.5966%8-bit (15 tensors)
Live interpretation — computed fresh from this exact refresh's real data
Right now, 481 of 481 completed tensors (100%) show a real reduction in Hessian-weighted error vs. naive rounding -- a strong, consistent real win. Most of them (481 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: 402 of these tensors (84%) are reused as-is from the seeded prior build and were NOT re-measured this run (avg 99.3921% reduction, inherited); the other 79 were genuinely computed live during this run (avg 99.2011% reduction, 79/79 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: 6.010 -- 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-v2-lmhead6-boundary16