| Age | Commit message (Collapse) | Author |
|---|
|
5 methods × 3 seeds on the SmallCNN (3 conv + BN + 1 FC + head, no
terminal LN) using existing checkpoints in results/cnn_baseline/.
Key findings:
BP CNN: 0.866 acc, max/block 1.3, trustworthy
State Bridge CNN: 0.633 acc, max/block 2.4, trustworthy
EP CNN: 0.512 acc, max/block 12, trustworthy
DFA CNN: 0.566 acc, max/block 237, walked back via (a)
Credit Bridge CNN: 0.325 acc, max/block 96, walked back via (a)
CRITICAL: diagnostic (b) ||g_L|| floor NEVER fires on CNN for any method.
The deepest BP grad is at ~1e-5 to 6e-1, all well above the 1e-7 floor.
This is the cleanest confirmation that terminal LayerNorm is the
structural cause of the catastrophic gradient collapse in (b). Without
out_ln, the BP grad does NOT collapse to the floor, even on DFA. The
scale pathology (a) still appears on DFA and CB, but the gradient
collapse pathology (b) is specific to terminal-LN architectures.
DFA CNN's accuracy (56.6%) is much higher than DFA ResMLP (30.8%) or
DFA ViT (23.7%) — partially because the scale pathology is less
catastrophic without the LN-driven gradient cancellation amplifying
it. This is the cross-architecture mechanism story made concrete.
|
|
5-epoch DFA training on CIFAR-10 + apply protocol + interpret verdict.
Self-contained, runs on CPU in <2 minutes. Demonstrates the API a future
paper author would use:
1. train your model (any FA-style method)
2. build eval_batches from your test loader
3. call diagnose(model, eval_batches, headline_acc, frozen_baseline_acc)
4. read report.verdict; walk back if 'needs walk-back'
Not run during this session to avoid GPU contention with the in-flight
direction-quality and ViT/ResNet experiments.
|
|
3-panel side-by-side showing per-epoch trajectories of vanilla DFA vs
DFA + lambda*||f||^2 penalty:
(a) ||h_L||: vanilla 4e8 vs penalty 4e4 (4 OOM rescue)
(b) ||g_L||: vanilla 5e-10 vs penalty ~1e-6 (4 OOM rescue)
(d) test acc: vanilla 0.31 vs penalty 0.36 vs frozen baseline 0.349 vs BP 0.61
The visual story: (a) and (b) show the penalty pulling the diagnostics
back into the healthy regime, but (d) shows the rescue translates to
only +1 pp above the DFA-shallow baseline and 24 pp below BP-trainable.
The two failure modes (scale + direction) are visually separable: scale
is fixed, direction is not.
Together with figure_audit_5method.png and figure_cross_arch_temporal_s42.png,
this is the third paper-ready figure for §3-§4.
|
|
4-panel layout (one per diagnostic), 5 methods sorted bottom-to-top by
ascending accuracy, color-coded healthy (BP/EP, blue) vs degenerate
(DFA/SB/CB, red), with threshold lines drawn:
(a) max per-block growth (log scale, threshold 50x)
(b) ||g_L|| (log scale, floor 1e-7)
(c) cross-batch stability (linear, ceiling 0.30)
(d) headline acc (linear, frozen baseline 0.349)
The visual layout makes it immediately obvious that:
- (a) and (b) cleanly split healthy from degenerate (4-7 OOM gap)
- (c) is bimodal and doesn't cleanly split — confirms it's a sub-mode
discriminator, not a primary detector
- (d) shows BP above the frozen baseline by ~25 pp while DFA/CB/SB
are at or below it
|
|
Same protocol applied to the 4-block d=512 ResMLP variant (vs the d=256
default). 4 methods × 3 seeds = 12 conditions:
BP @ d=512: trustworthy on all 3 seeds (acc 0.60-0.61)
DFA @ d=512: walked back on all 3 seeds via (a)+(b)
State Bridge @ d=512: walked back on all 3 seeds via (a)+(b), with
drift sub-mode on s123 (stability 0.879)
Credit Bridge @ d=512: walked back on all 3 seeds via (a)+(b)
Width effect: max-per-block growth is HIGHER at d=512 (6e3-7e4) than at
d=256 (~1e3). Larger width amplifies the explosion. The protocol
verdicts are robust to this — same binary outcome, more extreme
quantitative numbers.
This is the cross-width validation: the protocol's findings are not
d=256-specific. The §3 audit results generalize across the width
dimension.
|
|
3-seed analysis of DFA + lambda=1e-2 ||f||^2 penalty using only the data
already in the existing penalty JSON logs (no checkpoint or full layer
norms needed):
(a) per-block growth: avg ~8x per block (geom mean), well below 50x
threshold. PASS likely (with small caveat that max could differ
from mean).
(b) BP grad floor: g_2 = 8-10e-7 across 3 seeds, 10x above the
1e-7 floor. PASS exact.
(d) frozen baseline: margin = 1.35-1.45 pp (mean 1.38) < 2 pp
required. FIRE on all 3 seeds.
Aggregate partial verdict: protocol catches the SECOND failure mode
(direction quality / passive blocks) on penalized DFA even though it
PASSES the scale-related diagnostics. This is the cleanest possible
evidence that the two failure modes are separable: the penalty fixes
the scale failure but not the direction failure. The protocol's (d)
diagnostic is the right test for the second failure mode and it still
fires after the penalty rescue.
This is the §4 'two failure modes' evidence that doesn't depend on the
direction-quality direct test (which is still running). The (d)
diagnostic alone shows the separation.
|
|
3-column 3-row plot:
rows: ||h_L||, ||g_L||, test accuracy
cols: ResMLP (with LN) | ViT-Mini (cls + LN) | StudentNet (no LN)
BP and DFA trajectories overlaid. Floor threshold drawn on the ||g_L||
row. Visualizes the cross-architecture causal control: with-LN
architectures both show ||g_L|| collapse below 1e-7 (DFA hits the floor
within 5 epochs); without-LN architecture shows ||g_L|| stays in the
healthy regime even though ||h_L|| still grows (catastrophic vs mild).
|
|
For each diagnostic, sweeps threshold across orders of magnitude on the
3-seed audit data and reports the verdict at each value.
Key calibration findings (3 seeds):
Diagnostic (a) max per-block growth:
Healthy max (BP/EP): 11.0
Degenerate min (DFA/SB/CB): 694
Separation gap: 63x
Default threshold 50 sits comfortably in the middle.
Any threshold in [12, 693] gives the same verdicts.
Diagnostic (b) ||g_L|| at floor:
Healthy min (BP/EP): 1.02e-4
Degenerate max (DFA/SB/CB): 4.18e-9
Separation gap: 24,338x
Default threshold 1e-7 sits comfortably in the middle.
Any threshold in [4.2e-9, 1.0e-4] gives the same verdicts.
Diagnostic (c) cross-batch stability:
NOT a clean binary discriminator across seeds. BP s456=0.114
near threshold; DFA s42=0.047 (noise sub-mode) doesn't fire;
SB s456=0.035 (noise sub-mode) doesn't fire. (c) is for sub-mode
interpretation, not binary detection.
This is the calibration evidence answering the E&D reviewer question
'why these specific thresholds?'.
|
|
All 3 verified on the real DFA s42 checkpoint:
Bug 4: training Bs gives Γ=+0.068, 10 fresh Bs draws give Γ=+0.0043±0.007.
The 'alignment' is the network adapting to specific Bs.
Bug 5: 4 valid aggregation strategies give Γ in [-0.028, +0.074]. The
spread is 0.10 (3.45x ratio) and **the sign flips** between
strategies. Pick the wrong aggregation and DFA is anti-aligned;
pick the right one and DFA looks aligned.
Bug 6: Γ_layer0 = +0.429 dominates the mean +0.068. Hidden layers 1-4 are
all near zero or slightly negative. Mean of hidden layers only is
-0.022 (negative!). The deep blocks the paper claims to be
'training' have Γ ≈ 0 or below.
Bugs 5 and 6 are causally linked: 'median over layers' strategies pick a
negative deep layer; 'mean over layers' is dominated by the positive l0.
The catalog under-reported bug 5 (it said 2.5x, actual is 3.45x with sign
flip).
|
|
Demonstrates the practical use case of the protocol — not as a post-hoc
audit but as an in-training abort condition. Walks through the existing
per-epoch trace and shows when the protocol would have triggered an early
stop on DFA training and what the saved compute would be.
Result: DFA on 4-block d=256 ResMLP fires diagnostic (b) at epoch 4 with
test acc 0.3076. The final acc at epoch 100 is *also* 0.3076 (identical).
Stopping at epoch 4 saves 96% of compute with zero headline acc loss.
|
|
ResMLP (4-block d=256, with out_ln, CIFAR-10):
s42: DFA (a) ep 8, (b) ep 4, acc 0.308
s123: DFA (a) ep 11, (b) ep 4, acc 0.320
s456: DFA (a) ep 8, (b) ep 3, acc 0.300
ViT-Mini (4-block d=128, cls token + terminal LN, CIFAR-10):
s42: DFA (a) ep 1, (b) ep 3, acc 0.256
s123: DFA (a) ep 1, (b) ep 2, acc 0.202
s456: DFA (a) ep 1, (b) ep 3, acc 0.253
StudentNet (4-block d=128, NO terminal LN, synthetic alpha=1.0):
s42: DFA (a) ep 18, (b) NEVER, acc 0.332
s123: DFA (a) ep 14, (b) NEVER, acc 0.314
s456: DFA (a) ep 25, (b) NEVER, acc 0.336
BP: never fires on any seed x any architecture (9/9 sanity passes).
Key cross-architecture finding: diagnostic (b) is specifically the LN-
driven failure mode. Without out_ln, the BP grad never crosses the 1e-7
floor, even though (a) still fires (the residual stream still grows, just
without the LN-cancellation pathology that drives the BP grad to the
floor). This is the causal architectural control: (b) specifically tests
'is terminal-LN gradient cancellation active?' and (a) tests 'is the
residual stream growing without bound?'. They are linked but separable.
This is the §3 cross-architecture validation evidence.
|
|
Old metric: max(||h||) / max(||h_0||, eps). False-positives on ViT-style
architectures because the cls token at layer 0 (right after patch_embed)
has anomalously small magnitude (~0.3-1.5), inflating the ratio even on
healthy BP-trained ViTs.
New metric: max_l(||h_{l+1}|| / ||h_l||) — the largest single-block
residual amplification. Architecture-invariant.
Calibration:
- BP-trained, late training: <5x per block
- BP ViT, early epochs (cls token resolving): 13-25x max
- DFA-trained ResMLP/ViT: 100-4000x per block
Threshold raised from 10 to 50 to sit cleanly between healthy-early-
training (max 25) and failure-regime (min 100).
Re-verifications:
- smoke test (BP/DFA/EP): all 3 verdicts unchanged
- random init (3 seeds): trustworthy on all 3
- 5-method audit table single-seed: identical verdicts
- decision-utility ablation: identical (still 0/5 by S1, 3/5 by S_full)
- temporal evolution 3-seed: (b) now fires first at ep 3-4, (a) at ep
8-11. Both well before training ends. The 'protocol fires ~92 epochs
early' story still holds.
- ViT temporal evolution: BP no longer false-fires; DFA fires (a) ep 1,
(b) ep 3 — protocol works on the second architecture.
|
|
Each bug from the catalog has a synthetic reproducer that runs in <1 sec
without GPU:
Bug 1: x.norm(-1) on a 2x2 tensor returns 1.143 (L_{-1} of whole tensor)
instead of [5, 10] (per-row L_2 along dim=-1).
Bug 2: F.cosine_similarity(a, b) with ||b||=5e-10 returns +0.000905
instead of the true +0.018101. The clamp (eps=1e-8) underestimates
the divisor 20x.
Bug 3: 5e-10 in fp16 -> 0 (underflows smallest subnormal ~6e-8).
Downstream F.cosine_similarity returns NaN. bf16 works because it
shares fp32's exponent range.
Bugs 4-6 (Bs reproducibility, aggregation, layer-0 dominance) require a
trained network and are demonstrated inside audit_table and
ablation_decision_utility.
|
|
s42: (a)+(b) fire at epoch 4, DFA final acc 0.3076
s123: (a)+(b) fire at epoch 4, DFA final acc 0.3203
s456: (a)+(b) fire at epoch 3, DFA final acc 0.2998
BP never fires on any seed (final acc 0.61-0.63).
The 'protocol catches it 96 epochs early' finding is fully reproducible
across seeds.
|
|
Replays per-epoch logged data from results/snapshot_evolution_v2/ through
the protocol thresholds.
Result: diagnostics (a) ||h_l|| explosion AND (b) ||g_L|| at floor BOTH
first fire at epoch 4 of DFA training. At that point, DFA test acc is
0.308 — its final value at epoch 100 is also 0.308. The protocol could
have walked back the headline 96 epochs before training finished.
DFA's gamma hovers at 0.087-0.107 for all 100 epochs. A reviewer looking
at acc+gamma would conclude 'DFA is hovering at 31% acc with ~0.10
alignment, both reasonable'. Wrong on both counts.
BP never fires any diagnostic at any epoch. Stays bounded at ||h_L||~200,
||g_L||~3-5e-5, accuracy climbs to 0.61.
This is the temporal validation of decision utility: the protocol catches
the pathology AS IT HAPPENS, not just retrospectively.
|
|
3-seed random init ResMLP gives chance accuracy (~10%) but the protocol
verdict is 'trustworthy' on all 3 seeds:
- residual norms ~8.7 across all layers (no growth, bounded)
- BP gradient norms ~8e-3 (healthy, well above 1e-7 floor)
- cross-batch stability 0.08-0.18 (in the BP/EP range)
This is the answer to the likely reviewer question: 'is your protocol just
flagging anything that doesn't perform well?' Answer: no. Random init is
at chance and the protocol passes it. The walked-back trained methods are
walked back because of the *measurements*, not because of the accuracy.
Notable: random init g-norms (8e-3) are actually HIGHER than BP-trained
ones (4e-4) — BP training reduces the gradient magnitude as loss decreases.
So the protocol distinguishes 3 distinct regimes: (1) untrained healthy,
(2) trained-and-still-healthy (BP/EP), (3) trained-into-pathology (DFA/SB/CB).
|
|
3 seeds × 5 methods × 4 diagnostics = 60 measurements. Key reproducibility
findings:
- BP: trustworthy on all 3 seeds (acc 0.61-0.62, h_L ~200, g_L ~3-4e-4)
- EP: trustworthy on all 3 seeds (acc 0.29-0.36, h_L 3-8e3, g_L ~1e-4)
- DFA, SB, CB: walked back on all 3 seeds × all 3 of (a)/(b)/(d)
Diagnostic (c) is bimodal across seeds — confirms the prior memory finding:
- DFA s42=0.047 (noise), s123=0.436 (drift), s456=-0.005 (noise)
- SB s42=0.992 (drift), s123=0.561 (drift), s456=0.035 (noise)
- CB s42=0.352 (drift), s123=0.250 (~edge), s456=0.518 (drift)
(c) catches different methods on different seeds. (a)/(b)/(d) catch all 3
failing methods on all 3 seeds — robust binary detection.
|
|
Builds on the 5-method audit JSON. For each method, evaluates 7 reporting
strategies (S0=acc only, S1=+Γ field standard, S2-S5=+single diagnostic,
S_full=full protocol), and emits the verdict each strategy would have
reached.
Result: 3 of 5 methods (DFA/SB/CB) are walked back by S_full but NOT by S1.
Each of (a)scale, (b)floor, (d)frozen is independently sufficient for
binary detection of those 3 failures. Diagnostic (c)stability adds
sub-mode discrimination (drift vs noise) but not new positive detections.
This is the §3 protocol decision-utility evidence.
|
|
5-method audit table on 4-block d=256 ResMLP CIFAR-10 seed 42:
- BP: trustworthy (acc 0.615, h_L=2e2, g_L=4e-4, stab 0.099)
- DFA: walked back via (a)+(b)+(d) — h_L=4e8, g_L=4e-9, undercuts frozen
- State Bridge: walked back via all 4 diagnostics — stability 0.992 is the
cleanest possible drift-dominated case
- Credit Bridge: walked back via all 4 — stability 0.352, also drift mode
- EP: trustworthy (acc 0.359, h_L=3e3, g_L=2e-4, stab -0.036) — paper's
internal control case
This is the §2 audit evidence for the main-track paper. Confirms that
standard headline acc + Γ silently fails on 3 of 5 methods on this
architecture, while the 4-diagnostic protocol catches all three.
|
