diff options
| author | YurenHao0426 <Blackhao0426@gmail.com> | 2026-04-07 22:52:41 -0500 |
|---|---|---|
| committer | YurenHao0426 <Blackhao0426@gmail.com> | 2026-04-07 22:52:41 -0500 |
| commit | ede7cca3e4f9048e3fc6d99077f8842e9b598ff4 (patch) | |
| tree | 720562461cfd2672a6f9c1b74bad6fc088bd6872 /results/confirmatory/clean_sparsity/synth_dfa_s2048_a1.0_L4.json | |
| parent | 4420af372024ef12b28eac21678504dd75484dca (diff) | |
Add reproducers for pitfalls 1-3 in CHECKLIST.md
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.
Diffstat (limited to 'results/confirmatory/clean_sparsity/synth_dfa_s2048_a1.0_L4.json')
0 files changed, 0 insertions, 0 deletions
