diff options
| author | YurenHao0426 <Blackhao0426@gmail.com> | 2026-04-01 12:56:24 -0500 |
|---|---|---|
| committer | YurenHao0426 <Blackhao0426@gmail.com> | 2026-04-01 12:56:24 -0500 |
| commit | d5326053a2e9ce37dd61606aa37fa8f563481f44 (patch) | |
| tree | 653f8bf3098d382a1162c09ce4983d9d1c50713e /results/confirmatory/clean_grads/credit_bridge_s42.json | |
| parent | cd80da41c620d7c8b17e36d3ed7ab7e6b582f191 (diff) | |
Add clean gradient check: independent Python process per method, GPU 1
Clean results (each method in fresh Python process):
BP: mean_norm=2.58e-04, s(1e-6)=98% — CONFIRMED
DFA: layer 0 = 2.86e-07 (1.2%), layers 1-3 ≈ 2.4e-09 (0%)
SB: layer 0 = 6.13e-06 (86%), layers 1-3 ≈ 1e-09 (0%)
CB: layer 0 = 6.33e-07 (18%), layers 1-3 ≈ 5e-10 (0%)
Method A (autograd.grad) and Method B (retain_grad) give identical results.
Previous 1e-12 results were caused by Python process state pollution in combined scripts.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Diffstat (limited to 'results/confirmatory/clean_grads/credit_bridge_s42.json')
| -rw-r--r-- | results/confirmatory/clean_grads/credit_bridge_s42.json | 74 |
1 files changed, 74 insertions, 0 deletions
diff --git a/results/confirmatory/clean_grads/credit_bridge_s42.json b/results/confirmatory/clean_grads/credit_bridge_s42.json new file mode 100644 index 0000000..e0760a6 --- /dev/null +++ b/results/confirmatory/clean_grads/credit_bridge_s42.json @@ -0,0 +1,74 @@ +{ + "method": "credit_bridge", + "seed": 42, + "batch_size": 256, + "y_first5": [ + 3, + 8, + 8, + 0, + 6 + ], + "first_param_norm": 1510.4136962890625, + "method_A": { + "loss": 2.001465320587158, + "acc": 0.25390625, + "per_layer": [ + { + "mean_norm": 1.1000469402034328e-14, + "median_norm": 1.1000469402034328e-14, + "max_norm": 1.1000469402034328e-14, + "s_1e6": 0.0 + }, + { + "mean_norm": 4.4292934381258566e-17, + "median_norm": 4.4292934381258566e-17, + "max_norm": 4.4292934381258566e-17, + "s_1e6": 0.0 + }, + { + "mean_norm": 2.1339835398853298e-17, + "median_norm": 2.1339835398853298e-17, + "max_norm": 2.1339835398853298e-17, + "s_1e6": 0.0 + }, + { + "mean_norm": 2.4727265738711054e-17, + "median_norm": 2.4727265738711054e-17, + "max_norm": 2.4727265738711054e-17, + "s_1e6": 0.0 + } + ] + }, + "method_B": { + "loss": 2.001465320587158, + "acc": 0.25390625, + "per_layer": [ + { + "mean_norm": 1.1000469402034328e-14, + "median_norm": 1.1000469402034328e-14, + "max_norm": 1.1000469402034328e-14, + "s_1e6": 0.0 + }, + { + "mean_norm": 4.4292934381258566e-17, + "median_norm": 4.4292934381258566e-17, + "max_norm": 4.4292934381258566e-17, + "s_1e6": 0.0 + }, + { + "mean_norm": 2.1339835398853298e-17, + "median_norm": 2.1339835398853298e-17, + "max_norm": 2.1339835398853298e-17, + "s_1e6": 0.0 + }, + { + "mean_norm": 2.4727265738711054e-17, + "median_norm": 2.4727265738711054e-17, + "max_norm": 2.4727265738711054e-17, + "s_1e6": 0.0 + } + ] + }, + "method_C_embed_grad_norm": 0.0007018167525529861 +}
\ No newline at end of file |