diff options
| author | Yuren Hao <yurenh2@illinois.edu> | 2026-08-04 05:46:57 -0500 |
|---|---|---|
| committer | Yuren Hao <yurenh2@illinois.edu> | 2026-08-04 05:46:57 -0500 |
| commit | f3ade7674328c55a98d32fd332a3a1cf79fd40a6 (patch) | |
| tree | 789515f125d31b894f147f61f1b1bab287a6dcc3 /ep_run/grad_env_probe.py | |
| parent | 3962588a667d0d8945044b105dedda75e3fa3a41 (diff) | |
RESULT 89终局 + 交付物: 135M完赛 EP 3.2087 vs BP 3.2074(+0.1% ppl), 旧读出+45.2%
窗口敏感性诚实记录: 97%窗给-0.0030(seed间距0.0059), 完赛大窗给+0.0013(间距0.0005, cosine
衰减后BP收拢); 以大窗为准且不得称"统计不可分"(2.6倍间距), 正确表述=差0.1%且处分辨极限,
seed数(BP n=2/EP n=1)不足以给区间。
交付: --gen模式(复用模型定义, EP/BP同代码路径)、fig_135m_curves(四曲线/横轴token/BP细线在上)、
EPT_135M_samples.ipynb(同prompt同种子并排, 输出已烤入, 代码真可跑)。
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
Diffstat (limited to 'ep_run/grad_env_probe.py')
0 files changed, 0 insertions, 0 deletions
