diff options
| author | Yuren Hao <yurenh2@illinois.edu> | 2026-07-03 18:21:21 -0500 |
|---|---|---|
| committer | Yuren Hao <yurenh2@illinois.edu> | 2026-07-03 18:21:21 -0500 |
| commit | 6e78420da6e613964d93da06156b556e1a91caef (patch) | |
| tree | 9329b0bc134fbc0627a80e6fd095651fcf9e4975 /ep_run/floss_smoke.log | |
| parent | bcec9560cf5c9b113e9381a52d1a941daa8865f2 (diff) | |
floss-ept: graded finite-horizon LE penalty (--floss) + three-arm from-scratch ablation queueHEADmaster
- ep_step: floss block after resreg — unroll q=10 steps past z_T1 on a
sub-batch WITH graph, rho_hat = mean per-step delta growth, one-sided
relu(rho_hat - 0.995)^2, ramp keyed on (rho_hat - target) NOT resT1
(de-cliffed resreg: same fundamental path-LE quantity, linear early signal),
capped at floss fraction of task-grad norm (resreg convention).
- smoke: below-target = untouched (cos 1.0000); force-fire = finite grads,
capped perturbation (cos 0.9803).
- runs/abl3_queue.sh (runner live): waits for free GPU slots (0/1/3, GPU2
excluded), launches abl_floss (floss-only) / abl_resreg (resreg-only, never
cleanly run) / abl_pair (proven 2.09 recipe, control) with identical
remaining flags + seed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
Diffstat (limited to 'ep_run/floss_smoke.log')
| -rw-r--r-- | ep_run/floss_smoke.log | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/ep_run/floss_smoke.log b/ep_run/floss_smoke.log new file mode 100644 index 0000000..a476b9f --- /dev/null +++ b/ep_run/floss_smoke.log @@ -0,0 +1,8 @@ +[off ] res=5.006e-02 n_grads=18 +/home/yurenh2/ept/ep_run/lt_ep_train.py:248: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior. +Consider using tensor.detach() first. (Triggered internally at /pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:836.) + gf = torch.autograd.grad(Rf, blk.block, allow_unused=True) if float(Rf) > 0 else None +[floss] res=5.006e-02 rho_hat=0.9919 finite=True n_grads=18 +cos(g_off, g_floss)=1.0000 (below-target: should be 1.0 = untouched) +[fire ] rho_hat=0.9919 finite=True cos(g_off, g_fire)=0.9803 (should be <1 but >0.9: capped perturbation) +FLOSS_SMOKE_DONE |