wassnameandClaude Opus 4.7 19bc3edb2e add dW magnitude/direction ablation eval ...

Constructs four variants of a trained dW and evaluates each on daily
dilemmas at coeffs {-1, 0, +1}:
  full         original (control)
  dir_only     elementwise direction preserved, all tensors rescaled
               to a common Frobenius norm (flattens per-tensor magnitude)
  mag_only     random direction per tensor, original per-tensor norm
               (preserves which layers/modules carry the load)
  random_norm  random direction + common norm (control)

Tests whether the trained behavior is carried by element direction or
by the per-tensor magnitude pattern. Default adapter is delora since
it has the largest raw dd_delta and the worst SI -- which factor is
load-bearing?

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

2026-04-28 08:31:24 +08:00

..

eval

add dW magnitude/direction ablation eval

2026-04-28 08:31:24 +08:00

__init__.py

tidy

2026-04-25 19:27:53 +08:00

_log.py

logs

2026-04-26 11:12:11 +08:00

data.py

switch training/eval axis from sycophancy to honesty

2026-04-28 06:00:03 +08:00

diff.py

T6/T7/T8 ablations + lens-search hold pending multiseed

2026-04-27 19:05:20 +08:00

replicate.py

wip

2026-04-27 09:59:06 +08:00

run_demo.py

logs

2026-04-26 11:12:11 +08:00

run_subspace.py

logs

2026-04-26 11:12:11 +08:00

run_sweep.py

wip

2026-04-27 09:59:06 +08:00

steer.py

phase 0-2: HF+PEFT pipeline, smoke, subspace alignment

2026-04-25 20:14:07 +08:00

subspace.py

wip

2026-04-27 09:59:06 +08:00

train.py

wip

2026-04-27 09:59:06 +08:00