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

Browse Source

Rip Axolotl/vLLM, switch to HF+PEFT functional pipeline.
Add LoRA/DoRA/PiSSA/DeLoRA train, delta-W diff, weight_steer hook,
sycophancy logratio eval, and SVD top-k + weak-readout alignment.
Smoke runs end-to-end on tiny-random qwen3 with BEARTYPE=1.

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

...

This commit is contained in:

wassname

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

parent 4ad6971038

commit 7527688a40

17 changed files with 4117 additions and 57 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

uv.lock

Generated

+3132

File diff suppressed because it is too large Load Diff