results: QLoRA bs=3 ga=2 + lam_round_pow=-0.5 extends movement to r6 (peak -0.37 vs -0.60)

Browse Source

- plot: Panel A now tracks top-moving trait (care for love demo, auth for authority)
  instead of hardcoded auth_nats; Panel C already did this, Panel A now consistent
- README: update table with new run (lam decay extends saturation r4→r6), refresh diary
  from new run's outputs, update trajectory plot
- AGENTS.md: correct gotchas -- tau<operating_KL is the key constraint (tau=2.0 not 4.0);
  QLoRA + bs=3 ga=2 is the right default for better heal gradient estimates

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>

...

This commit is contained in:

wassname

2026-06-10 07:36:09 +08:00

parent 2b884c2fb9

commit 09349894ce

4 changed files with 64 additions and 49 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

docs/love_loop.png

BIN

Binary file not shown.

Side by Side Swipe Overlay Before Width:  |  Height:  |  Size: 248 KiB After Width:  |  Height:  |  Size: 180 KiB