item 22 pre-registered: E2-L rung B internalization ladder (front-first deletion, 10 pauses/step, warm-started d=1..3); trainer gains --drop-steps/--warm-start/--steps/--lr; smoke-tested 2 steps on Spark

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Nils
2026-07-16 10:13:01 +02:00
co-authored by Claude Fable 5
parent 272a7b1d1f
commit 20f7014f5b
3 changed files with 85 additions and 4 deletions
+13
View File
@@ -0,0 +1,13 @@
# gpuq-in: results-loop/star_data.json results-loop/gsm_cot_data.json results-loop/adapter_carrycot_e400.pt
# gpuq-out: results-loop/eval_gsm_carrycot_b*.json results-loop/train_carrycot_b*_log.json results-loop/adapter_carrycot_b*.pt
git pull origin main -q 2>/dev/null
P=/home/nils/jspace/.venv/bin/python
export JLENS_MODEL=google/gemma-4-E2B-it LOOP_OUT=/home/nils/jspace/results-loop
cd /home/nils/jspace/scripts
WARM=$LOOP_OUT/adapter_carrycot_e400.pt
for D in 1 2 3; do
$P train_carry_cot.py --drop-steps $D --warm-start $WARM --steps 200 --lr 3e-4
WARM=$LOOP_OUT/adapter_carrycot_b${D}_e200.pt
$P eval_carry_cot.py --adapter $WARM --tag gsm_carrycot_b$D \
--grid 0:0,2:$((2+10*D)),2:$((6+10*D)) --n 256
done