scripts/run21.sh

#!/bin/bash
# run21.sh - launch the v21 shared-pool PEER pretrain (resumable).
cd /root/dna
set -a; . ./.env; set +a
export HF_TOKEN=$HUGGING_FACE
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
export TORCHINDUCTOR_COMPILE_THREADS=4
export MUON_EVERY=2 NS_STEPS=3 MUT_EVERY=16
if ! pgrep -f "python stream_data.py" >/dev/null; then
  setsid nohup python stream_data.py >> stream.log 2>&1 < /dev/null &
fi
exec python train_peer_v21.py \
  --batch 20 --seq 512 --nk 704 --topk 8 --pheads 1 \
  --target-tokens 3e9 --max-hours "${MAXH:-20}" \
  --hf-repo jaivial/dna-diskchat-2b-peer-v21 \
  --out /root/dna/ckpt-peer21 --resume /root/dna/ckpt-peer21/resume.pt