code/training/scripts/remote/resume.sh

#!/usr/bin/env bash
# Resume on a fresh GPU: pull code + data + latest checkpoint from your private HF repos.
# Usage: resume.sh <hf_token> [checkpoint_subdir]   e.g. resume.sh hf_xxx checkpoint-800
set -uo pipefail
export HF_TOKEN="${1:?usage: resume.sh <hf_token> [checkpoint]}"
CKPT="${2:-}"
export HF_HOME=/workspace/hf-cache HF_HUB_CACHE=/workspace/hf-cache/hub PIP_BREAK_SYSTEM_PACKAGES=1
mkdir -p /workspace/infosec/data/processed /workspace/infosec/checkpoints
cd /workspace/infosec
echo "== code from HF (adhikjoshi/infosec-v1:code/) =="
hf download adhikjoshi/infosec-v1 --include "code/**" --repo-type model --local-dir /workspace/_codedl 2>&1 | tail -1
cp -r /workspace/_codedl/code/* /workspace/infosec/ 2>/dev/null || true
echo "== v2 training data from HF =="
hf download adhikjoshi/infosec-training-data v2/v2_train.chatml.jsonl v2/v2_val.chatml.jsonl --repo-type dataset --local-dir /workspace/_datadl 2>&1 | tail -1
cp /workspace/_datadl/v2/*.jsonl /workspace/infosec/data/processed/ 2>/dev/null || true
if [ -n "$CKPT" ]; then
  echo "== checkpoint $CKPT from HF =="
  hf download adhikjoshi/infosec-v1 --include "$CKPT/**" --repo-type model --local-dir /workspace/infosec/checkpoints 2>&1 | tail -1
fi
echo "resume complete: code + data ${CKPT:+(+ $CKPT)} pulled. Next: bash training/scripts/remote/bootstrap.sh"