#!/usr/bin/env bash # Full eval battery for the 50M line: main + researcher + red-team. # Usage: ./stage_eval_50m.sh [ckpt] (default: latest in ckpt/hybrid50m_lora) set -euo pipefail cd "$(dirname "$0")" export PYTHONPATH=$PWD export MALLOC_ARENA_MAX=2 SRC="${1:-ckpt/hybrid50m_lora}" EVAL_LOG="logs/eval_50m_$(date +%Y%m%d_%H%M).log" echo "== main battery ==" > "$EVAL_LOG" .venv/bin/python research/eval.py --ckpt "$SRC" --tok data/tokenizer16k.json --threads 6 --probes data/eval_probes.jsonl >> "$EVAL_LOG" 2>&1 echo "== researcher battery ==" >> "$EVAL_LOG" .venv/bin/python research/eval.py --ckpt "$SRC" --tok data/tokenizer16k.json --threads 6 --probes data/probes_researcher.jsonl >> "$EVAL_LOG" 2>&1 echo "== red-team battery ==" >> "$EVAL_LOG" .venv/bin/python research/eval.py --ckpt "$SRC" --tok data/tokenizer16k.json --threads 6 --probes data/eval_redteam.jsonl >> "$EVAL_LOG" 2>&1 echo "saved $EVAL_LOG" tail -6 "$EVAL_LOG"