#!/bin/sh
# Lenient re-score of the base model's answers (no Beam tokens): llama.cpp on 8092, then rescore_lenient.py.
ROOT="$HOME/work/Projects/shortlist"; cd "$ROOT" || exit 1
(cd vendor/llama/llama-b11968 && LD_LIBRARY_PATH=. exec ./llama-server -m ../../models/Qwen3-0.6B-Q8_0.gguf --jinja \
  -c 4096 -np 2 -t 8 --host 127.0.0.1 --port 8092 > "$ROOT/runs/rescore-server.log" 2>&1) &
S=$!
for i in $(seq 1 60); do curl -s -m 2 http://127.0.0.1:8092/health | grep -q ok && break; sleep 2; done
.venv-train/bin/python training/rescore_lenient.py > runs/rescore.log 2>&1
kill $S