#!/usr/bin/env bash
set -euo pipefail
MODEL_DIR="${1:-/tmp/bitnet-work/train/best_model}"
OUT="${2:-/home/user/bitnet-tinydialogues/gguf}"
LLAMA_CPP="${LLAMA_CPP:-/tmp/bitnet-work/llama.cpp}"
BUILD="${BUILD:-/tmp/bitnet-work/llama-build}"
mkdir -p "$OUT"
python "$LLAMA_CPP/convert_hf_to_gguf.py" "$MODEL_DIR" --outfile "$OUT/bitnet-tinydialogues-f32.gguf" --outtype f32
"$BUILD/bin/llama-quantize" --token-embedding-type Q8_0 "$OUT/bitnet-tinydialogues-f32.gguf" "$OUT/bitnet-tinydialogues-TQ1_0.gguf" TQ1_0 2
"$BUILD/bin/llama-quantize" --token-embedding-type Q8_0 "$OUT/bitnet-tinydialogues-f32.gguf" "$OUT/bitnet-tinydialogues-TQ2_0.gguf" TQ2_0 2
ls -lh "$OUT"