#!/usr/bin/env bash
# Reproduces the benchmark table in the README (zero-shot, lm-evaluation-harness).
set -euo pipefail
pip install -q lm-eval
MODEL=${1:-IvmeLabs/Ivme-Conversate-P-v1}
python -m lm_eval --model hf \
  --model_args pretrained=$MODEL,trust_remote_code=True,dtype=float32 \
  --tasks wikitext,blimp,arc_easy,arc_challenge,hellaswag,piqa \
  --batch_size 64 --output_path eval-results