evaluation/compare_models.sh

#!/bin/bash

cd /home/ubuntu/RLVR/Absolute-Zero-Reasoner/evaluation/math_eval

# Base model (Qwen2.5-7B) 평가
echo "=== Evaluating Base Model: Qwen2.5-7B ==="
bash eval_math_nodes.sh \
    --run_name qwen25_7b_base \
    --init_model "Qwen/Qwen2.5-7B" \
    --template qwen25 \
    --tp_size 1 \
    --temperature 0 \
    --top_p 0.95 \
    --max_tokens 16000 \
    --benchmarks aime24,aime25,amc23,math500,olympiadbench,minerva_math \
    --n_sampling 1 \
    --just_wandb false \
    --seed 42

# AZR Coder 7B 평가
echo "=== Evaluating AZR Coder 7B ==="
bash eval_math_nodes.sh \
    --run_name azr_coder_7b_hf \
    --init_model "andrewzh/Absolute_Zero_Reasoner-Coder-7b" \
    --template azr \
    --tp_size 1 \
    --temperature 0 \
    --top_p 0.95 \
    --max_tokens 16000 \
    --benchmarks aime24,aime25,amc23,math500,olympiadbench,minerva_math \
    --n_sampling 1 \
    --just_wandb false \
    --seed 42

# AZR Base 7B 평가
echo "=== Evaluating AZR Base 7B ==="
bash eval_math_nodes.sh \
    --run_name azr_base_7b_hf \
    --init_model "andrewzh2/Absolute_Zero_Reasoner-Base-7b" \
    --template azr \
    --tp_size 1 \
    --temperature 0 \
    --top_p 0.95 \
    --max_tokens 16000 \
    --benchmarks aime24,aime25,amc23,math500,olympiadbench,minerva_math \
    --n_sampling 1 \
    --just_wandb false \
    --seed 42

echo "=== All evaluations completed! ==="