eval.sh

set -ex
# export CUDA_VISIBLE_DEVICES=0,1

############## SimpleRL ##############
# MODEL_NAME_OR_PATH=Qwen/Qwen2.5-MATH-7B
# PROMPT_TYPE="qwen25-math-cot"
MODEL_NAME_OR_PATH=FreedomIntelligence/openPangu-Embedded-7B-V1.1
PROMPT_TYPE="pangu"

DATA_NAME=${1:-"amc23,var_amc23,aime24,var_aime24,aime25,var_aime25"}
N_SAMPLES=${2:-"1"}

python3 -u math_eval.py \
        --model_name_or_path ${MODEL_NAME_OR_PATH} \
        --data_name $DATA_NAME \
        --output_dir "score_pass_${N_SAMPLES}/openPangu-Embedded-7B-V1.1" \
        --prompt_type "pangu" \
        --num_test_sample "-1" \
        --temperature 0.6 \
        --n_sampling ${N_SAMPLES} \
        --top_p 1 \
        --use_vllm \
        --max_tokens_per_call 8192 \
        --pangu_think_mode "slow" \
        --apply_chat_template \
        --save_outputs