ASCEND_RT_VISIBLE_DEVICES=4 python main.py --mode eval \
  --device npu \
  --model_name FreedomIntelligence/openPangu-Embedded-7B \
  --test_data processed_data/AMC/test.jsonl \
  --load_checkpoint checkpoints/MATH/pangu/best_checkpoint.pt \
  --task_type math \
  --results_dir results/AMC/ \
  --token_budget 4096 \