- python == 3.11.5
- torch == 2.1.2
- numpy == 1.26.3
- pandas == 2.1.4
- tqdm == 4.65.0
- python-chess == 1.10.0
Modify the data path in
./code/fetch_data.sh
and run:
sh ./code/fetch_data.sh
We train our model with 2 Nvidia A100 GPUs and 48 AMD EPYC 7V13 CPUs. The data preprocessing is integrated in the training process. To train Maia-2, simply run:
cd ./code
nohup python main.py --verbose 0 --data_root your_data_root >your_log_file.log 2>&1 &
Download the model from here. To reproduce the main results, simply run:
cd ./code
python inference.py --model_path your_model_path