This is the official code repo for Paper Importance Weighting can Help Large Language Models Self-Improve in AAAI 2025. For the version with full supplementary materials, please refer to https://arxiv.org/abs/2408.09849
To let LLM self-generate samples, use the script generate_sample_parallel.sh
To compute DS weights, use the script compute_weights.sh
To start training, use the script baseline.sh.
To evaluate the model, use the script evaluation.sh.
We would recommend that you refer to the arxiv version for more details.