IWSI

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

Self-Generate samples

To let LLM self-generate samples, use the script generate_sample_parallel.sh

Compute DS weights

To compute DS weights, use the script compute_weights.sh

Train

To start training, use the script baseline.sh.

Evaluation

To evaluate the model, use the script evaluation.sh.

We would recommend that you refer to the arxiv version for more details.