jieyibi / amdkd Goto Github PK

Hi there 👋🏼

Great job with your paper and code implementation of AMDKD~

Dropping by since we are actively developing RL4CO, a library for all things Reinforcement Learning for Combinatorial Optimization. The idea is to make everything as easy to use and modular as possible in the hope of helping practitioners and researchers with the help of some recent software (such as TorchRL, TensorDict, PyTorch Lightning, FlashAttention, and Hydra).

We did not have enough time to include your work in the library so far, but believe that AMDKD would be fairly easy to integrate and it would also contribute in terms of generalization studies. If you are interested in contributing/collaborating, feel free to leave PRs/issues or contact us anytime 😉

Thank you for your insightful work and for releasing the code. I am wondering how to implement AMDKD+EAS with your code? Could you please share the source code for AMDKD+EAS?

About May I ask how to solve a large number of Inf problems in both student_logp and teacher_logp during model training?

Thanks for your insightful work. I would like to ask if 【Table 11: Full generalization results on CVRPLIB】 was obtained using a student model trained on the cvrp-100 and decoded by [sampling-1280]?