yd-kwon / pomo Goto Github PK

Hi there 👋🏼

First of all, great job on your work, it has served as an inspiration to our research group!
We are actively developing RL4CO, a library for all things Reinforcement Learning for Combinatorial Optimization. We reimplemented POMO in the library and modularized it to be compatible with problems other than TSP and CVRP. We also used some recent software (such as TorchRL, TensorDict, PyTorch Lightning, and Hydra) as well as routines such as FlashAttention. We tried to make everything as easy to use as possible in the hope of helping practitioners and researchers.

Feel free to check RL4CO out 🚀 ^ ^

Hello.
Thank you for your contribution.
I'm running TSP training with default setting. It took 14 days already and still learning. I wonder how long time this TSP training takes. I'm using V100-DGX station with cuda 10.1.

Hi i want to ask how can i see the distance and tour

Thanks for the nice work!

However, could you please add more information about the Python environment (e.g. package versions) to the README? It would be very helpful (avoid potential compatibility issues).

Best

Hello, I ran the train_n100.py file directly, why the loss is negative

I'm trying to wrap POMO as a routing solver module, so I want to know if there are pre-trained model for cvrp20 and cvrp50 in NEW_py_ver?

hello,I want to know how vrp100_test_seed1234.pt is obtained

What i need to do if i want to add time windows to this code ?