Open bywbilly opened 2 years ago
I have addressed the problem above. You need to directly run solve-20.sh instead of running generate_lib.sh first. I have one more question: when I try to solve TSP20, the program may stack at some certain instances like the 25-th data point. And some data points may get very huge like ~1e10 results. I am wondering do you have any idea why this can happen. @Spider-scnu
Hi @bywbilly, just curious, are you using the suggested CUDA version from the authors? I've been trying to get their code to work on a machine with CUDA 11.3, but I can't seem to get it to work.
Hi @bywbilly, just curious, are you using the suggested CUDA version from the authors? I've been trying to get their code to work on a machine with CUDA 11.3, but I can't seem to get it to work.
I use 9.2 and it works.
Hi @bywbilly , I have not encountered this problem during the process of experiment.
Hi @Spider-scnu, I have a problem when running the GPU version of the search. I managed to make the code work, but the GPU version of the search is very slow (and the GPU is being used as I can see in the resources usage information). In fact, it runs slower than the CPU version. Do you know why this could happen and if it can be fixed?
Moreover, I get the same mistake as @bywbilly regarding the search getting stuck on the 25-th instance.
Thank you in advance.
Thanks for your excellent work and open souring the code. I encountered one issue when trying to compile the GPU MCTS.
Here is the error when I run ./generate_lib.sh, it seems that we miss some header files here? probably curand_kernal.h?
I am wondering can you give it a look? @Spider-scnu
Thanks!