Inquiry about Experimental Setup and Runtime

[yw3l]

Dear @bosung
First of all, thank you for sharing your excellent work on MTL-KGC. I am particularly interested in reproducing the results presented in this paper. To ensure that I accurately replicate the experiments, I would like to ask for some details regarding the experimental setup:
Hardware Configuration: Could you please share the details of the hardware configuration used for running the experiments? Specifically, information on the GPU, would be very helpful.
Runtime: I would also appreciate it if you could give an estimate of the time it took to run the main experiments. This will help me plan my experiments more effectively.
Best regards,
yw3l

[bosung]

Hi,
I used GTX 1080s when I published the paper (which was 4 years ago), but any GPUs can be used to run experiments.
For the runtime, training doesn't take that long ~3hrs for a single task and ~8hrs for multi-task. However, evaluation takes very long time; it used to take over 3 days for one Link Prediction evaluation on the WN18RR dataset. But I guess it will take less with more advanced recent GPUs. When I tried with one A10G, it takes around 27 hours. Also this code base is really outdated, you might need to update PyTorch version and speed up with Huggingface's Accelerate. Hope this would help!

[yw3l]

Thank you for your quick response and helpful suggestions! I will follow your advice and try it out. I really appreciate your assistance!