search

xjw-nlp / SimMCS

Apache License 2.0
3 stars 0 forks source link

Computing resource consumption #2

Open forkliu0826 opened 10 months ago

forkliu0826 commented 10 months ago

Hello author, I am very interested in your article, I see that your article did not report computing resource requirements and running time, I would like to ask you about the relevant information

xjw-nlp commented 10 months ago

I'm sorry I misunderstood some information. In this work, we used four 3090 GPUs and a training epoch took about 16 hours on the CNNDM dataset and about 10 hours on the XSum dataset.

forkliu0826 commented 10 months ago

Hello author, I see that your epoch default value is 100, does it take that long to train？

xjw-nlp commented 10 months ago

In fact, it’s not necessary to train so many rounds, we only need to train a few rounds (< 10).

forkliu0826 commented 9 months ago

Sorry author, I have one more question, I would like to know what instruction is running as a result of the test in your experiment, do_reranking or do_generation or something else.

xjw-nlp commented 9 months ago

do_ranking is for the experiment used for selecting candidate summaries, while we employ do_generation to evaluate our approach. The author is currently occupied with other tasks. Subsequently, there will be further code organization and updates to the README.md will be provided.