Could you please share a more detailed hypermeter setting for your Fine-tuning Process, especially for the CorpusBrain(ππ‘+π΅πΏπΌππΎ) trail? Like max-tokens/GPU, the num of GPUs you used, update-freq, seed, and max steps you use.
And how did you make the train and dev corpus for CorpusBrain(ππ‘+π΅πΏπΌππΎ)? Did you just mix and shuffle the BLINK train corpus + KILT train for training, and mix all KILT devs for the development set?
Hi, thanks for your awesome work!
Could you please share a more detailed hypermeter setting for your Fine-tuning Process, especially for the CorpusBrain(ππ‘+π΅πΏπΌππΎ) trail? Like max-tokens/GPU, the num of GPUs you used, update-freq, seed, and max steps you use. And how did you make the train and dev corpus for CorpusBrain(ππ‘+π΅πΏπΌππΎ)? Did you just mix and shuffle the BLINK train corpus + KILT train for training, and mix all KILT devs for the development set?
Thanks so much! @Chriskuei