Oleh Shliazhko

Скажите, а какие планы по этой фиче? Когда примерно можно ожидать релиза?

Пробовал его сегодня, не все кейсы вытягивает (

``` print(dateparser.parse('через 30 секунд', languages=['ru'])) >>None print(dateparser.parse('5 часов', languages=['ru'])) >>2018-02-07 07:15:57.283900 # построил как "5 часов назад" print(dateparser.parse('в 5 часов', languages=['ru'])) >>2018-02-07 07:16:35.001085 # построил как "5 часов назад" ```

Круто, спасибо! Я сегодня въезжал в текущий yargy, попробую пока набросать грамматику для времени по аналогии с имеющейся для дат.

Sometimes this issue arises during fp16 training. We recommend to: 1. Try to decrease learning rate 2-4 times and resume training from last saved successful step. 2. If it doesn't...

I've used this script while working on the MMLU eval for LLaMA models in lm-evaluation-harness. I've replicated the numbers for the two of the LLaMA models, 7B & 13B. [Results...

Thanks for pointing it out! Could you provide the complete code snippet to reproduce the issue?

1. In this repo, it's true that the space is not inserted at the end of the sample. I did it precisely the same way it was [implemented in this...

Followup thought, for the logprob selection approach to be correct we can use the following method. Encode the whole line of the possible answer `Answer: A`, where the choice is...

> 2. for Eleuther I don't understand your point here nor the one in the other thread. I thought the whole point of your PR was to align Harness on...