HackerTrans
TopNewTrendsCommentsPastAskShowJobs

jumpCastle

no profile record

comments

jumpCastle
·geçen yıl·discuss
It was a good benchmark until it entered the training set.
jumpCastle
·geçen yıl·discuss
But open source like aider
jumpCastle
·geçen yıl·discuss
Deepseek v3 was FP8
jumpCastle
·geçen yıl·discuss
It's quite similar to muP

https://github.com/microsoft/mup
jumpCastle
·2 yıl önce·discuss
Stock prices cannot go up without if the planet is destroyed
jumpCastle
·2 yıl önce·discuss
Attention was invented because Bengio lab had to be disciplined about a black box (google had more compute)
jumpCastle
·2 yıl önce·discuss
Aren't services like runpod solve half of these concerns?
jumpCastle
·2 yıl önce·discuss
Are the updates really slow?
jumpCastle
·2 yıl önce·discuss
Spend more time with my side projects.
jumpCastle
·2 yıl önce·discuss
You can try his book. https://www.math.ias.edu/avi/book
jumpCastle
·2 yıl önce·discuss
You can use older models with the api.
jumpCastle
·2 yıl önce·discuss
Also the parameters are optimized also with loss of future tokens in the sequence.