HackerTrans
TopNewTrendsCommentsPastAskShowJobs

jwan584

no profile record

Submissions

Mistral Flash Answers Run on Cerebras

cerebras.ai
5 points·by jwan584·anno scorso·1 comments

100x defect tolerance: How we solved the yield problem

cerebras.ai
331 points·by jwan584·anno scorso·179 comments

Cerebras CS-3: the fastest and most scalable AI accelerator

cerebras.net
2 points·by jwan584·2 anni fa·0 comments

comments

jwan584
·anno scorso·discuss
The point about using FP32 for training is wrong. Mixed precision (FP16 multiplies, FP32 accumulates) has been use for years – the original paper came out in 2017.
jwan584
·anno scorso·discuss
A good talk on how Cerebras does power & cooling (8min) https://www.youtube.com/watch?v=wSptSOcO6Vw&ab_channel=Appli...
jwan584
·2 anni fa·discuss
batch size by Q4 will be solid double digits (cerebras employee)