rajhans

4 ms·

1. 1.▲
   by rajhans 2y ago

   We have published some insights here. https://medium.com/snowflake/snowflake-arctic-cookbook-serie...
2. 2.▲
   by rajhans 2y ago

   Arctic dev here. Yes keeping all experts in memory is the recommendation here and understandably that is a barrier to some. But once you have 1 H100 node or two (gpu middle-class I guess...?), then a few things to note: 1. FP6/FP8 infe
3. 3.▲
   by rajhans 2y ago

   One of the modelers working on Arctic. We have done no alignment training whatsoever.
4. 4.▲
   by rajhans 9y ago

   On day one, Spoke's ML and NLP algorithm can respond with the right answer that is relatively directly related (i.e. has a good amount of keyword overlap with the query) without any training. For complex queries, if Spoke does not find
5. 5.▲
   by rajhans 10y ago

   Great work! A few questions for the author(s): In the article, you have listed 9 feature extractors/templates. In the final model, what's the total number (or rough magnitude) of features? How much data (or ballpark estimate) did