__vec__

4 ms·

1. 1.▲
   The tide is shifting: 1.3B outperforms 7B Llama 2(arxiv.org)

   63 points by __vec__ 3y ago|15 comments
2. 2.▲
   by __vec__ 3y ago

   Textbooks Are All You Need II: phi-1.5 technical report "Perhaps achieving ChatGPT’s level of capability at the one billion parameters scale is actually achievable?"
3. 3.▲
   by __vec__ 3y ago

   The difference is an informative error message at compile time when such fails to satisfy the 'refinement'
4. 4.▲
   by __vec__ 3y ago

   DML link in the article is broken - Might I instead point you to a relevant ATS-lang (successor 2 of DML) documentation about `Dependent Types in ATS2` https://ats-lang.sourceforge.net/DOCUMENT/INTPROGINATS/HTML&#x
5. 5.▲
   by __vec__ 3y ago

   Anyone recreate all of the results with un"contaminated datasets"?