neilkale

4 ms·

1. 1.▲
   Modifying RL to train models on production traces(trajectory.ai)

   1 points by neilkale 5mo ago|0 comments
2. 2.▲
   by neilkale 2y ago

   Reminds me of the Physics of LM papers (Part 1 here https://arxiv.org/abs/2305.13673 )
3. 3.▲
   by neilkale 2y ago

   You don't even need to use quantization- most benchmarks can be broken with just prompting. https://arxiv.org/abs/2410.02879