Announcement
prime-rl v0.10.0 and verifiers v0.4.0 released with IPO as the default algorithm
PrimeIntellect says IPO is tailored for stability in asynchronous reinforcement learning.
TLDR
PrimeIntellect announced the release of prime-rl v0.10.0 and verifiers v0.4.0. It says its new IPO reinforcement-learning algorithm is now the default and is tailored for stability in asynchronous RL. The company also cites performance improvements and new model support.
Combined views
203
11 Sources, first seen ago
