v0.0.2-alpha13
Add changes for SFT-6 training (#2554)
Subclass dataset (follow-up for https://github.com/LAION-AI/Open-Assistant/pull/2671) (#2686)
Flash attention support for Llama (#2277)
Add HFSummaryPairs class & fix AnthropicRLHF parsing (#2362)
Prepare oasst data export to HuggingFace (#2305)
Fix DeepSpeed 0.8.3 training (#2299)
Feature/remove reward instructor (#2289)
Add GPTNeoXRewardModel (#2182)
add dataset counts script (#2294)
fix bug
augment rank results for reward model and some other improvements for RM training (#2321)
Add rng_seed parameter to trainers (#2254)
Add max_replies parameter for trainer_rm (#2433)
Fix 'ConfigError: Attempted to change value of key "val_max_length" ...' and remove val_max_len variable (#2425)