Reinforcement learnjng through human feedback

[zitterbewegung]

import pykoi as pk

use huggingface sft and reward model

config = pk.RLHFConfig( base_model_path="meta-llama/Llama-2-7b-hf", reward_model_path="goldmermaid/rlhf_reward_model", dataset_type="huggingface", dataset_name="goldmermaid/stack_exchange_rank_10k_dataset", dataset_subset_rl="data", )

rlhf_step3_rl = pk.RLFinetuning(config) rlhf_step3_rl.train_and_save("./models/rlhf_step3_rl")