yoavg

Reinforcement Learning for Language Models

Yoav Goldberg, April 2023.

Why RL?

With the release of the ChatGPT model and followup large language models (LLMs), there was a lot of discussion of the importance of "RLHF training", that is, "reinforcement learning from human feedback". I was puzzled for a while as to why RL (Reinforcement Learning) is better than learning from demonstrations (a.k.a supervised learning) for training language models. Shouldn't learning from demonstrations (or, in language model terminology "instruction fine tuning", learning to immitate human written answers) be sufficient? I came up with a theoretical argument that was somewhat convincing. But I came to realize there is an additional argumment which not only supports the case of RL training, but also requires it, in particular for models like ChatGPT. This additional argument is spelled out in (the first half of) a talk by John Schulman from OpenAI. This post pretty much

tae-jun

ARG UBUNTU_VERSION=18.04
ARG CUDA_VERSION=10.2
FROM nvidia/cuda:${CUDA_VERSION}-base-ubuntu${UBUNTU_VERSION}
# An ARG declared before a FROM is outside of a build stage,
# so it can’t be used in any instruction after a FROM
ARG USER=reasearch_monster
ARG PASSWORD=${USER}123$
ARG PYTHON_VERSION=3.8
# To use the default value of an ARG declared before the first FROM,
# use an ARG instruction without a value inside of a build stage:

nraw

""" Kedro Torch Model IO

Models need to be imported and added to the dictionary
as shown with the ExampleModel

Example of catalog entry:

modo:
  type: kedro_example.io.torch_model.TorchLocalModel
  filepath: modo.pt