From 33876d124f19e6996c3a09b16b43183da2ef6cf3 Mon Sep 17 00:00:00 2001 From: Nicklas Hansen Date: Sun, 7 Jan 2024 11:55:07 -0800 Subject: [PATCH] add instructions for distributed training --- README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/README.md b/README.md index 94534a4..f724fa2 100755 --- a/README.md +++ b/README.md @@ -112,6 +112,8 @@ $ python train.py task=walker-walk obs=rgb We recommend using default hyperparameters for single-task online RL, including the default model size of 5M parameters (`model_size=5`). Multi-task offline RL benefits from a larger model size, but larger models are also increasingly costly to train and evaluate. Available arguments are `model_size={1, 5, 19, 48, 317}`. See `config.yaml` for a full list of arguments. +**As of Jan 7, 2024 the TD-MPC2 codebase also supports multi-GPU training for multi-task offline RL experiments**; use branch `distributed` and argument `world_size=N` to train on `N` GPUs. We cannot guarantee that distributed training will yield the same results, but they appear to be similar based on our limited testing. + ---- ## Citation