Nathan Lambert

Closes #https://github.com/lvwerra/trl/issues/215 if correct on point 1 @younesbelkada !

I tested the logging change with my code in H4 #https://github.com/huggingface/h4/pull/73, and it fixed my problem!

I'll test `tensorboard` today. FYI this is needed for the script in H4, so I'll be motivated to get this working soon. If `tensorboard` doesn't work, I'll prolly do an...

@younesbelkada I think I ran this with `tensorboard` (just changed the config to as follows and it didn't error). Seems good to me? The term I changed `tracker_kwargs` was not...

Hey @idobenshaul10, this is not currently implemented. I just took a look and its a bit of a heavy lift due to the block arrangement in OLMo (which is much...

Hey @hxdtest, this is a good start. Looks like theres maybe a github action issue with 3.8, which I can look at later, and linting is needed for the style...

Yeah this looks good, I'm new to OLMo maintenance, so will see if we can get this merged. Can you bump the version here back to 3.9 https://github.com/allenai/OLMo/commit/d9c09937e4ad3daef97d1cfdaf2bd948d9316296

Can you say more about your dataset and tooling used? Is it with the OLMo repo? What model are you training from?

Thanks @geyang, this is still useful!

@jaku-jaku A lot of RL repositories still include `mujoco_py` as a dependency, so people are going to be dealing with this for years to come. Though I totally agree that...