Olmo / OLMo consistency

[Muennighoff]

🐛 Describe the bug

In the HF code we use OLMo but in training it's Olmo - This creates some inconsistencies when importing from the training modeling file ---- I think we should settle on one for both - wdyt? Probably OLMo as it's the actual name; happy to open a pr (filing as a bug as e.g. _no_split_modules = ["OLMoBlock"] doesn't work i think)

cc @AkshitaB @dirkgr

Versions

latest

[dirkgr]

Please file a PR! In the code, they should all be "OlmoSomething".

[Muennighoff]

Please file a PR! In the code, they should all be "OlmoSomething".

Sure so I'll rename everything in here https://github.com/allenai/OLMo/blob/main/hf_olmo/modeling_olmo.py to Olmo?

[dirkgr]

👍🏻

[Muennighoff]

Sounds good - will just wait for @AkshitaB to give her okay, as I will need to change a bunch of things including the config of the model on the hub https://huggingface.co/allenai/OLMo-7B/blob/main/config.json

[AkshitaB]

Huggingface recommended the convention to be OLMoXYZ for the naming of the model. Also, I will recommend not making changes to the model on the hub unless absolutely necessary, as it's not just 1 model, it's ALL the 1000+ checkpoints for all 3 models.

[dirkgr]

Ok, then we'll go with Huggingface's suggestion. That means we rename everything to OLMo*, right?

[Muennighoff]

Fixed it here: https://github.com/allenai/OLMo/pull/466