pythia
pythia copied to clipboard
EleutherAI
A change to address the issue below. ## Issue: Documentation for Step Viewing Currently, the documentation for `utils/batch_viewer.py` reads as follows: ``` parser.add_argument( "--start_iteration", type=int, default=0, help="What train step to...
According to the `20B_tokenizer.json`, the end of document (EOD) token has id 0 and is denoted ``. Some people have raised in ealier issues that there are no EOD tokens...
Hi, thank you for developing such a useful tool! My requirement is that, modify the parameter of the trained model (checkpoint) and then reload it to the models to get...
We were using madgraph to study processes at the EP collider with Pythia8 and Delphes enabled, but were informed this approach is incorrect. This is because Pythia8 cannot be used...
Hi, I think the shuffle_idx.npy for seed2 is corrupted. Can someone else reproduce these results? (a copy of the issue exists on huggingface: https://huggingface.co/datasets/EleutherAI/pile-preshuffled-seeds/discussions/1) Expected: The shuffle_idx file is a...
