Here, you'll find the raw scores for the HERM project.
The repository is structured as follows.
├── best-of-n/ <- Nested directory for different completions on Best of N challenge
| ├── alpaca_eval/ └── results for each reward model
| | ├── tulu-13b/{org}/{model}.json
| | └── zephyr-7b/{org}/{model}.json
| └── mt_bench/
|… See the full description on the dataset page:
https://huggingface.co/datasets/allenai/reward-bench-results.
