I am looking to fine-tune a summarization model (Bart-Large-CNN). I would like to do this on a system without internet access. However, when I call load_metric(‘rouge’), a file is downloaded which is then used in computing the metric (4.28kB size file). What is this file? Where can I download it? Thanks!
I tested it out in a Colab journal, it just downloads and caches the rouge.py file (datasets/rouge.py at master · huggingface/datasets · GitHub). The metrics aren’t actually included in the datasets library, load_metric fetches them from the repo when they’re needed. If your system doesn’t have internet access, just place the rouge.py file somewhere on the system and then you can pass the path to the load_metric call, e.g.:
load_metric("./rouge.py")
More info here! Loading methods
4 Likes