Update README.md

README.md

@@ -1 +1,17 @@
-This model uses pretrained weights from distilbert-base-uncased and fine tunes the classification head to the manually annotated news corpus. Long articles are chunked in order to fit in the 512 token limit.  
+---
+language: en
+datasets:
+- valurank/news-small
+---
+
+# DistilROBERTA fine-tuned for news classification
+
+This model is based on [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) pretrained weights, with a classification head fine-tuned to classify news articles into 3 categories (bad, medium, good).
+
+## Training data
+
+The dataset used to fine-tune the model is [news-small](https://huggingface.co/datasets/valurank/news-small), the 300 article news dataset manually annotated by Alex.
+
+## Inputs
+
+Similar to its base model, this model accepts inputs with a maximum length of 512 tokens.