Update README.md
Browse filesGreat work on this! 🙌
Propose adding a space to the query/document prepend instructions, as that seem to be what is used in the config.
README.md
CHANGED
|
@@ -48,7 +48,7 @@ For more information, please read our [blogpost](https://huggingface.co/blog/lig
|
|
| 48 |
|
| 49 |
## DenseOn
|
| 50 |
|
| 51 |
-
**DenseOn** is a dense (single-vector) retrieval model built on ModernBERT (149M parameters), trained by [LightOn](https://lighton.ai). It encodes queries and documents independently using cosine similarity with `query:`/`document:` prefixes and CLS pooling.
|
| 52 |
|
| 53 |
DenseOn achieves **56.75** average NDCG@10 on BEIR (14 datasets) and **57.71** on decontaminated BEIR (12 datasets), topping all base-size dense models and outperforming models up to 4x larger.
|
| 54 |
Notably it:
|
|
|
|
| 48 |
|
| 49 |
## DenseOn
|
| 50 |
|
| 51 |
+
**DenseOn** is a dense (single-vector) retrieval model built on ModernBERT (149M parameters), trained by [LightOn](https://lighton.ai). It encodes queries and documents independently using cosine similarity with `query: `/`document: ` prefixes and CLS pooling.
|
| 52 |
|
| 53 |
DenseOn achieves **56.75** average NDCG@10 on BEIR (14 datasets) and **57.71** on decontaminated BEIR (12 datasets), topping all base-size dense models and outperforming models up to 4x larger.
|
| 54 |
Notably it:
|