Update README.md

README.md

@@ -31,7 +31,7 @@ pipeline_tag: sentence-similarity
 
 ---
 
-GritHopper is the first **decoder-based** multi-hop dense retrieval model and achieves **state-of-the-art performance** on both **in-distribution and out-of-distribution** benchmarks for **decomposition-free multi-hop dense retrieval**. Built on [GRITLM](https://github.com/ContextualAI/gritlm), it is trained across diverse datasets spanning **question-answering** and **fact-checking**. Unlike traditional decomposition-based approaches, GritHopper iteratively retrieves passages without explicit sub-question decomposition, concatenating retrieved evidence with the query at each step.
+GritHopper-7B is the first **decoder-based** multi-hop dense retrieval model and achieves **state-of-the-art performance** on both **in-distribution and out-of-distribution** benchmarks for **decomposition-free multi-hop dense retrieval**. Built on [GRITLM](https://github.com/ContextualAI/gritlm), it is trained across diverse datasets spanning **question-answering** and **fact-checking**. Unlike traditional decomposition-based approaches, GritHopper iteratively retrieves passages without explicit sub-question decomposition, concatenating retrieved evidence with the query at each step.
 
 Using the decoder model in an encoder-only approach (like [MDR](https://github.com/facebookresearch/multihop_dense_retrieval)), it performs each retrieval step in a single forward pass. In contrast to previous SOTA BERT-based approaches (like [BeamRetriever](https://github.com/canghongjian/beam_retriever) or [MDR](https://github.com/facebookresearch/multihop_dense_retrieval)), GritHopper generalizes significantly better to **out-of-distribution** data.
 
@@ -61,7 +61,7 @@ from grithopper import GritHopper
 
 # Initialize GritHopper with your GRITLM model checkpoint or huggingface path
 hopper = GritHopper(
-    model_name_or_path="UKPLab/GritHopper",  
+    model_name_or_path="UKPLab/GritHopper-7B",  
     device="cuda"  # or "cpu"
 )
 ```