Add paper and code links to model card

README.md

@@ -1,29 +1,27 @@
 ---
+base_model: Qwen/Qwen3-VL-8B-Instruct
+language:
+- en
 library_name: transformers
-pipeline_tag: visual-document-retrieval
 license: apache-2.0
-base_model: Qwen/Qwen3-VL-8B-Instruct
+pipeline_tag: visual-document-retrieval
 tags:
-  - reranker
-  - rerank
-  - listwise-reranker
-  - visual-document-retrieval
-  - multimodal
-  - document-understanding
-  - qwen3-vl
-  - rankgpt
-  - mmdocir
-language:
-  - en
+- reranker
+- rerank
+- listwise-reranker
+- visual-document-retrieval
+- multimodal
+- document-understanding
+- qwen3-vl
+- rankgpt
+- mmdocir
 ---
 
 # ZipRerank
 
-**ZipRerank** is a **listwise reranker for visual documents**, built on top of
-[`Qwen/Qwen3-VL-8B-Instruct`](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct).
+**ZipRerank** is a **listwise reranker for visual documents**, introduced in the paper [Very Efficient Listwise Multimodal Reranking for Long Documents](https://huggingface.co/papers/2605.11864). The official implementation is available on [GitHub](https://github.com/dukesun99/ZipRerank).
 
-Given a text query and a set of document page images (typically rendered from a PDF),
-ZipRerank scores every page and returns them ordered from most to least relevant.
+Built on top of [`Qwen/Qwen3-VL-8B-Instruct`](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct), ZipRerank is designed for high-efficiency multimodal reranking. Given a text query and a set of document page images (typically rendered from a PDF), the model scores every page and returns them ordered from most to least relevant in a single forward pass.
 
 ZipRerank can be used either as:
 
@@ -116,7 +114,8 @@ def create_ranking_prompt(query: str, num_passages: int) -> str:
         "The output format should be [A] > [B], etc.",
         "Only output the ranking results, do not say anything else.",
     ]
-    return "\n".join(lines)
+    return "
+".join(lines)
 
 
 @torch.no_grad()
@@ -239,4 +238,4 @@ top-K pages with ZipRerank.
 - Training focused on English documents; multilingual performance has not been evaluated,
   so results on non-English content may vary.
 - The window size is capped at 20 pages per forward pass (letters `A`–`T`); longer documents
-  rely on the sliding-window procedure described above.
+  rely on the sliding-window procedure described above.