Add pipeline tag, link to paper

This PR improves the model card, by:
- making sure the model can be found at https://huggingface.co/models?pipeline_tag=video-text-to-text&sort=trending
- is linked to https://huggingface.co/papers/2410.17434

Would be great to update all the other model cards with this!

README.md

@@ -4,6 +4,7 @@ datasets:
 - shenxq/VideoChat2
 base_model:
 - Vision-CAIR/LongVU_Qwen2_7B_img
+pipeline_tag: video-text-to-text
 model-index:
 - name: llava-onevision-qwen-7b-ov
   results:
@@ -50,6 +51,8 @@ model-index:
 ---
 # LongVU
 
+This repository contains the model based on Qwen2-7B as presented in [LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding](https://huggingface.co/papers/2410.17434).
+
 Play with the model on the [HF demo](https://huggingface.co/spaces/Vision-CAIR/LongVU).
 
 <div align="left">