Add pipeline tag and update Block-Sparse Attention note

This PR improves the model card for FlashVSR by:

- Adding the `pipeline_tag: image-to-image` to the metadata, which enhances the model's discoverability on the Hugging Face Hub under the relevant task category (Video Super-Resolution).
- Updating the "3️⃣ Install Block-Sparse Attention (Required)" section with more detailed information, including an important `⚠️ Note` from the GitHub repository regarding GPU compatibility. This provides clearer guidance for users on optimal hardware usage.

The existing structure, links to the paper (via arXiv), project page, and usage instructions are maintained.

README.md

@@ -1,6 +1,8 @@
 ---
 license: apache-2.0
+pipeline_tag: image-to-image
 ---
+
 # ⚡ FlashVSR
 
 **Towards Real-Time Diffusion-Based Streaming Video Super-Resolution**
@@ -68,7 +70,7 @@ pip install -r requirements.txt
 
 #### 3️⃣ Install Block-Sparse Attention (Required)
 
-FlashVSR **requires** the **Block-Sparse Attention** backend for inference:
+FlashVSR relies on the **Block-Sparse Attention** backend to enable flexible and dynamic attention masking for efficient inference.
 
 ```bash
 git clone https://github.com/mit-han-lab/Block-Sparse-Attention
@@ -77,6 +79,7 @@ pip install packaging
 pip install ninja
 python setup.py install
 ```
+**⚠️ Note:** The Block-Sparse Attention backend currently achieves ideal acceleration only on NVIDIA A100 or A800 GPUs (Ampere architecture). On H100/H800 (Hopper) GPUs, due to differences in hardware scheduling and sparse kernel behavior, the expected speedup may not be realized, and in some cases performance can even be slower than dense attention.
 
 #### 4️⃣ Download Model Weights from Hugging Face
 
@@ -167,4 +170,4 @@ We gratefully acknowledge the following open-source projects:
       primaryClass={cs.CV},
       url={https://arxiv.org/abs/2510.12747}, 
 }
-```
+```