Improve model card for SEAgent: Add metadata and description

This PR significantly enhances the model card for SEAgent by:
- Adding the `pipeline_tag: image-text-to-text`, which helps users discover the model when filtering by task on the Hugging Face Hub.
- Specifying the `library_name: transformers` to indicate compatibility with the `transformers` library, enabling the "How to use" widget on the model page.
- Providing a concise yet informative description of the model's capabilities and purpose, derived from the paper's abstract.
- Updating the paper link from arXiv to the official Hugging Face papers page for better integration and user experience.

README.md

@@ -1,13 +1,24 @@
 ---
-license: apache-2.0
-datasets:
-- xlangai/ubuntu_osworld
 base_model:
 - ByteDance-Seed/UI-TARS-7B-DPO
+datasets:
+- xlangai/ubuntu_osworld
+license: apache-2.0
+pipeline_tag: image-text-to-text
+library_name: transformers
 ---
 
-Checkout our github repo and arxiv paper for inference!
+# SEAgent: Self-Evolving Computer Use Agent with Autonomous Learning from Experience
+
+This repository hosts `SEAgent`, an innovative agentic self-evolving framework designed to empower Large Vision-Language Models (LVLMs) to function as Computer Use Agents (CUAs). SEAgent enables these agents to autonomously master novel and specialized software environments, particularly in scenarios where human annotations are scarce.
+
+The framework achieves this through experiential learning, allowing agents to explore new software, learn via iterative trial-and-error, and progressively tackle auto-generated tasks organized from simple to complex. Key innovations include a **World State Model** for step-wise trajectory assessment and a **Curriculum Generator** that generates increasingly diverse and challenging tasks. The agent's policy is updated through a novel experiential learning approach, combining adversarial imitation of failure actions and Group Relative Policy Optimization (GRPO) for successful ones. Furthermore, SEAgent introduces a specialist-to-generalist training strategy to develop a stronger generalist CUA capable of continuous autonomous evolution.
+
+SEAgent demonstrates significant performance improvements, achieving a 23.2% increase in success rate on five novel software environments within OS-World, surpassing competitive open-source CUAs like UI-TARS.
+
+For more detailed information, please refer to our paper and code:
 
-https://github.com/SunzeY/SEAgent
+*   **Paper:** [SEAgent: Self-Evolving Computer Use Agent with Autonomous Learning from Experience](https://huggingface.co/papers/2508.04700)
+*   **Code:** [https://github.com/SunzeY/SEAgent](https://github.com/SunzeY/SEAgent)
 
-https://arxiv.org/abs/2508.04700
+For inference and usage instructions, please consult the official GitHub repository.