Improve model card: Add tags, links, introduction, and usage example

README.md

@@ -1,7 +1,55 @@
 ---
-license: mit
 base_model:
 - GSAI-ML/LLaDA-8B-Instruct
+license: mit
+pipeline_tag: text-generation
+library_name: transformers
 ---
 
-Post-Training Lora models on math task based on LLaDA-8B-Instruct for the paper Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspective
+# ESPO: Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspective
+
+This repository contains Post-Training LoRA models on various tasks based on LLaDA-8B-Instruct, as presented in the paper [Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspective](https://huggingface.co/papers/2512.03759).
+
+**ESPO (ELBO-based Sequence-level Policy Optimization)** is a principled reinforcement learning (RL) framework for diffusion Large Language Models (dLLMs). It addresses the fundamental challenges of adapting RL methods to dLLMs by treating the **entire sequence generation as a single action** and using the **ELBO** as a tractable sequence-level likelihood proxy. ESPO introduces per-token normalization of importance ratios and robust KL-divergence estimation for stable large-scale training, demonstrating significant performance improvements on mathematical reasoning, coding, and planning tasks.
+
+*   [Paper](https://huggingface.co/papers/2512.03759)
+*   [Code](https://github.com/ML-GSAI/ESPO)
+*   [Project Page](https://jingyangou.github.io/ESPO-Demo/)
+
+## Checkpoints
+We release ESPO-fine-tuned checkpoints built on LLaDA-8B-Instruct.
+
+ESPO-Code is released as a full fine-tuned model (no LoRA). ESPO-GSM8K, ESPO-Math, ESPO-Countdown, and ESPO-Sudoku are provided as LoRA adapters, which can be loaded on top of the base LLaDA-8B-Instruct model for lightweight and efficient fine-tuning.
+
+## Quick Start
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+from eval.generate_utils import generate
+base_model_path = 'GSAI-ML/LLaDA-8B-Instruct'
+peft_model_path = 'GSAI-ML/ESPO-Math'
+tokenizer = AutoTokenizer.from_pretrained(base_model_path)
+model = AutoModelForCausalLM.from_pretrained(
+    base_model_path, trust_remote_code=True,torch_dtype="bfloat16", device_map="cuda")
+peft_model = PeftModel.from_pretrained(model, peft_model_path, device_map="cuda")
+prompt = "The point $(0,0)$ is reflected over the vertical line $x=1$.  When its image is then reflected over the line $y=2$, what is the resulting point?
+
+Write your answer in the form $(x, y)$ where $x$ and $y$ are real numbers."
+messages = [{"role": "user", "content": prompt}]
+input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")
+output_ids = generate(peft_model, input_ids,tokenizer, steps=128,    gen_length=256, temperature=0.9,remasking="low_confidence",)
+output_text = tokenizer.batch_decode(output_ids[:, input_ids.shape[1]:], skip_special_tokens=True)[0]
+print(output_text)
+```
+
+## Citation
+If you find ESPO useful in your research, please consider to cite our paper:
+
+```bibtex
+@article{ou2025principledrldiffusionllms,
+      title={Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspective}, 
+      author={Jingyang Ou and Jiaqi Han and Minkai Xu and Shaoxuan Xu and Jianwen Xie and Stefano Ermon and Yi Wu and Chongxuan Li},
+      journal={arXiv preprint arXiv:2512.03759},
+      year={2025},
+}
+```