Instructions to use Joinn/UserMirrorrer-Qwen-DPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Joinn/UserMirrorrer-Qwen-DPO with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Joinn/UserMirrorrer-Qwen-DPO")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Joinn/UserMirrorrer-Qwen-DPO")
model = AutoModelForCausalLM.from_pretrained("Joinn/UserMirrorrer-Qwen-DPO", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Joinn/UserMirrorrer-Qwen-DPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Joinn/UserMirrorrer-Qwen-DPO"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Joinn/UserMirrorrer-Qwen-DPO",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Joinn/UserMirrorrer-Qwen-DPO

SGLang

How to use Joinn/UserMirrorrer-Qwen-DPO with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Joinn/UserMirrorrer-Qwen-DPO" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Joinn/UserMirrorrer-Qwen-DPO",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Joinn/UserMirrorrer-Qwen-DPO" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Joinn/UserMirrorrer-Qwen-DPO",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Joinn/UserMirrorrer-Qwen-DPO with Docker Model Runner:
```
docker model run hf.co/Joinn/UserMirrorrer-Qwen-DPO
```

Joinn

nielsr HF Staff commited on Apr 21

Commit

a21392d

1 Parent(s): 2ada39e

Improve model card and add metadata (#1)

Browse files

- Improve model card and add metadata (ec520b373ef7343eae9ad107882203aeda915a43)

Co-authored-by: Niels Rogge <nielsr@users.noreply.huggingface.co>

Files changed (1) hide show

README.md +35 -11

README.md CHANGED Viewed

@@ -1,21 +1,45 @@
 ---
 library_name: transformers
-tags: []
 ---
-# Model Card for Model ID
-<!-- Provide a quick summary of what the model is/does. -->
-## UserMirrorrer-Qwen-DPO
-This is the fine-tuned checkpoint using our constructed dataset [UserMirrorer](https://huggingface.co/datasets/MirrorUser/UserMirrorer).
-Please refer to our paper: "Mirroring Users: Towards Building Preference-aligned User Simulator with Recommendation Feedback".
 ## Model Details
-The finetuned model is based on [Qwen-2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct).
-The fine-tuning process involves 2 stage:
-1. Supervised Finetuning for 1 epoch
-2. DPO Finetuning for 2 epochs

 ---
 library_name: transformers
+pipeline_tag: text-generation
+base_model: Qwen/Qwen2.5-3B-Instruct
+license: cc-by-nc-4.0
+tags:
+- recommendation-system
+- user-simulation
 ---
+# UserMirrorrer-Qwen-DPO
+This is a fine-tuned user simulator model introduced in the paper "[Mirroring Users: Towards Building Preference-aligned User Simulator with User Feedback in Recommendation](https://huggingface.co/papers/2508.18142)".
+The model is designed to simulate user behavior in recommender systems (RSs) by leveraging extensive user feedback to achieve better preference alignment. It uses decision-making processes as explanatory rationales to reduce ambiguity in simulation samples.
 ## Model Details
+- **Base Model:** [Qwen-2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct)
+- **Fine-tuning Process:**
+  1. **Supervised Finetuning (SFT):** 1 epoch.
+  2. **Direct Preference Optimization (DPO):** 2 epochs.
+- **Dataset:** [UserMirrorer](https://huggingface.co/datasets/MirrorUser/UserMirrorer)
+## Resources
+- **Paper:** [Mirroring Users: Towards Building Preference-aligned User Simulator with User Feedback in Recommendation](https://arxiv.org/abs/2508.18142)
+- **GitHub Repository:** [Joinn99/UserMirrorer](https://github.com/Joinn99/UserMirrorer)
+- **Training Data:** [UserMirrorer Training Set](https://huggingface.co/datasets/Joinn/UserMirrorer)
+## Citation
+If you find this work useful in your research, please consider citing the following paper:
+```bibtex
+@misc{wei2025mirroringusersbuildingpreferencealigned,
+      title={Mirroring Users: Towards Building Preference-aligned User Simulator with User Feedback in Recommendation},
+      author={Tianjun Wei and Huizhong Guo and Yingpeng Du and Zhu Sun and Huang Chen and Dongxia Wang and Jie Zhang},
+      year={2025},
+      eprint={2508.18142},
+      archivePrefix={arXiv},
+      primaryClass={cs.HC},
+      url={https://arxiv.org/abs/2508.18142},
+}
+```