Instructions to use phanviethoang1512/Qwen3-8B-UserRL-Agent with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use phanviethoang1512/Qwen3-8B-UserRL-Agent with Transformers:

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

pipe = pipeline("text-generation", model="phanviethoang1512/Qwen3-8B-UserRL-Agent")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("phanviethoang1512/Qwen3-8B-UserRL-Agent")
model = AutoModelForCausalLM.from_pretrained("phanviethoang1512/Qwen3-8B-UserRL-Agent", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use phanviethoang1512/Qwen3-8B-UserRL-Agent with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "phanviethoang1512/Qwen3-8B-UserRL-Agent"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "phanviethoang1512/Qwen3-8B-UserRL-Agent",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/phanviethoang1512/Qwen3-8B-UserRL-Agent

SGLang

How to use phanviethoang1512/Qwen3-8B-UserRL-Agent 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 "phanviethoang1512/Qwen3-8B-UserRL-Agent" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "phanviethoang1512/Qwen3-8B-UserRL-Agent",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "phanviethoang1512/Qwen3-8B-UserRL-Agent" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "phanviethoang1512/Qwen3-8B-UserRL-Agent",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use phanviethoang1512/Qwen3-8B-UserRL-Agent with Docker Model Runner:
```
docker model run hf.co/phanviethoang1512/Qwen3-8B-UserRL-Agent
```

Qwen3-8B-UserRL-Agent

The UserRL baseline from MIMESIS: Qwen3-8B trained with multi-turn GRPO, with GPT-5.5 as the simulated user, following UserRL. The paper reports it as "GRPO w. GPT-5.5" in the main text and as UserRL in Appendix D.

Paper · Code · Project page · Collection

Results

Mean task score across eight Gym environments, three of them held out from training, under nine evaluation user models that were not used in training:

Training condition	Mean score
GRPO w. GPT-5.5 (this model)	26.10
GRPO w. MIMESIS-9B	29.54

With the GRPO objective fixed, replacing GPT-5.5 with MIMESIS-9B as the training user improves the mean score under every evaluation user. Per-environment and per-user scores are in Appendix D of the paper.

Usage

The agent acts in the UserRL Gym environments through a tool call. Training and evaluation scripts, including evaluation under any user model, are in the agent directory of the code release.

Citation

@article{phan2026mimesis,
  title   = {{MIMESIS}: Learning User Simulators as Training Environments for Interactive Agents},
  author  = {Phan, Hoang and Huynh, Dat and Zhmoginov, Andrey and Zeng, Qi and Mu, Wancen and Cao, Yue and Bi, Shengjie and He, Yun and Oh, Changdae and Lei, Deren},
  journal = {arXiv preprint arXiv:2610.09484},
  year    = {2026}
}