Instructions to use ttttonyhe/Qwen3-4B-Instruct-RETA-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ttttonyhe/Qwen3-4B-Instruct-RETA-Base with Transformers:

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

pipe = pipeline("text-generation", model="ttttonyhe/Qwen3-4B-Instruct-RETA-Base")
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("ttttonyhe/Qwen3-4B-Instruct-RETA-Base")
model = AutoModelForCausalLM.from_pretrained("ttttonyhe/Qwen3-4B-Instruct-RETA-Base", 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 ttttonyhe/Qwen3-4B-Instruct-RETA-Base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ttttonyhe/Qwen3-4B-Instruct-RETA-Base"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ttttonyhe/Qwen3-4B-Instruct-RETA-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ttttonyhe/Qwen3-4B-Instruct-RETA-Base

SGLang

How to use ttttonyhe/Qwen3-4B-Instruct-RETA-Base 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 "ttttonyhe/Qwen3-4B-Instruct-RETA-Base" \
    --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": "ttttonyhe/Qwen3-4B-Instruct-RETA-Base",
		"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 "ttttonyhe/Qwen3-4B-Instruct-RETA-Base" \
        --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": "ttttonyhe/Qwen3-4B-Instruct-RETA-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ttttonyhe/Qwen3-4B-Instruct-RETA-Base with Docker Model Runner:
```
docker model run hf.co/ttttonyhe/Qwen3-4B-Instruct-RETA-Base
```

Qwen3-4B-Instruct-RETA-Base

This checkpoint was initialized through supervised fine-tuning (SFT) to defend a tool-using agent against weak static prompt-injection attacks in AgentDojo. After each tool call, the model is trained to reason about whether the returned content aligns with the user's task, ignore conflicting instructions in the tool output, and continue the task.

This model is a fine-tuned version of Qwen/Qwen3-4B-Instruct-2507. It has been trained using TRL. Based on this checkpoint, we further trained it via a 250-step RL climb and then OPSD, the final checkpoint is available at https://huggingface.co/ttttonyhe/Qwen3-4B-Instruct-RETA, which has much higher robustness.

Quick start

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "ttttonyhe/defender_sft_qwen3_4b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype="auto")