Instructions to use Henryoung/WRIT-Qwen3-4B-Instruct-2507 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Henryoung/WRIT-Qwen3-4B-Instruct-2507 with Transformers:

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

pipe = pipeline("text-generation", model="Henryoung/WRIT-Qwen3-4B-Instruct-2507")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Henryoung/WRIT-Qwen3-4B-Instruct-2507")
model = AutoModelForCausalLM.from_pretrained("Henryoung/WRIT-Qwen3-4B-Instruct-2507")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Henryoung/WRIT-Qwen3-4B-Instruct-2507 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Henryoung/WRIT-Qwen3-4B-Instruct-2507"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Henryoung/WRIT-Qwen3-4B-Instruct-2507",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Henryoung/WRIT-Qwen3-4B-Instruct-2507

SGLang

How to use Henryoung/WRIT-Qwen3-4B-Instruct-2507 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 "Henryoung/WRIT-Qwen3-4B-Instruct-2507" \
    --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": "Henryoung/WRIT-Qwen3-4B-Instruct-2507",
		"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 "Henryoung/WRIT-Qwen3-4B-Instruct-2507" \
        --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": "Henryoung/WRIT-Qwen3-4B-Instruct-2507",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Henryoung/WRIT-Qwen3-4B-Instruct-2507 with Docker Model Runner:
```
docker model run hf.co/Henryoung/WRIT-Qwen3-4B-Instruct-2507
```

WRIT icon

WRIT-Qwen3-4B-Instruct-2507

This repository hosts a WRIT model checkpoint based on Qwen/Qwen3-4B-Instruct-2507.

Paper: WRIT: Write-Read Intensive Trajectory Synthesis for Multi-Turn User-Facing Agents.

Project homepage: https://hengrui-gu.github.io/WRIT/.

Model Summary

Method: WRIT
Base model: Qwen/Qwen3-4B-Instruct-2507
Primary use case: research on multi-turn, user-facing, tool-using agents
Checkpoint format: standard Transformers-compatible safetensors checkpoint

For Qwen3 inference, use non-thinking mode when your serving stack exposes a thinking toggle.

Quick Start

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Henryoung/WRIT-Qwen3-4B-Instruct-2507"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)

For tool-use experiments, use the tokenizer chat template included in this repository and the same tool schema format expected by your evaluation environment.

Intended Use

This checkpoint is intended for research on multi-turn tool-use, customer-service agents, and trajectory synthesis methods. It is not intended for deployment without task-specific validation and safety testing.

Limitations

The model may produce incorrect tool calls, incomplete task execution, or unsupported actions outside the target evaluation setting. Users should validate behavior carefully before applying it to any real user-facing workflow.

Citation

@misc{gu2026writwritereadintensivetrajectory,
      title={WRIT: Write-Read Intensive Trajectory Synthesis for Multi-Turn User-Facing Agents},
      author={Hengrui Gu and Xiaotian Han and Kaixiong Zhou},
      year={2026},
      eprint={2606.02908},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2606.02908},
}