Instructions to use THU-KEG/LongWriter-Zero-32B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use THU-KEG/LongWriter-Zero-32B with Transformers:

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

pipe = pipeline("text-generation", model="THU-KEG/LongWriter-Zero-32B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("THU-KEG/LongWriter-Zero-32B")
model = AutoModelForCausalLM.from_pretrained("THU-KEG/LongWriter-Zero-32B")
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]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use THU-KEG/LongWriter-Zero-32B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "THU-KEG/LongWriter-Zero-32B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "THU-KEG/LongWriter-Zero-32B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/THU-KEG/LongWriter-Zero-32B

SGLang

How to use THU-KEG/LongWriter-Zero-32B 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 "THU-KEG/LongWriter-Zero-32B" \
    --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": "THU-KEG/LongWriter-Zero-32B",
		"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 "THU-KEG/LongWriter-Zero-32B" \
        --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": "THU-KEG/LongWriter-Zero-32B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use THU-KEG/LongWriter-Zero-32B with Docker Model Runner:
```
docker model run hf.co/THU-KEG/LongWriter-Zero-32B
```

Can you share more information about the Writer Reward Model training ?

by cizhenshi - opened Jun 25, 2025

Discussion

cizhenshi

Jun 25, 2025

such as datasets size, data source

bys0318

Knowledge Engineer Group @ Tsinghua University org Jun 26, 2025

Our writing reward model (RM) is trained in a manner similar to the RM training approach described in https://arxiv.org/abs/2404.00934, except that our dataset includes a larger number of writing-related prompts and a more diverse collection of model responses.

cizhenshi

Jun 26, 2025

Thank you for your detailed response! It's very helpful to understand the training approach and the diversity of the dataset in the writing domain. I was wondering if it might be possible to share a general idea of the scale of reward pairs used in this area—would it be in the range of a few thousand, tens of thousands, or perhaps even hundreds of thousands? Of course, I completely understand if this information is sensitive. I really appreciate the insights you've shared so far!

bys0318

Knowledge Engineer Group @ Tsinghua University org Jun 26, 2025

•

edited Jun 26, 2025

Thanks for your understanding. The scale of reward pairs is above hundreds of thousands.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment