# Ring-lite

<p align="center">
    <img src="https://huggingface.co/inclusionAI/Ring-lite/resolve/main/ant-bailing.png" width="100"/>
<p>

<p align="center">
          🤗 <a href="https://huggingface.co/inclusionAI">Hugging Face</a>
<p>

## Introduction

Ring-lite is an fully open-source MoE LLM provided by InclusionAI, which has 16.8B parameters with 2.75B activated parameters. It was derived from [Ling-lite-1.5](https://huggingface.co/inclusionAI/Ling-lite-1.5) through a training process involving reasoning SFT, reasoning RL and general SFT. This model delivers performance comparable to [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) on reasoning benchmarks, while activating only one-third of their parameters.

## Model Downloads

<div align="center">

|     **Model**      | **#Total Params** | **#Activated Params** | **Context Length** | **Download** |
| :----------------: | :---------------: | :-------------------: | :----------------: | :----------: |
| Ring-lite |       16.8B       |         2.75B         |        64K         |      [🤗 HuggingFace](https://huggingface.co/inclusionAI/Ring-lite) |

</div>

## Evaluation
In order to fully evaluate the model's reasoning performance, we examined Ring-lite on several reasoning benchmarks, including MATH-500, AIME-24, AIME-24, Livecodebench, Codeforces and GPQA.

<p align="center">
    <img src="https://huggingface.co/inclusionAI/Ring-lite/resolve/main/performance.png" width="500"/>
<p>



More details will be reported in our technical report. [TBD]

## Quickstart

### 🤗 Hugging Face Transformers
Here is a code snippet to show you how to use the chat model with `transformers`:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "inclusionAI/Ring-lite-distill-preview"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "Give me a short introduction to large language models."
messages = [
    {"role": "system", "content": "You are Ring, an assistant created by inclusionAI"},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=8192
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
```

## Dataset
The training data of Ring-lite-distill-preview will be released soon. 

## Deployment
Please refer to [GitHub](https://github.com/inclusionAI/Ring/blob/main/README.md)

## License
This code repository is licensed under [the MIT License](https://huggingface.co/inclusionAI/Ring-lite-distill/blob/main/LICENSE).

## Citation
[TBD]