Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
license: mit
|
| 2 |
+
language:
|
| 3 |
+
- zh
|
| 4 |
+
- en
|
| 5 |
+
base_model:
|
| 6 |
+
- inclusionAI/Ling-lite
|
| 7 |
+
pipeline_tag: text-generation
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# Ring-lite
|
| 11 |
+
|
| 12 |
+
<p align="center">
|
| 13 |
+
<img src="https://huggingface.co/inclusionAI/Ring-lite-distill-preview/resolve/main/ant-bailing.png" width="100"/>
|
| 14 |
+
<p>
|
| 15 |
+
|
| 16 |
+
<p align="center">
|
| 17 |
+
🤗 <a href="https://huggingface.co/inclusionAI">Hugging Face</a>
|
| 18 |
+
<p>
|
| 19 |
+
|
| 20 |
+
## Introduction
|
| 21 |
+
|
| 22 |
+
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 parameter. . This demonstrates that Ring-lite-distill is a more balanced and versatile model. Additionaly, it maintains competitive latency and throughput compared to other reasoning LLMs of similar size.
|
| 23 |
+
|
| 24 |
+
## Model Downloads
|
| 25 |
+
|
| 26 |
+
<div align="center">
|
| 27 |
+
|
| 28 |
+
| **Model** | **#Total Params** | **#Activated Params** | **Context Length** | **Download** |
|
| 29 |
+
| :----------------: | :---------------: | :-------------------: | :----------------: | :----------: |
|
| 30 |
+
| Ring-lite-distill-preview | 16.8B | 2.75B | 64K | [🤗 HuggingFace](https://huggingface.co/inclusionAI/Ring-lite-distill) |
|
| 31 |
+
|
| 32 |
+
</div>
|
| 33 |
+
|
| 34 |
+
## Evaluation
|
| 35 |
+
In order to fully evaluate the model's performance, we examined Ring-lite-distill-preview in terms of both reasoning ability and general ability.
|
| 36 |
+
### Reasoning ability
|
| 37 |
+
|
| 38 |
+
<div align="center">
|
| 39 |
+
|
| 40 |
+
| **Model** | **AIME24** | **MATH-500** | **GPQA-diamond** | **LiveCodeBench** |
|
| 41 |
+
| :----------------: | :---------------: | :-------------------: | :----------------: | :----------: |
|
| 42 |
+
| DeepSeek-R1-Distill-Qwen-7B (reported) | 55.5 | 92.8 | 49.1 | 37.6 |
|
| 43 |
+
| DeepSeek-R1-Distill-Qwen-7B (reproduce) | 53.2 | 93.7 | 50.4 | 36.5 |
|
| 44 |
+
| Ring-lite-distill-preview | 56.3 | 93.7 | 46.2 | 31.9 |
|
| 45 |
+
|
| 46 |
+
</div>
|
| 47 |
+
|
| 48 |
+
### General ability
|
| 49 |
+
|
| 50 |
+
<div align="center">
|
| 51 |
+
|
| 52 |
+
| **Model** | **IFEval** | **T-eval** | **BFCL_v2** | **MMLU** |
|
| 53 |
+
| :----------------: | :---------------: | :-------------------: | :----------------: | :----------: |
|
| 54 |
+
| DeepSeek-R1-Distill-Qwen-7B (reproduce) | 39.3 | 26.9 | 38.9 | 44.1 |
|
| 55 |
+
| Ring-lite-distill-preview | 75.3 | 81.3 | 63.0 | 63.3 |
|
| 56 |
+
|
| 57 |
+
</div>
|
| 58 |
+
More details will be reported in our technical report. [TBD]
|
| 59 |
+
|
| 60 |
+
## Quickstart
|
| 61 |
+
|
| 62 |
+
### 🤗 Hugging Face Transformers
|
| 63 |
+
Here is a code snippet to show you how to use the chat model with `transformers`:
|
| 64 |
+
|
| 65 |
+
```python
|
| 66 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 67 |
+
|
| 68 |
+
model_name = "inclusionAI/Ring-lite-distill-preview"
|
| 69 |
+
|
| 70 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 71 |
+
model_name,
|
| 72 |
+
torch_dtype="auto",
|
| 73 |
+
device_map="auto"
|
| 74 |
+
)
|
| 75 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 76 |
+
|
| 77 |
+
prompt = "Give me a short introduction to large language models."
|
| 78 |
+
messages = [
|
| 79 |
+
{"role": "system", "content": "You are Ring, an assistant created by inclusionAI"},
|
| 80 |
+
{"role": "user", "content": prompt}
|
| 81 |
+
]
|
| 82 |
+
text = tokenizer.apply_chat_template(
|
| 83 |
+
messages,
|
| 84 |
+
tokenize=False,
|
| 85 |
+
add_generation_prompt=True
|
| 86 |
+
)
|
| 87 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
| 88 |
+
|
| 89 |
+
generated_ids = model.generate(
|
| 90 |
+
**model_inputs,
|
| 91 |
+
max_new_tokens=8192
|
| 92 |
+
)
|
| 93 |
+
generated_ids = [
|
| 94 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
| 95 |
+
]
|
| 96 |
+
|
| 97 |
+
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
| 98 |
+
```
|
| 99 |
+
|
| 100 |
+
## Dataset
|
| 101 |
+
The training data of Ring-lite-distill-preview will be released soon.
|
| 102 |
+
|
| 103 |
+
## Deployment
|
| 104 |
+
Please refer to [GitHub](https://github.com/inclusionAI/Ring/blob/main/README.md)
|
| 105 |
+
|
| 106 |
+
## License
|
| 107 |
+
This code repository is licensed under [the MIT License](https://huggingface.co/inclusionAI/Ring-lite-distill/blob/main/LICENSE).
|
| 108 |
+
|
| 109 |
+
## Citation
|
| 110 |
+
[TBD]
|