|
--- |
|
license: apache-2.0 |
|
language: |
|
- en |
|
- zh |
|
pipeline_tag: text-generation |
|
tags: |
|
- ' TransNormerLLM' |
|
--- |
|
|
|
<div align="center"> |
|
<h1> |
|
TransNormerLLM3 -- A Faster and Better LLM |
|
</h1> |
|
</div> |
|
|
|
# Introduction |
|
|
|
This official repository unveils the TransNormerLLM3 model along with its open-source weights for every 50 billion tokens processed during pre-training. |
|
|
|
[TransNormerLLM](https://arxiv.org/abs/2307.14995) evolving from [TransNormer](https://arxiv.org/abs/2210.10340), standing out as the first LLM within the linear transformer architecture. Additionally, it distinguishes itself by being the first non-Transformer LLM to exceed both traditional Transformer and other efficient Transformer models (such as, RetNet and Mamba) in terms of speed and performance. |
|
|
|
|
|
# TransNormerLLM3 |
|
- **TransNormerLLM3-15B** features **14.83 billion** parameters. It is structured with **42 layers**, includes **40 attention heads**, and has a total **embedding size of 5120**. |
|
- **TransNormerLLM3-15B** is purely intergrated with **[Lightning Attention-2](http://arxiv.org/abs/2401.04658)**, which can maintain a **stable TGS** during training of **unlimited sequence lengths**, up until encountering firm limitations like GPU memory constraints. |
|
- **Titoken** tokenizer is used with a total **vocabulary size** of about **100,000**. |
|
- It incorporates **Simple GLU** for its channel mixer, **GLA** in the token mixer, and **SRMSNorm** for normalization. |
|
- In terms of position encoding, the first layer employs **LRPE with exponential decay**, whereas the subsequent layers continue with **exponential decay encoding**. |
|
|
|
### Pre-training Logbook |
|
* Realtime Track: https://api.wandb.ai/links/opennlplab/kip314lq |
|
* Join to dicussion: [discord](https://discord.gg/MYQh6BWN) <<<>>> [wechat group](https://github.com/OpenNLPLab/TransnormerLLM/blob/main/images/contact_me_qr.png) |
|
|
|
> --23.12.25-- startup: [WeChat - 预训练启航](https://mp.weixin.qq.com/s/YjUY-uy89WkF75_-rBTuKw) <<<>>> [Twitter - Pre-training Commences ](https://twitter.com/opennlplab/status/1739568669502611825) <<<>>> [YouTube Recording](https://t.co/wk7svS4o5r) <<<>>> [bilibili 回放](https://www.bilibili.com/video/BV11j411J7Dy) |
|
> --24.01.02-- first week review: [WeChat - 第一周概览](https://mp.weixin.qq.com/s/zwGnZZI3itNPoxzzXkuU2w) <<<>>> [Twitter - First Week Review](https://twitter.com/opennlplab/status/1742187694078501038) |
|
> --24.01.09-- second week review: [WeChat - 第二周概览](https://mp.weixin.qq.com/s/6D0qi-0aBier05OKuHfPEA) <<<>>> [Twitter - Second Week Review](https://twitter.com/opennlplab/status/1744720007299523063) |
|
|
|
|
|
# Released Weights |
|
|
|
| param | token | Hugging Face | Model Scope | Wisemodel | |
|
| :-----: | :---: | :----------: | :---------: | :-------: | |
|
| **15B** | 50B | 🤗 | 🤖 | 🐯 | |
|
|
|
# Benchmark Results |
|
The evaluations of all models are conducted using the official settings and the [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) framework. |
|
|
|
| Model | P | T | BoolQ | PIQA | HS | WG | ARC-e | ARC-c | OBQA | MMLU | C-Eval | |
|
| ----------------------- | --- | ---- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ----- | ------ | |
|
| **TransNormerLLM3-15B** | 15 | 0.05 | 62.08 | 72.52 | 55.55 | 57.14 | 62.12 | 31.14 | 32.40 | 27.50 | 26.18 | |
|
| **TransNormerLLM3-15B** | 15 | 0.10 | 63.98 | 74.70 | 61.09 | 61.33 | 65.95 | 34.64 | 35.60 | 25.38 | 27.40 | |
|
| **TransNormerLLM3-15B** | 15 | 0.15 | 60.34 | 75.08 | 63.99 | 62.04 | 64.56 | 34.90 | 35.20 | 22.64 | 26.60 | |
|
|
|
|
|
|
|
> **P**: parameter size (billion). **T**: tokens (trillion). **BoolQ**: acc. **PIQA**: acc. **HellaSwag**: acc_norm. **WinoGrande**: acc. **ARC-easy**: acc. **ARC-challenge**: acc_norm. **OpenBookQA**: acc_norm. **MMLU**: 5-shot acc. **C-Eval**: 5-shot acc. |
|
|
|
|
|
# Acknowledgments and Citation |
|
|
|
## Acknowledgments |
|
Our project is developed based on the following open source projects: |
|
- [tiktoken](https://github.com/openai/tiktoken) for the tokenizer. |
|
- [metaseq](https://github.com/facebookresearch/metaseq) for training. |
|
- [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) for evaluation. |
|
|
|
|
|
## Citation |
|
If you wish to cite our work, please use the following reference: |
|
``` |
|
@article{qin2023scaling, |
|
title={Scaling transnormer to 175 billion parameters}, |
|
author={Qin, Zhen and Li, Dong and Sun, Weigao and Sun, Weixuan and Shen, Xuyang and Han, Xiaodong and Wei, Yunshen and Lv, Baohong and Yuan, Fei and Luo, Xiao and others}, |
|
journal={arXiv preprint arXiv:2307.14995}, |
|
year={2023} |
|
} |
|
|
|
@misc{qin2024lightning, |
|
title={Lightning Attention-2: A Free Lunch for Handling Unlimited Sequence Lengths in Large Language Models}, |
|
author={Zhen Qin and Weigao Sun and Dong Li and Xuyang Shen and Weixuan Sun and Yiran Zhong}, |
|
year={2024}, |
|
eprint={2401.04658}, |
|
archivePrefix={arXiv}, |
|
primaryClass={cs.CL} |
|
} |
|
``` |
|
|
|
<p align="center"> |
|
- OpenNLPLab @2024 - |
|
</p> |
