qkvflow

KO | EN

gitlite — search

#javascript #hacktoberfest #llm #golang #android #macos #nodejs #windows #cli #kubernetes #vue #pytorch #ios #mcp

(★ 21)

Neural ODE Transformers (ICLR 2025)

qkvflow 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Neural ODE Transformers: Analyzing Internal Dynamics and Adaptive Fine-tuning

[![ICLR 2025](https://img.shields.io/badge/ICLR-2025-blue)](https://iclr.cc/)

This repository contains the official implementation of the paper **"Neural ODE Transformers: Analyzing Internal Dynamics and Adaptive Fine-tuning"**, accepted at the International Conference on Learning Representations (ICLR) 2025.

## Abstract

Recent advancements in large language models (LLMs) based on transformer architectures have sparked significant interest in understanding their inner workings. In this paper, we introduce a novel approach to modeling transformer architectures using highly flexible non-autonomous neural ordinary differential equations (ODEs). Our proposed model parameterizes all weights of attention and feed-forward blocks through neural networks, expressing these weights as functions of a continuous layer index. Through spectral analysis of the model's dynamics, we uncover an increase in eigenvalue magnitude that challenges the weight-sharing assumption prevalent in existing theoretical studies. We also leverage the Lyapunov exponent to examine token-level sensitivity, enhancing model interpretability. Our neural ODE transformer demonstrates performance comparable to or better than vanilla transformers across various configurations and datasets, while offering flexible fine-tuning capabilities that can adapt to different architectural constraints.

## Model Architecture

Our model formulates transformers as neural ODEs with highly flexible non-autonomous vector fields. Instead of shared weights across layers, we parameterize all weights through neural networks that express these weights as functions of a continuous layer index (time).
The model includes:

- Time-dependent weights for attention components (Q, K, V)
- Time-dependent weights for feed-forward networks
- Representation of weights using a time-dependent unit that embeds time information in the Fourier domain

## Experimental Results

### Language Modeling
Comparable or better performance than vanilla transformers across various configurations
![Perplexity](https://raw.githubusercontent.com/SDML-KU/qkvflow/main/assets/llama_vs_our_llama_1b.jpg)

Significant improvements in downstream tasks, particularly in reading comprehension
![Downstream](https://raw.githubusercontent.com/SDML-KU/qkvflow/main/assets/five_shot_llama.png)

### Adaptive finetune
Flexible fine-tuning capabilities that can adapt to different architectural constraints
<div style="display: flex; justify-content: space-between;">
  <img src="assets/lora_wiki.jpg" alt="OWT to Wikitext" width="24%" />
  <img src="assets/full_wiki.jpg" alt="OWT to Wikitext" width="24%" />
  <img src="assets/lora_owt.jpg" alt="Wikitext to OWT" width="24%" />
  <img src="assets/full_owt.jpg" alt="Wikitext to OWT" width="24%" />
</div>

## Implementation
The implementation is built on JAX, utilizing an ecosystem that includes [Equinox](https://github.com/patrick-kidger/equinox), [Haliax](https://github.com/stanford-crfm/haliax), and the [Levanter](https://github.com/stanford-crfm/levanter) framework.

## Citation

If you find this work useful, please consider citing:

```bibtex
@inproceedings{
tong2025neural,
title={Neural {ODE} Transformers: Analyzing Internal Dynamics and Adaptive Fine-tuning},
author={Anh Tong and Thanh Nguyen-Tang and Dongeun Lee and Duc Nguyen and Toan Tran and David Leo Wright Hall and Cheongwoong Kang and Jaesik Choi},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=XnDyddPcBT}
}
```

## Acknowledgements

This work was supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grants funded by the Korea government(MSIT) (No. RS-2019-II190079, Artificial Intelligence Graduate School Program(Korea University); No. RS-2019-II190075, Artificial Intelligence Graduate School Program (KAIST); No. RS-2022-II220984, Development of Artificial Intelligence Technology for Personalized Plug-and-Play Explanation and Verification of Explanation; No. RS-2024-00457882, AI Research Hub Project) and the New Faculty Settlement Research Fund by Korea University. This work was supported by the New Faculty Settlement Research Fund by Korea University and Artificial intelligence industrial convergence cluster development project funded by the Ministry of Science and ICT(MSIT, Korea) & Gwangju Metropolitan City. This work is supported by the Google Cloud Research Credits program with the award GCP19980904 in the early stage of the project.

## License
[MIT License](LICENSE)

qkvflow

qkvflow 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

qkvflow

qkvflow 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README