license: mit
pipeline_tag: text-generation
ELF: Embedded Language Flows
Embedded Language Flows (ELF) is a class of continuous diffusion language models based on continuous-time Flow Matching. Unlike existing diffusion language models (DLMs), ELF predominantly stays within the continuous embedding space until the final time step, where it maps to discrete tokens using a shared-weight network. This formulation makes it straightforward to adapt established techniques from image-domain diffusion models, such as classifier-free guidance (CFG).
- Paper: ELF: Embedded Language Flows
- Repository: https://github.com/lillian039/ELF
Description
Experiments show that ELF substantially outperforms leading discrete and continuous DLMs, achieving better generation quality with fewer sampling steps. These results suggest that ELF offers a promising path toward effective continuous DLMs.
Usage
To use these models, please follow the installation instructions in the official repository.
Inference and Evaluation
You can run evaluation for unconditional generation (e.g., using ELF-B) with the following command:
cd src/
python eval.py \
--config configs/training_configs/train_owt_ELF-B.yml \
--checkpoint_path embedded-language-flows/ELF-B-owt
For conditional tasks like translation or summarization, use the corresponding configuration files:
# XSum (summarization)
python eval.py \
--config configs/training_configs/train_xsum_ELF-B.yml \
--checkpoint_path embedded-language-flows/ELF-B-xsum
# WMT14 De-En (translation)
python eval.py \
--config configs/training_configs/train_de-en_ELF-B.yml \
--checkpoint_path embedded-language-flows/ELF-B-de-en
Citation
@article{elf2026,
title={ELF: Embedded Language Flows},
author={Hu, Keya and Qiu, Linlu and Lu, Yiyang and Zhao, Hanhong and Li, Tianhong and Kim, Yoon and Andreas, Jacob and He, Kaiming},
journal={arXiv preprint arXiv:2605.10938},
year={2026}
}