TESS 2: A Large-Scale Generalist Diffusion Language Model

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tae, Jaesung, Ivison, Hamish, Kumar, Sachin, Cohan, Arman
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918042525499392
author Tae, Jaesung
Ivison, Hamish
Kumar, Sachin
Cohan, Arman
author_facet Tae, Jaesung
Ivison, Hamish
Kumar, Sachin
Cohan, Arman
contents We introduce TESS 2, a general instruction-following diffusion language model that outperforms contemporary instruction-tuned diffusion models, as well as matches and sometimes exceeds strong autoregressive (AR) models. We train TESS 2 by first adapting a strong AR model via continued pretraining with the usual cross-entropy as diffusion loss, and then performing further instruction tuning. We find that adaptation training as well as the choice of the base model is crucial for training good instruction-following diffusion models. We further propose reward guidance, a novel and modular inference-time guidance procedure to align model outputs without needing to train the underlying model. Finally, we show that TESS 2 further improves with increased inference-time compute, highlighting the utility of diffusion LMs in having fine-grained controllability over the amount of compute used at inference time. Code and models are available at https://github.com/hamishivi/tess-2.
format Preprint
id arxiv_https___arxiv_org_abs_2502_13917
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle TESS 2: A Large-Scale Generalist Diffusion Language Model
Tae, Jaesung
Ivison, Hamish
Kumar, Sachin
Cohan, Arman
Computation and Language
We introduce TESS 2, a general instruction-following diffusion language model that outperforms contemporary instruction-tuned diffusion models, as well as matches and sometimes exceeds strong autoregressive (AR) models. We train TESS 2 by first adapting a strong AR model via continued pretraining with the usual cross-entropy as diffusion loss, and then performing further instruction tuning. We find that adaptation training as well as the choice of the base model is crucial for training good instruction-following diffusion models. We further propose reward guidance, a novel and modular inference-time guidance procedure to align model outputs without needing to train the underlying model. Finally, we show that TESS 2 further improves with increased inference-time compute, highlighting the utility of diffusion LMs in having fine-grained controllability over the amount of compute used at inference time. Code and models are available at https://github.com/hamishivi/tess-2.
title TESS 2: A Large-Scale Generalist Diffusion Language Model
topic Computation and Language
url https://arxiv.org/abs/2502.13917