---
library_name: safetensors
license: other
license_name: bagel-derived
license_link: https://huggingface.co/ByteDance-Seed/BAGEL-7B-MoT
pipeline_tag: any-to-any
tags:
- multimodal
- any-to-any
- modus
gated: manual
extra_gated_heading: Access MODUS 16-modality weights
extra_gated_prompt: These weights are derived from ByteDance BAGEL-7B-MoT and fine-tuned
  on the MODUS 15-modality dataset. By requesting access you agree to use them for
  research purposes only, consistent with the upstream BAGEL license.
extra_gated_fields:
  Name: text
  Affiliation: text
  I will use these weights for research purposes only: checkbox
extra_gated_button_content: Request access
---

# MODUS

MODUS for any-to-any generation (15 aligned modalities) as presented in [MODUS: Decoder-Only Any-to-Any Modeling of Diverse Modalities](https://huggingface.co/papers/2607.25948).

Project page: https://modus-multimodal.epfl.ch/
Code: https://github.com/EPFL-VILAB/Modus

**Inference config (important):**
- modality config: `conf/modalities/instruction_16mod_stage2.yaml`
- weights are **bf16**.

## Files
- `model.safetensors` — trained weights (bf16)
- `ae.safetensors` — VAE (image decode)
- `config.json` / `llm_config.json` / `vit_config.json` — architecture config
- `vocab.json` / `merges.txt` / `tokenizer_config.json` — tokenizer