#4949th of 52 in Open modelsFor image, video and design, image and design
bagel
Open multimodal model for image understanding, text-to-image and image editing
Live demo ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 6.2k
- License
- Apache-2.0
- Last commit
- May 2026
- Language
- Python
5,720 on 3 Apr 20266,205 on 30 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
BAGEL is a 7B-active (14B total) Mixture-of-Transformer-Experts model that handles visual understanding, text-to-image generation and image editing in one checkpoint. The repo provides inference code, a Gradio web app with NF4 and INT8 quantization modes, and training and evaluation scripts. Weights are on Hugging Face as BAGEL-7B-MoT.
Who it is for: Researchers running or tuning a unified image understanding/generation model
Strengths
- One checkpoint covers understanding, generation and editing; reports GenEval 0.82 and MMBench 85.0
- Includes training (scripts/train.sh) and evaluation scripts for VLM, T2I and editing benchmarks
- Gradio app offers an NF4 mode for 12-32GB VRAM GPUs
- Apache-2.0; community ComfyUI node, Dockerfiles and DF11/INT8 weights exist
Weaknesses
- Full-precision app needs a 32GB+ VRAM GPU or multiple GPUs
- Requires flash_attn==2.5.8, which complicates setup
- README advises against the INT8 mode; quantized inference is community-contributed
- No tagged release; Dockerfiles are linked from issues, not shipped in the repo
What it needs
- GPU required
- Needs Python 3.10, conda, flash_attn 2.5.8, Hugging Face Hub
- Models: BAGEL-7B-MoT safetensors, NF4, INT8, DF11
Quick answers
Is bagel open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Does bagel need a GPU?
Yes, it needs a GPU.
Also in Open models
See all 52| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 95 out of 100 |
| 2 | llmfitRust tool that checks which LLMs fit your RAM, CPU and GPU | 87 out of 100 |
| 3 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 86 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 77 out of 100 |
| #5 | yolov5PyTorch YOLOv5 object detection, segmentation and classification models with training code | 75 out of 100 |