ZeroModels is a collection of pretrained models built entirely in Keras 3. It spans a broad range of tasks, including image classification, object detection, segmentation, monocular depth estimation, feature extraction, vision-language modeling (VLMs), speech recognition, speech-aware language modeling, text encoding and masked language modeling, large language models (LLMs), text-to-text encoder-decoder modeling, multimodal vision-language generation, and more.
Hi, I’m Gitesh, the creator of ZeroModels.
ZeroModels started as a small side project because I wanted modern pretrained models in Keras without being locked into a single backend. It has since grown into an open-source model ecosystem covering 100+ architectures across LLMs, vision, VLMs, speech, detection, segmentation, depth estimation, and more.
The core idea is simple: write your model code once in Keras 3 and run it on JAX, PyTorch, or TensorFlow.
ZeroModels also handles pretrained weight loading, Hugging Face checkpoint conversion, numerical parity validation, and native INT8, INT4, and FP8 weight-only quantization.
I’ve been building and maintaining it largely solo, and there’s still a lot I want to improve.
I’d especially love feedback from ML engineers and researchers: