CantorAI ModelZoo 🦓

This repository is the central hub for hosting pre-trained models, optimized weights, and AI assets used across the CantorAI ecosystem.

Because many models require massive data files (often exceeding GitHub's 100MB file limit), the core repository acts as a lightweight index. The actual model binaries are distributed as packaged archives on the Releases page.

📦 Getting Started

To download models for deployment:

1. Navigate to the Releases page.
2. Download the ZIP or TAR archive for the specific model suite you need (e.g., xworld_onnx_models_small_v1.zip).
3. Extract the contents directly into your application's models/ directory.

⚙️ Model Formats

The ModelZoo hosts various architectures tailored for different deployment targets:

• .onnx: Framework-agnostic graphs ready to be compiled into hardware-specific TensorRT .engine files.
• .pth / .pt: Native PyTorch checkpoints for fine-tuning or direct inference.
• .safetensors: Secure, fast-loading weight formats for Large Language Models.

⚖️ Licensing

Models hosted in this repository are converted or fine-tuned from open-source baselines. Unless otherwise specified in a specific release, the models are distributed under the permissive Apache 2.0 License.

When using these models, please respect the original licenses of the foundational architectures (e.g., Google SigLIP, Megvii YOLOX, Qwen, etc.).