Instructions to use vedaco/Tera.v3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Keras
How to use vedaco/Tera.v3 with Keras:
# Available backend options are: "jax", "torch", "tensorflow". import os os.environ["KERAS_BACKEND"] = "jax" import keras model = keras.saving.load_model("hf://vedaco/Tera.v3") - Notebooks
- Google Colab
- Kaggle
| language: en | |
| license: other | |
| tags: | |
| - multimodal | |
| - tensorflow | |
| - keras | |
| - sovereign-intelligence | |
| - dense-elite | |
| pipeline_tag: image-text-to-text | |
| # Tera.V3: Sovereign Intelligence | |
| <p align="center"> | |
| <img src="https://huggingface.co/vedaco/Tera.v3/resolve/main/star_logo.png" width="200"> | |
| </p> | |
| **Tera.V3** is a high-efficiency Multimodal "Dense-Elite" architecture designed for private, sovereign deployment. (Proprietary / Non-Open Source) | |
| ### Key Features | |
| - **Multimodal**: Integrated `TeraVisionEncoder` for simultaneous image and text processing. | |
| - **Sovereign Efficiency**: Designed to outperform massive models through architectural precision. | |
| - **Stable Core**: Utilizes `LogSquaredReLU` and `TokenShift` for deep sequential modeling. | |
| ### Model Details | |
| - **Optimizer**: LionSovereign | |
| - **Architecture**: Multimodal Transformer-style | |
| - **Status**: Pre-trained on synthetic multimodal data (5,000 steps) | |
| ### Usage | |
| Weights are stored in `.weights.h5` format for Keras 3 / TensorFlow compatibility. | |