| | --- |
| | license: mit |
| | organization_profile: true |
| | --- |
| | |
| | # 🏛️ Celeste Imperia | High-Efficiency AI Weights |
| |
|
| | **The official repository for local-first, hardware-optimized AI models.** ⚡ |
| |
|
| | This organization is dedicated to making advanced AI accessible on consumer hardware. We specialize in porting heavy encoders and LLMs to run on **NPUs** and **ARM** architectures. |
| |
|
| | ### 📦 What you’ll find here: |
| | * **NPU-Optimized Encoders:** CLIP and T5 variants converted for **Intel OpenVINO** and **Qualcomm AI Stack**. |
| | * **Consistent Character LoRAs:** High-fidelity character models trained for perfect persistence across frames. |
| | * **Edge-Ready LLMs:** Quantized and ported models specifically tuned for local CPU/NPU inference. |
| |
|
| | ### 🛠️ Hardware Focus |
| | Our models are tested and optimized on local rigs (RTX A4000) to ensure they work for creators, not just data centers. |
| |
|
| | --- |
| | 📫 **Inquiries:** [celesteimperia@gmail.com](mailto:celesteimperia@gmail.com) |
| | "Forging the future of Edge AI, one model at a time." |
