---
title: Celeste Imperia | Hardware-Aware AI Forge
emoji: 🌌
colorFrom: blue
colorTo: purple
sdk: static
pinned: true
thumbnail: >-
  https://cdn-uploads.huggingface.co/production/uploads/697f74cb005a67fc114da2b4/OCwnTMS-RIt9cPQR0Gnxf.png
license: apache-2.0
tags:
- edge-ai
- openvino
- whisper
- sdxl
- quantization
- windows-on-arm
---

# Celeste Imperia | Hardware-Aware AI Forge

**Bridging the gap between frontier AI architectures and consumer silicon.**

Celeste Imperia specializes in the precision optimization and deployment of Large Language Models (LLMs), Multimodal Vision (VLMs), and Speech architectures for edge hardware. We provide high-fidelity, hardware-validated weights optimized for private, zero-latency execution.

---

## 🛠️ The Development Forge
All models are validated on our standard consumer-grade benchmark rig to ensure "masses-ready" stability:
- **Processor:** Intel Core i5-11400
- **Compute:** NVIDIA RTX A4000 (16GB VRAM)
- **Memory:** 40GB RAM
- **Validation:** Every port undergoes strict logic consistency and vision-token alignment checks.

---

## 📂 Active Repositories

### 🧠 Language Reasoning (GGUF Trinity)
Optimized for the `llama.cpp` ecosystem, providing **Master (FP16)**, **Pro (Q8_0)**, and **Mobile (Q4_K_M)** weights.
* **[Llama-3.2-1B-Instruct-GGUF](https://huggingface.co/CelesteImperia/Llama-3.2-1B-Instruct-GGUF)** - Meta's industry standard for edge reasoning.
* **[Phi-3.5-mini-instruct-GGUF](https://huggingface.co/CelesteImperia/Phi-3.5-mini-instruct-GGUF)** - Microsoft’s 128k context logic powerhouse.

### 🎙️ Speech & Audio (OpenVINO Optimized)
High-speed transcription and translation optimized for Intel CPUs/NPUs.
* **[Whisper-Large-V3-Turbo-OpenVINO](https://huggingface.co/CelesteImperia/whisper-large-v3-turbo-openvino)** - **[LATEST]** Near real-time transcription with 8-bit quantization for Intel hardware.

### 🎨 Generative Vision (Diffusion)
* **[SDXL-OpenVINO-Trinity](https://huggingface.co/CelesteImperia/celeste-imperia-sdxl-openvino)** - **[LATEST]** Full 4-step generation with fused LCM and TinyVAE. Optimized for i5/i7 hardware.
* **[Qwen2-VL-2B-Instruct-OpenVINO-INT4](https://huggingface.co/CelesteImperia/Qwen2-VL-2B-Instruct-OpenVINO-INT4)** - State-of-the-art vision-language reasoning for OCR and scene analysis.

---

## 🚀 Optimization Matrix

| Platform Target | Optimization Backend | Architecture Focus |
| :--- | :--- | :--- |
| **Snapdragon X Elite** | GGUF / QNN / ONNX | ARM64 high-speed inference |
| **Intel Core Ultra / Arc** | OpenVINO / NPU | Low-power background execution |
| **Edge CPUs (Mobile/Linux)** | GGUF (INT4/INT8) | Resource-constrained "Agent" logic |

---

**New here?** Check out our **[Getting Started Guide](https://huggingface.co/CelesteImperia/SDXL-Base)** to find the right model for your CPU!

**Connect with the architect:** [Abhishek Jaiswal on LinkedIn](https://www.linkedin.com/in/abhishek-jaiswal-524056a/)