126 lines
4.3 KiB
Markdown
126 lines
4.3 KiB
Markdown
---
|
||
license: apache-2.0
|
||
language:
|
||
- en
|
||
tags:
|
||
- rag
|
||
- retrieval-augmented-generation
|
||
- knowledge-base
|
||
- emotion-detection
|
||
- dual-stage-retrieval
|
||
- coding
|
||
- math
|
||
- science
|
||
- history
|
||
- nyxis
|
||
- quantasparklabs
|
||
pipeline_tag: text-generation
|
||
base_model:
|
||
- QuantaSparkLabs/NYXIS-1.1B
|
||
library_name: transformers
|
||
datasets:
|
||
- QuantaSparkLabs/NYXIS-AEGIS-Knowledge
|
||
---
|
||
<p align="center">
|
||
<img src="https://huggingface.co/QuantaSparkLabs/NYXIS-Pro/resolve/main/preview imgagee.png"
|
||
alt="NYXIS Logo"
|
||
width="160"
|
||
height="160"
|
||
style="border-radius: 50%; object-fit: cover;">
|
||
</p>
|
||
|
||
<p align="center">
|
||
<img src="https://huggingface.co/QuantaSparkLabs/NYXIS-Pro/resolve/main/logoname.png"
|
||
alt="NYXIS Name"
|
||
width="700"
|
||
style="border-radius: 18px;">
|
||
</p>
|
||
# NYXIS-PRO 🛡️
|
||
|
||
<p align="center">
|
||
<a href="https://huggingface.co/QuantaSparkLabs/NYXIS-1.1B"><img src="https://img.shields.io/badge/Base-NYXIS--1.1B-purple" alt="Base Model"></a>
|
||
<a href="https://huggingface.co/QuantaSparkLabs/NYXIS-AEGIS-KB"><img src="https://img.shields.io/badge/Knowledge-AEGIS-blue" alt="AEGIS KB"></a>
|
||
<a href="#"><img src="https://img.shields.io/badge/Retrieval-Dual%20Stage-green" alt="Dual Stage"></a>
|
||
<a href="#"><img src="https://img.shields.io/badge/Emotion-1--10%20Scale-orange" alt="Emotion Engine"></a>
|
||
<a href="https://www.apache.org/licenses/LICENSE-2.0"><img src="https://img.shields.io/badge/License-Apache%202.0-yellow" alt="License"></a>
|
||
</p>
|
||
|
||
**The all-in-one reasoning engine with the AEGIS knowledge shield.**
|
||
|
||
NYXIS-PRO combines the NYXIS1.1B base model with a curated 5,000+ chunk knowledge base covering coding, math, science, history, and common knowledge. A dual‑stage retriever grounds every factual answer in verified data — **zero hallucination.**
|
||
|
||
## 🧠 What Makes It Different
|
||
|
||
| Feature | Description |
|
||
|---------|-------------|
|
||
| **AEGIS Knowledge Base** | 5,426 curated Wikipedia, Trivia QA, and ArXiv chunks |
|
||
| **Dual‑Stage Retriever** | Dense recall (bge‑small) + Cross‑encoder rerank (ms‑marco‑MiniLM) |
|
||
| **Emotion Engine** | Detects user emotion on a 1‑10 scale and matches tone |
|
||
| **4‑Tier Knowledge Cascade** | AEGIS → External Search → Model Brain → Honest Fallback |
|
||
| **Zero Hallucination** | If it doesn't know, it says so — no fabrication |
|
||
|
||
## 🚀 Quick Start
|
||
|
||
```python
|
||
from huggingface_hub import snapshot_download
|
||
import sys
|
||
|
||
model_dir = snapshot_download("QuantaSparkLabs/NYXIS-Pro")
|
||
sys.path.insert(0, model_dir)
|
||
|
||
from pipeline import NYXISPro
|
||
nyxis = NYXISPro(model_dir)
|
||
|
||
result = nyxis.generate("What is a binary search tree?")
|
||
print(result["response"])
|
||
```
|
||
|
||
> **Requirements:** `sentence-transformers`, `faiss-cpu`, `transformers`, `accelerate`, `bitsandbytes`
|
||
> **VRAM:** ~2.1 GB (4‑bit) | **Hardware:** T4 or better
|
||
|
||
## 📦 What's Inside
|
||
|
||
| Component | Model | Purpose |
|
||
|-----------|-------|---------|
|
||
| **Base LLM** | NYXIS1.1B) | Text generation |
|
||
| **Dense Retriever** | bge‑small‑en‑v1.5 | Initial candidate recall |
|
||
| **Cross‑Encoder** | ms‑marco‑MiniLM‑L‑6‑v2 | Precision re‑ranking |
|
||
| **FAISS Index** | 5,426 chunks | Knowledge storage |
|
||
| **Emotion Engine** | Custom keyword‑based | Tone matching |
|
||
|
||
## 🤖 Emotion Intelligence
|
||
|
||
NYXIS-PRO detects your emotional state from text and adjusts its personality:
|
||
|
||
| Level | Emotion | Response Style |
|
||
|:---:|---------|----------------|
|
||
| 1‑3 | Sad / Angry | Gentle, supportive, empathetic |
|
||
| 4‑6 | Neutral | Balanced, warm, helpful |
|
||
| 7‑9 | Happy / Excited | Energetic, playful |
|
||
| 10 | Overjoyed | Celebratory |
|
||
|
||
## 🛡️ Knowledge Cascade
|
||
|
||
When you ask a factual question, NYXIS-PRO follows a strict 4‑tier protocol:
|
||
|
||
1. **AEGIS** — Searches internal verified knowledge base
|
||
2. **External Search** — Falls back to web search (user‑provided API)
|
||
3. **Model Brain** — Uses training data with honesty guard
|
||
4. **Honest Fallback** — Admits "I don't have enough information"
|
||
|
||
It **never fabricates** an answer.
|
||
|
||
## Limitations
|
||
|
||
- Knowledge base covers ~5,400 chunks — not infinite
|
||
- Emotion detection is keyword‑based, not deep sentiment analysis
|
||
- External search requires user‑provided API function
|
||
- English‑only, 1.5B model size limits complex reasoning
|
||
|
||
## License
|
||
|
||
Apache‑2.0
|
||
|
||
---
|
||
|
||
*Built with 🛡️ by QuantaSparkLabs* |