Update README.md
Browse files
README.md
CHANGED
|
@@ -4,7 +4,53 @@ emoji: 🏃
|
|
| 4 |
colorFrom: pink
|
| 5 |
colorTo: gray
|
| 6 |
sdk: static
|
| 7 |
-
pinned:
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
colorFrom: pink
|
| 5 |
colorTo: gray
|
| 6 |
sdk: static
|
| 7 |
+
pinned: true
|
| 8 |
---
|
| 9 |
|
| 10 |
+
# ⚡ Neuronic
|
| 11 |
+
|
| 12 |
+
**High-Performance Agents for the Edge.**
|
| 13 |
+
|
| 14 |
+
Neuronic is a specialized AI research and development lab focused on building hyper-optimized, low-parameter models for function calling, tool use, and agentic routing. We believe that intelligence doesn't always require massive compute—it requires precision.
|
| 15 |
+
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
## 🎯 Our Focus
|
| 19 |
+
|
| 20 |
+
* **Function-Calling Specialists:** We train models to act as deterministic routers, prioritizing strict syntax, reliable JSON output, and accurate tool utilization over general chat.
|
| 21 |
+
* **Edge Computing:** By optimizing models in the 0.5B to 7B parameter range, we make local, high-speed agentic workflows possible on edge devices, mobile hardware, and consumer GPUs.
|
| 22 |
+
* **Low Latency:** Designed for environments where every millisecond counts.
|
| 23 |
+
|
| 24 |
+
---
|
| 25 |
+
|
| 26 |
+
## 🚀 Featured Models
|
| 27 |
+
|
| 28 |
+
### **Nero1-0.5B**
|
| 29 |
+
Our flagship router. A 500-million parameter model fine-tuned specifically for agentic tool use and structural reliability.
|
| 30 |
+
* **Base Model:** `Qwen/Qwen2.5-Coder-0.5B-Instruct`
|
| 31 |
+
* **Dataset:** smirki/Agentic-Coding-Tessa
|
| 32 |
+
* **Use Case:** Local Python toolkits, fast API routing, and strictly typed outputs.
|
| 33 |
+
|
| 34 |
+
### **Nero-0.5B-GGUF**
|
| 35 |
+
The quantized versions of Nero for immediate deployment on CPU or low-VRAM environments.
|
| 36 |
+
* **Formats:** `Q8_0` (Maximum accuracy), `Q4_K_M` (Maximum speed)
|
| 37 |
+
* **Framework:** Ready for `llama.cpp` and LM Studio.
|
| 38 |
+
|
| 39 |
+
---
|
| 40 |
+
|
| 41 |
+
## 🛠️ Usage Guidelines
|
| 42 |
+
|
| 43 |
+
Neuronic models are heavily biased toward action over conversation. For optimal performance:
|
| 44 |
+
1. **Use ChatML formatting.**
|
| 45 |
+
2. **Define tools explicitly** in your system prompt.
|
| 46 |
+
3. **Keep temperatures low** (`0.1 - 0.3`) to prevent structural hallucinations during complex JSON generation.
|
| 47 |
+
|
| 48 |
+
---
|
| 49 |
+
|
| 50 |
+
## 🔗 Links & Resources
|
| 51 |
+
|
| 52 |
+
* [GitHub] *(Coming Soon)*
|
| 53 |
+
* [Documentation] *(Coming Soon)*
|
| 54 |
+
* [Community Discord](#) *(Coming Soon)*
|
| 55 |
+
|
| 56 |
+
> *"Precision over parameters."*
|