---
library_name: gguf
base_model: meta-llama/Llama-3.2-3B-Instruct
tags:
- llama
- llama-3
- gguf
- cyberspace
- presales
- vertical-ai
language:
- en
---

# Cyberspace Presales Co-Pilot — Llama 3.2 3B (Q5_K_M)

Fine-tuned Llama 3.2 3B Instruct for generating Cyberspace Limited technical proposals.

## Training
- Base: `meta-llama/Llama-3.2-3B-Instruct`
- Method: LoRA via Unsloth on RunPod A100
- Dataset: 132 structured proposals (real + augmented), ChatML format
- Loss: 1.70 (target range 1.0–1.8)
- Template-bleed mitigation: `train_on_responses_only` with category-specific system prompts

## Usage (Ollama)
```bash
ollama create cyberspace-llama32-3b:latest -f Modelfile
ollama run cyberspace-llama32-3b:latest
```