Update README.md

README.md

@@ -1,3 +1,120 @@
----
-license: mit
----
+---
+license: mit
+language:
+- en
+tags:
+- gguf
+- llama.cpp
+- qwen
+- instruct
+- lora
+- roleplay
+- christmas
+- holiday
+- storytelling
+pipeline_tag: text-generation
+---
+
+<!--
+GHOSTAI • CHRISTMAS SPIRIT GGUF (2B)
+Professional Hugging Face README for a GGUF release.
+-->
+
+<p align="center">
+  <img src="https://capsule-render.vercel.app/api?type=waving&color=0:0b1d12,50:8b0000,100:0b1d12&height=170&section=header&text=GHOSTAI%20—%20CHRISTMAS%20SPIRIT%20GGUF%20(2B)&fontSize=40&fontColor=F7F7F7&animation=twinkling" />
+</p>
+
+<p align="center">
+  <img alt="Format" src="https://img.shields.io/badge/Format-GGUF-0B1D12?style=for-the-badge">
+  <img alt="Runtime" src="https://img.shields.io/badge/Runtime-llama.cpp-14532D?style=for-the-badge">
+  <img alt="Model Size" src="https://img.shields.io/badge/Model-2B-1F6FEB?style=for-the-badge">
+  <img alt="Theme" src="https://img.shields.io/badge/Theme-Christmas%20Spirit-8B0000?style=for-the-badge">
+</p>
+
+<p align="center">
+  <strong>GHOSTAI — Christmas Spirit GGUF (2B)</strong><br/>
+  A compact, holiday-forward <strong>2B</strong> model tuned for cozy storytelling, cheerful roleplay, and warm seasonal vibes — distributed in <strong>GGUF</strong> format for the <strong>llama.cpp</strong> ecosystem.<br/>
+  <sub>Optimized for low-RAM systems while retaining strong instruction-following and narrative tone.</sub>
+</p>
+
+---
+
+## Overview
+
+**GHOSTAI: Christmas Spirit (2B)** is a lightweight, personality-driven model focused on:
+
+- Cozy winter storytelling  
+- Festive dialogue and roleplay  
+- Family-friendly holiday themes  
+- Cheerful, wholesome tone control  
+
+This release targets **efficient inference** on laptops, desktops, and small servers while maintaining a consistent seasonal voice.
+
+The model is distributed as **GGUF**, making it compatible with:
+- CPU-only inference
+- GPU offload (CUDA / Metal / Vulkan via llama.cpp)
+
+Quantization choice controls memory and speed. GPU usage is configured at runtime.
+
+---
+
+## Base model
+
+- **Base:** Qwen-2.5-2B-Instruct  
+- **Training:** LoRA fine-tune, merged and exported to GGUF  
+- **Format:** llama.cpp GGUF v3+
+
+---
+
+## Files (this release)
+
+Approximate sizes for a 2B model. Actual sizes may vary slightly by build.
+
+| File | Quant | Approx size | Rough RAM needed (4k ctx) |
+|---|---:|---:|---:|
+| `christmas_qwen2b_v1.f16.gguf` | f16 | ~4.0 GB | ~5–6 GB |
+| `christmas_qwen2b_v1.Q8_0.gguf` | Q8_0 | ~2.3 GB | ~3.5–4 GB |
+| `christmas_qwen2b_v1.Q6_K.gguf` | Q6_K | ~1.8 GB | ~3 GB |
+| `christmas_qwen2b_v1.Q5_K_M.gguf` | Q5_K_M | ~1.6 GB | ~2.5–3 GB |
+| `christmas_qwen2b_v1.Q4_K_M.gguf` | Q4_K_M | ~1.4 GB | ~2–2.5 GB |
+| `christmas_qwen2b_v1.Q3_K_M.gguf` | Q3_K_M | ~1.1 GB | ~2 GB |
+| `christmas_qwen2b_v1.Q2_K.gguf` | Q2_K | ~0.9 GB | ~1.5–2 GB |
+
+**RAM notes**
+- Assumes ~4k context.
+- For 8k context, add ~0.5–1 GB.
+- GPU offload can move some memory to VRAM, but system RAM is still required.
+
+---
+
+## Recommended downloads
+
+- **Best default:** `Q4_K_M`
+- **Higher quality:** `Q5_K_M`, `Q6_K`, `Q8_0`
+- **Low RAM systems:** `Q3_K_M`, `Q2_K`
+- **Maximum fidelity:** `f16`
+
+---
+
+## Prompting style
+
+The model responds best to:
+- Clear role or tone definition
+- Warm, descriptive prompts
+- Explicit seasonal framing
+
+Example system prompt:
+> *You are GHOSTAI Christmas Spirit. Your responses are cozy, kind, and festive, with a warm holiday tone.*
+
+---
+
+## Quickstart (llama.cpp)
+
+### CPU-only
+
+```bash
+./llama-cli \
+  -m christmas_qwen2b_v1.Q4_K_M.gguf \
+  -ngl 0 \
+  -c 4096 \
+  -p "You are GHOSTAI Christmas Spirit. Write a cozy winter story set in a quiet town preparing for Christmas Eve."