Update README.md

README.md

@@ -15,4 +15,78 @@ tags:
 - gguf
 - llama.cpp
 - inference-optimization
----
+---
+
+# DeepSeekV4Flash Quantization Repository
+
+![v4-benchmark-2](https://cdn-uploads.huggingface.co/production/uploads/671ab90d28ec35263e09152f/WXhyPJ5E8r3B2p0TO8us6.png)
+
+This repository provides scripts and guidelines for quantizing the **DeepSeek V4 Flash** model, enabling reduced model size and optimized inference performance.
+
+![v4-efficiency](https://cdn-uploads.huggingface.co/production/uploads/671ab90d28ec35263e09152f/m4HSN3MmYyW2SHZytAbFE.png)
+
+---
+
+## 🚀 Purpose
+- Reduce model size (BF16 → Q3/Q4/Q5/Q8, etc.)
+- Improve inference speed
+- Enable deployment on limited GPU/CPU resources
+
+---
+
+## 🌍 Languages
+- English (en)
+- Vietnamese (vi)
+
+---
+
+## 🧠 Base Model
+- DeepSeek-V4-Flash (https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash)
+
+---
+
+## 📦 Contents
+- Model conversion and quantization scripts
+- Usage examples for llama.cpp / GGUF workflows
+- Common quantization configurations
+
+---
+
+## 🛠️ Requirements
+- Python >= 3.12
+- Latest version of llama.cpp (with GGUF support)
+- HuggingFace Transformers (if converting from HF format)
+- Sufficient RAM/VRAM depending on model size
+
+---
+
+## ⚙️ Example Usage
+
+```bash
+python convert_hf_to_gguf.py   --model deepseek-ai/DeepSeek-V4-Flash   --outfile models/DeepSeekV4Flash.gguf
+
+./llama-quantize models/DeepSeekV4Flash.gguf Q4_K_M
+```
+
+---
+
+## 📌 Notes
+- Quantization may require significant system memory depending on model size
+- Some quantization formats may not be compatible with all runtimes or versions
+- Always validate output quality after quantization
+
+---
+
+## 👤 Author
+
+- Email: tecaprovn@gmail.com
+- Telegram: https://t.me/tamndx
+
+---
+
+## 📄 License
+
+This repository follows the original DeepSeek model license.
+
+- Base model: Apache 2.0 (DeepSeek)
+- Only conversion scripts included, no weight modification