---
license: apache-2.0
---
# Typhoon2.5-Qwen3-30B-A3B (GGUF)

GGUF-format conversions of the [**scb10x/typhoon2.5-qwen3-30b-a3b**](https://huggingface.co/scb10x/typhoon2.5-qwen3-30b-a3b) model for efficient inference with **llama.cpp** and compatible runtimes.  

Converted using `llama.cpp`’s `convert_hf_to_gguf.py` and quantize tool. No additional training or fine-tuning was performed.

---
## 🧩 Variants

| Variant  | File size |
|----------|-----------|
| **BF16** | 61.1 GB   |
| **F16**  | 61.1 GB   |
| **Q8_0** | 32.5 GB   |
| **Q_K_M**| 18.6 GB   |

---
## 📝 Notes

- Source model: [**scb10x/typhoon2.5-qwen3-30b-a3b**](https://huggingface.co/scb10x/typhoon2.5-qwen3-30b-a3b)
- Tokenizer and metadata preserved during conversion
- Choose **BF16** for best fidelity, **F16** for GPU does not support BF16, **Q8_0** for balance, **Q4_K_M** for lowest memory

---
## ⚖️ License

Weights inherit the upstream model’s license.  
This repository redistributes format-converted copies only.  
Please review and comply with the upstream terms before use.

---
## 📝 Acknowledgments

Original model by **SCB10X**.  
GGUF conversion with quantization performed via **llama.cpp** tooling.