why only release gguf format for qat model.
why don't release qat safetensors format model, since we can convert safetensors to gguf, and the safetensors format is "common".
Hi @Fouye , Apologies for the delay,
This a very insightful. You're right that safetensors is a common and excellent format for model distribution, and it's the standard for many models on Hugging Face.
We chose to release the QAT models directly in GGUF format to provide the most optimized and ready-to-use experience for the llama.cpp community. It streamlines the process and ensures a high-quality, pre-quantized file. We appreciate the feedback and will consider adding the safetensors format in future releases.
Thank you.
@fouye Maybe this is what you are looking for https://huggingface.co/google/gemma-3-4b-it-qat-q4_0-unquantized