Gemma 3 12B quantization-aware training for NVFP4
A Gemma 3 configuration that includes NVFP4 weight and activation quantization in the training setup.
Published by Axolotl
How to use it
The example trains Gemma 3 12B IT on Alpaca and configures NVFP4 for weights and activations. It includes the Liger plugin and has a matching baseline configuration in the same publisher directory.
Evaluate the quantized and baseline configurations on the same held-out tasks. Confirm that your runtime supports the configured quantization path and that exported checkpoints load where you intend to serve them.
- Framework
- Axolotl
- Dataset
- tatsu-lab/alpaca
- Hardware
- Minimum VRAM unspecified. Verify NVFP4 support in your stack.
- Access
- Public publisher configuration; model and dataset terms apply.