Pre-Quantized Checkpoints: Gemma 4 models

[rnett]

Detailed description of the requested feature

It would be great to have gemma 4 models in your published optimized models (i.e. https://huggingface.co/collections/nvidia/inference-optimized-checkpoints-with-model-optimizer).

Timeline

Soon, but not blocking.

Describe alternatives you've considered

Target hardware/use case

For me, a 5080 for local inference. But it would be broadly useful for anyone running gemma 4 models.

[Edwardf0t1]

Thanks @rnett for opening the issue. We already released https://huggingface.co/nvidia/Gemma-4-31B-IT-NVFP4

[rnett]

Thanks for the pointer. Do you only plan on releasing the top end models for each generation? I notice the same thing with Qwen 3.5, where only Qwen3.5-397B-A17B-NVFP4 is released. As someone trying to run local models on Blackwell hardware, smaller models would be great too!