fp8 quantized Newbie-image for ComfyUI.

All credit belongs to the original model author. License is the same as the original model.

TensorCoreFP8 (tcfp8):

Scaled fp8 + Mixed precision + Hardware fp8 support

On supported GPU, ComfyUI will automatically do calculations in FP8 directly, instead of dequantizing + BF16. torch.compile is recommended, if you can get it up and running.

More info about hw fp8: https://civarchive.com/models/2172944

Old:

Scaled fp8 + Mixed precision. Does not support hardware fp8 (no calibration data).

Gemma 3 4b:

Scaled fp8 + Mixed precision.

Jina clip (TBD):

Jina clip is very small, seems not necessary.