Install: https://github.com/KONAKONA666/q8_kernels

Usage

import torch
from q8_kernels.models.T5EncoderFP8 import T5EncoderModelFP8
text_encoder = T5EncoderModelFP8.from_pretrained(
    "konakona/t5xxl_encoder_fp8", torch_dtype=torch.bfloat16
)

Needed dX for token training(PTI and textual inversion) in LTX.
dX is in bf16, calculations in FP8
Needs Ada GPU

Downloads last month
11
Safetensors
Model size
4.76B params
Tensor type
F32
BF16
F8_E4M3
Inference API
Unable to determine this model鈥檚 pipeline type. Check the docs .