Can this run with 8GB of VRAM?

[jdgom]

Hello, this is an honest question,
can this run on windows? I'm getting all sorts of problems trying to runnit in a venv for windows

[hben35096]

Loading the model requires 6GB, with a peak of 10GB during generation and 27GB during decoding. If the value of decode_chunk_size is reduced, it can be used with 12GB.

[wilbur4321]

For the sake of people looking, this value isn't in config. You have to change src/pipelines/pipeline_pose2vud_motenc_long.py line 136:

    def decode_latents_svd(
        self, latents: torch.FloatTensor, decode_chunk_size: int = 25
    ):

My 4090 (24gb vram) in wsl, changing this to 15 worked fine! THANKS!