MiniMax-H3 VAE decoder loads in fp32 but CUDA decode runs it in fp16

[asomoza]

Describe the bug

The VAE keeps everything in fp32 no matter what dtype you pass, but the decode block wraps vae.decode in a fp16 autocast, so the decoder gets downcast to fp16 anyway on every call. On CUDA this just costs VRAM, RAM and disk space.

Decoding 192 frames at 1344x768:

	weights	peak
main	9.70 GiB	16.69 GiB
decode in fp16	5.19 GiB	7.67 GiB

As seen on the table, this prevents to use this model on 16GB GPUs for that resolution and duration.

Reproduction

Just normal inference on a cuda GPU will reproduce it

import torch

from diffusers import ComponentsManager, ModularPipeline


manager = ComponentsManager()
manager.enable_auto_cpu_offload(device="cuda")

pipe = ModularPipeline.from_pretrained("MiniMaxAI/MiniMax-H3", components_manager=manager)
pipe.load_components(workflow="t2va", dtype=torch.bfloat16)

results = pipe(
    prompt="An astronaut hiking through the mountains, humming a tune",
    num_frames=192,
    output=["videos", "audio", "sampling_rate"],
)

code to test a PoC fix:

import torch

from diffusers import AutoencoderKLMiniMaxH3


class Fp16DecoderVae(AutoencoderKLMiniMaxH3):
    _keep_in_fp32_modules = [
        "encoder",
        "quant_conv",
        "post_quant_conv",
        "norm1",
        "norm2",
        "norm_out",
        "scale1",
        "scale2",
    ]

    @classmethod
    def from_pretrained(cls, *args, **kwargs):
        kwargs.pop("torch_dtype", None)
        kwargs["dtype"] = torch.float16
        return super().from_pretrained(*args, **kwargs)


vae = Fp16DecoderVae.from_pretrained("MiniMaxAI/MiniMax-H3", subfolder="vae", dtype=torch.bfloat16)

main

decoded_fp32pin.mp4

PoC fix

decoded_fp16decoder.mp4

System Info

diffusers from main
any cuda platform

Who can help?

[yiyixuxu]

thanks @asomoza!
feel free to open a PR if you verify that the output is identical