I used to have the exact same issue with 3.5 model. Upon some trial and error it appears that the use_auth_token argument might be already deprecated.

However I managed to resolve the issue by following steps here: https://stability.ai/learning-hub/setting-up-and-using-sd3-medium-locally

In particular I executed huggingface-cli login from the console with the same virtualenv as my jupyter runtime. It asks you to paste the token to console and confirm a few details. Upon restarting the runtime a plain command went withnout any issues:

pipe = diffusers.StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3.5-medium",
    torch_dtype=torch.bfloat16)