StreamDiffusion
StreamDiffusion copied to clipboard
Silent error on wrong path
I'm struggling to set an absolute path in the demo to an already downloaded model. I get a 'Triton not found' error but then nothing, when I cancel the process it shows an exception noting that I've supplied an invalid path (I'm on windows).
What is the workflow for pointing to a downloaded model? The included models directory doesn't seem to be used anywhere...
The Triton errors seem to be fine if you're on Windows, they haven't affected my ability to use the tool.
I ran into similar issues. I ended up modifying a few places and got some local models working.
Don't know if there's something I missed though that makes these changes unnecessary.
Also, this is for examples/screen but may be similar for other demos.
I set this value
https://github.com/cumulo-autumn/StreamDiffusion/blob/62596fd39b86a3088e28b2f2c58caa36bc069c50/examples/screen/main.py#L197
To my local model path:
model_id_or_path: str = "F:\AI\img\StreamDiffusion\models\Model\\turbo.safetensors",
And then adjusted the wrapper a little bit.
https://github.com/cumulo-autumn/StreamDiffusion/blob/62596fd39b86a3088e28b2f2c58caa36bc069c50/utils/wrapper.py#L410-L413
Modified that to try from_single_file first:
try: # Load from local directory
pipe: StableDiffusionPipeline = StableDiffusionPipeline.from_single_file(
model_id_or_path,
).to(device=self.device, dtype=self.dtype)
Thank you for your contribution.
Yes thanks - i didn’t have much luck with your advice so I’ll have to dig into it a bit more
On Mon, 25 Dec 2023 at 20:14, Ryslik @.***> wrote:
Thank you for your contribution.
— Reply to this email directly, view it on GitHub https://github.com/cumulo-autumn/StreamDiffusion/issues/45#issuecomment-1869090364, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABSNQQF5CRCVWN7JALZ6VV3YLHGAFAVCNFSM6AAAAABBBP3RQOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNRZGA4TAMZWGQ . You are receiving this because you authored the thread.Message ID: @.***>