StableCascade image generation #1979
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
aleksandr-mokrov
requested review from
a team and
apaniukov
and removed request for
a team
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2024-04-30T11:28:37Z "nncf>=2.10.0" aleksandr-mokrov commented on 2024-05-01T11:01:18Z Updated |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2024-04-30T11:28:37Z Line #7. prior = StableCascadePriorPipeline.from_pretrained("stabilityai/stable-cascade-prior", torch_dtype=torch.float32)
do we really need to load and run pt model every time? I think we can reduce running time and memory consumption if we skip it and will load pytorch model only if we need to convert model aleksandr-mokrov commented on 2024-05-01T11:04:52Z Loading doesn't affect, models are replaced one by one with a larger one. But inference increase memory consumption. Added a widget with the choice to perform inference from the original model or not, skipping inference by default. |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2024-04-30T11:28:38Z not sure that it is suitable title as code below only select device without model compilation aleksandr-mokrov commented on 2024-05-01T11:05:08Z Changed |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2024-04-30T11:28:39Z Line #2. def __init__(self, prior_path): could you please add oportunity to provide device as argument of init? (it gives opportunity to experiment with devices on user, e.g. offload pipepilne parts on different devices) aleksandr-mokrov commented on 2024-05-01T11:05:15Z Added |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2024-04-30T11:28:40Z Line #10. gr.Slider(2, 20, step=1, label="Prior guidance scale"), could you please provide some explanation for demo parameters? what is prior guidance scale responsible for? how this value affect the result? aleksandr-mokrov commented on 2024-05-01T11:07:34Z Added extra information and the separate guidance scale for the decoder pipeline was added |
… into stable-cascade-image-generation
|
Updated View entire conversation on ReviewNB |
|
Loading doesn't affect, models are replaced one by one with a larger one. But inference increase memory consumption. Added a widget with the choice to perform inference from the original model or not, skipping inference by default. View entire conversation on ReviewNB |
|
Changed View entire conversation on ReviewNB |
|
Added View entire conversation on ReviewNB |
|
Added extra information and the separate guidance scale for the decoder pipeline was added View entire conversation on ReviewNB |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2024-05-01T12:55:09Z please remove gradio cell output aleksandr-mokrov commented on 2024-05-01T15:32:29Z removed |
|
removed View entire conversation on ReviewNB |
CVS-139009