Kernel died
apking2000 opened this issue · 4 comments
apking2000 commented
import torch
from one_peace.models import from_pretrained
device = "cuda" if torch.cuda.is_available() else "cpu"
model = from_pretrained("ONE-PEACE", device=device, dtype="float32")
when i try to load model my kernel died can you tell me the specification of device like memory requirements, gpu memory allocation etc?
logicwong commented
We only tested on a 40GB A100. You can try setting dtype='float16'.
apking2000 commented
Hi can you tell me how you use the multi modality to retrieve the data ex:-
text+image to image?? It will be very helpful for me??
Thanks
Aakash
…On Mon, 8 Jan 2024, 12:53 Wang Peng, ***@***.***> wrote:
We only tested on a 40GB A100. You can try setting dtype='float16'.
—
Reply to this email directly, view it on GitHub
<#46 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AUXQ2ET3GN2CUTGH64MCNBDYNONI5AVCNFSM6AAAAABBJVZLI2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOBQGQ4TEOJXGI>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
apking2000 commented
Like i want to know how we combine multiple embedding to retrieve the image
data
…On Mon, 15 Jan 2024, 17:28 aakash panda, ***@***.***> wrote:
Hi can you tell me how you use the multi modality to retrieve the data
ex:- text+image to image?? It will be very helpful for me??
Thanks
Aakash
On Mon, 8 Jan 2024, 12:53 Wang Peng, ***@***.***> wrote:
> We only tested on a 40GB A100. You can try setting dtype='float16'.
>
> —
> Reply to this email directly, view it on GitHub
> <#46 (comment)>,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/AUXQ2ET3GN2CUTGH64MCNBDYNONI5AVCNFSM6AAAAABBJVZLI2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOBQGQ4TEOJXGI>
> .
> You are receiving this because you authored the thread.Message ID:
> ***@***.***>
>
logicwong commented
@apking2000 For guidance on implementing text+image to image retrieval, you can refer to multi-modal-embedding. To accomplish this, simply calculate the mean of text_features and image_features, and then employ the combined features to retrieve images.