Thank you for your work. How should this model be used? Do you have code? How much VRAM is needed to run and how long does it take to reason once

#3
by michaelj - opened

Thank you for your work. How should this model be used? Do you have code? How much VRAM is needed to run and how long does it take to reason once

Hi,

Do you have code?

Oh I thought I included a gradio app, looks like I forgot. Someone seems to be building a HF space for it: Gustav0-Freind/Bagel-7B-Demo

My code is quite hacky trying to get the weights on a single GPU so Gustav0-Friend's space code is probably going to be better.

how long does it take to reason once

As in to describe an image with ?

28 seconds with RTX3090.

How much VRAM is needed to run

Here's the model loaded on my rig:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.133.07             Driver Version: 570.133.07     CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3090        On  |   00000000:01:00.0 Off |                  N/A |
| 53%   44C    P8             39W /  390W |   10525MiB /  24576MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  NVIDIA GeForce RTX 3090        On  |   00000000:07:00.0 Off |                  N/A |
|  0%   49C    P8             27W /  350W |    4841MiB /  24576MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

16GB to load the model. And this is generating an image:

|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3090        On  |   00000000:01:00.0 Off |                  N/A |
| 57%   62C    P0            334W /  390W |   12015MiB /  24576MiB |     72%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  NVIDIA GeForce RTX 3090        On  |   00000000:07:00.0 Off |                  N/A |
| 43%   61C    P0            197W /  350W |    6901MiB /  24576MiB |     11%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

20GB

Sign up or log in to comment