GPU requirements

#31
by fedorn - opened

For anyone wondering regarding the GPU required to run the model, I was able to run the model on AWS EC2 instance g6e.xlarge with 32 GiB RAM and one NVIDIA L40S GPU with 48 GiB VRAM. The sentence-transformers code didn't require any modifications, and with HuggingFace Transformers the model can be loaded with model = AutoModel.from_pretrained('nvidia/NV-Embed-v2', trust_remote_code=True, device_map="cuda"). After loading the model takes around 30 GiB VRAM:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.127.08             Driver Version: 550.127.08     CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| 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 L40S                    Off |   00000000:30:00.0 Off |                    0 |
| N/A   29C    P0             79W /  350W |   30393MiB /  46068MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A      2859      C   .../bin/python                              30386MiB |
+-----------------------------------------------------------------------------------------+

after embedding the example it takes almost all available VRAM:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.127.08             Driver Version: 550.127.08     CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| 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 L40S                    Off |   00000000:30:00.0 Off |                    0 |
| N/A   32C    P0             78W /  350W |   45499MiB /  46068MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A      2859      C   .../bin/python                              45492MiB |
+-----------------------------------------------------------------------------------------+

Sign up or log in to comment