TLDR; I'm trying to run h2ogpt locally, specifically with the h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b-v3 model in a conda environment. For some reason my GPU can't be found.
Long version:
I've activated the env, installed requirements text and extra torch requirement but when I try to run this command, Powershell 7 says "no GPU detected":
python generate.py -base_model=h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b-v3 -score_model=None -prompt_type=human_bot -cli=True -load_8bit=True
It does continue to download/load the weights but the computer can't cope, makes screen output stutter etc. I would try load_4bit but definitely says "no GPU".
I have run nvidia-smi:
| NVIDIA-SMI 536.67 Driver Version: 536.67 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA GeForce GTX 1070 WDDM | 00000000:08:00.0 On | N/A |
| 0% 37C P0 31W / 151W | 1900MiB / 8192MiB | 0% Default |
| | | N/A |
I've just tried to installed CUDA toolkit to make sure I wasn't being silly - had it installed already, all drivers up-to-date too.
I assume Powershell is fine and can "see" my GPU because it can run 'nvidia-smi' and bring back info. Double checking the github docs I meet hardware requirements.
Looking through the github read me for h2ogpt I found a setting "CUDA_VISIBLE_DEVICES" which seems to be under 'What ENVs can I pass to control h2oGPT?'. I believe this is for if there are multiple GPUs but may be mistaken.
Thanks for reading.