Closed bschilder closed 3 months ago
Mon Jun 3 08:41:31 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.161.08 Driver Version: 535.161.08 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| 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 A30 Off | 00000000:00:07.0 Off | 0 |
| N/A 34C P0 46W / 165W | 3MiB / 24576MiB | 0% Default |
| | | Disabled |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| No running processes found |
+---------------------------------------------------------------------------------------+
ahhh, actually the issue was I had forgotten to add --nlayers 33
This works now!
!python eval_single_anndata.py --adata_path data/ot_l2g.h5ad --dir output/ot/ --species human --nlayers 33 --model_loc model_files/33l_8ep_1024t_1280.torch --batch_size 25 --filter False --skip False
That said, if you'd find it helpful I'd be happy to make a PR with the conda env. It took me a couple weeks to get a version of it that worked and it could save others quite a bit of time.
Hi Brian which packages had issues? I think the up-to date versions of the needed libraries all work.
cuda drivers, as always 😆
the trick was using cuda-toolkit
(which has versions >12.0), NOT cudatoolkit
(which only has versions <12.0).
Exactly what is "up-to-date" changes over time, and which versions are the latest will depend on the distributor. For those reasons I always find it helpful to have some minimum requirements set in the yaml, along with the conda channels (or pip) to install them from.
Also, it seems you used pytorch >2.0, inferred from the fact that my other envs with earlier versions of pytorch didn't work.
Made PR here #42
Hi @Yanay1 ,
I seem to be running into some issues with the 33-layer model that I hadn't observed with the 4-layer model.
It may be due to some versioning conflict. If so, would you mind sharing the exact versions of the packages you used? (especially the ones listed here)
Command
Output
Conda
yaml
I'm using a conda env I constructed on a Linux machine with an NVIDIA GPU as follows:
versions