Sean Owen
Sean Owen
How are you running it?
I think this is an issue with that demo more than Dolly but it's a reasonable place to discuss it here. If you're doing this all in one notebook, then...
You can discuss at github.com/mlflow/mlflow. I think they are going to add support for other chain classes.
No, you're just out of OS memory: CPU Virtual Memory: used = 29.71 GB, percent = 95.3% Offload is also offloading optimizer state into main mem. Can you use a...
You probably have an incompatible old pytorch version. What are you using? are you installing the requirements.txt as shown in this repo?
What torch version do you have? maybe that needs to be pinned
What version of torch are you using?
I updated the list of requirements to require later deepspeed, and fix a version of torch. Try again?
You probably don't quite have the right cublas version for your CUDA? The training script shows the right installs for CUDA 11.3 and 11.7. I'm not sure where you're loading...
Hm, I can't reproduce that. Start over with exactly the example on the repo page?