Brenton Carter

Edit your launch.py file, find commandline_args = os.environ.get('COMMANDLINE_ARGS', "") and change it to commandline_args = os.environ.get('COMMANDLINE_ARGS', "--skip-torch-cuda-test --no-half --use-cpu all")

The link is valid for 7 days with unlimited executions. I would just delete and re-do.

That sounds like just an error with how the link is input in the download.sh The weights link should be fine, just check your brackets and values. If it still...

You will have to manually add nccl. Make sure you have full privileges before choosing your install from nvidia. HPC-SDK is easiest, but downloading the tar and extracting to usr\local...

You can't resolve nccl issues without nvidia. There are other process's that could be used as opposed to nccl, there are also other libraries which allow parallel work around but...

CLA check complete.

You need to ensure you have permission to write and that the folder exists (use * ls -ld ~/llama//65B/ * to check permission settings). If it doesn't exist, open with...

You need to understand linux environments, python, cuda and a few other things in order to understand how to compile Llama. The link that was sent to you is to...

Love this!

NCCL is a pain. I'm assuming you are running this on windows in conda or similar environment? The easiest way is to just deal with hpc-sdk as it includes nccl....