Error in ollama doesn't result in valid output

[geerlingguy]

As a quick demonstration:

~/Downloads/ollama-benchmark$ ./obench.sh -m llama2:13b -c 3 --markdown
Running benchmark 3 times using model: llama2:13b

| Run | Eval Rate (Tokens/Second) |
|-----|-----------------------------|

~/Downloads/ollama-benchmark$ ollama run llama2:13b
Error: model requires more system memory (14.2 GiB) than is available (13.7 GiB)
~/Downloads/ollama-benchmark$ echo $?
1

If there's an error in ollama, then that error should be propagated up to the output.

The exit code (1) is returned correctly, likely due to the script's set -e, but it's confusing from an end-user's perspective when you see the output.

Found as a result of testing on the 16 GB RAM Radxa Oryon (#13).