Missing shared memory requirement in models #3
Description
Currently we're missing the required amount of shared memory in our spec and templates, leading to issues with atleast Qwen3-32B. It would be good if model-specific requirement was available in templates, so people wouldn't need to start containers to figure this out.