We recently found out that trans_required_pattern.search(name) is more robust than trans_required_pattern.match(name), which is used for exact match in granite and llama model implementations.

This issue proposes to change it in all the models e.g., Granite here, and Llama here ,