Code Review

This pull request introduces whitelist-based plugin filtering in base_model.py and ensures the trust_remote_code flag is correctly passed from the Trainer to the model loading process. However, two significant issues were identified: the filter_plugin_in_whitelist function is imported but its implementation is missing from the PR, which will cause an ImportError. Additionally, the security check logic appears to be inverted, potentially blocking safe, whitelisted plugins when trust_remote_code is disabled.