ai-agents-config-validation-loading-a-qwen2-vl-config-with-rope-scaling-type-s-6ac92163

THEN Add an `extra_ignore_key` parameter to the `rope_config_validation` function in Transformers to allow models like Qwen2-VL to pass additional valid keys (e.g., `mrope_section`) without altering the original `type` field. This ensures backward compatibility and correct config loading.

def rope_config_validation(config, extra_ignore_keys=None):
    if extra_ignore_keys is None:
        extra_ignore_keys = []
    # existing logic with extra allowed keys
    ...