Hi! Does anybody tried training contentvec with larger hubert models (xlarge) or with Whisper models?