we are not using shards over here, bcos due to less data , the llm halucinates , so we are using adapters over here instead of shards