What is the strategy for initializing the task_embedding, layer_id_embeddings, and adapters_block_type embeddings?

[jianghaojun]

@rabeehk It seems all these embeddings are initialized from a pytorch default gassian normal distribution with N(0, 1).