Find centralized, trusted content and collaborate around the technologies you use most.
Teams
Q&A for work
Connect and share knowledge within a single location that is structured and easy to search.
可能是疯狂的问题:但我想下载 gpt-2 模型框架,但我希望权重随机初始化。因此,就好像模型仍然需要在 reddit 内容(包括 json、词汇、元和索引文件等)上进行微调。这可能吗?
亲切的问候!
是的,可以随机初始化并从头开始训练。例如对于 T5,您可以执行以下操作:
from transformers import GPT2Model, GPT2Config config = GPT2Config.from_pretrained() # config as in the paper model = GPT2Model(config)