Leonid Sinev, Vladislav Vetrov.
The authors are working on improving generative language models to ensure they perform well in conversational assistants. The presentation plan: classical approaches to accelerating and compressing models; practical experience in accelerating inference of Large-sized GPT-type models; acceleration through architectural changes and deploying the model across multiple GPUs.
Continue reading — it's free
Google or a one-time email code. No password.