Talk structure: a brief overview of the transformer architecture, then three acceleration cases (voice assistant on CPU, GPT-J on GPU, structured pruning for language and vision transformers), then questions.
Continue reading — it's free
Google or a one-time email code. No password.