Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
In addition, our model CodeGen (with up to 16B parameters trained on TPU-v4) outperforms OpenAI's Codex on the HumanEval benchmark.