Tiny but Mighty: MSR's Small Models Crush HumanEval
·43 words·1 min
Amazing work by our group at Microsoft Research is finally public!
Can you achieve 50% on HumanEval with a mere 1.3B code generation model? Yes you can! 😇
How about cracking 45% with a “tiny” 350M model? No problem! 🤯
https://arxiv.org/abs/2306.11644 https://x.com/SebastienBubeck/status/1671326369626853376