Both models also benchmark impressively well against commercial models. replit-finetuned-v1-3b is by far the smallest model on the table and it outperformed Codex and LLaMA. PaLM-Coder is 200x larger and we’re closing in on their performance with much better latency.
Replit Fine-Tuned Model Outperforms Codex with Better Efficiency
By
–
