A YEAR OF FAST APPLY: THE PATH TO 10K TOKENS PER SECOND

Summary from the TLDR Dev newsletter

Relace explains how it built a small model that applies code edits at a reported 10,000 tokens per second using high-quality merge data, LoRA fine-tuning, FP8 quantization, and speculative decoding. Its training pipeline filtered 200,000 candidate examples down to about 145,000, using a judge calibrated against 500 manually reviewed merges. Relace Apply 3 supports a 256,000-token context and multiple diff formats, although improving frontier-model edit accuracy could eventually reduce the need for specialized apply models.