I’ve been experimenting with next-edit prediction for a while and wrote up how we trained the edit model that powers our Tab completion feature. This post is part of a broader series where we share how we built this feature from the low-level modeling right up to the editor extension.
The cool part is we fine-tuned Gemini Flash Lite with LoRA instead of an OSS model, helping us avoid all the infra overhead and giving us faster responses with lower compute cost.
Comments
I’ve been experimenting with next-edit prediction for a while and wrote up how we trained the edit model that powers our Tab completion feature. This post is part of a broader series where we share how we built this feature from the low-level modeling right up to the editor extension.
The cool part is we fine-tuned Gemini Flash Lite with LoRA instead of an OSS model, helping us avoid all the infra overhead and giving us faster responses with lower compute cost.