LLaDA 2.2 can delete, insert, and substitute tokens mid-generation
omarsar0 · x · 2026-07-26
The post explains LLaDA 2.2’s Levenshtein editing loop in more detail.
- The model performs four primitive operations on its own output: KEEP, SUBSTITUTE, DELETE, and INSERT.
- Those operations are trained with longest-common-subsequence alignment.
- The purpose is to allow mid-generation recovery, so a broken span can be deleted or a missing one inserted before the error becomes permanent.
More from Models
- Benchmark scores drop from 89% to 19% on new evals — how benchmaxxing breaks leaderboard trust — airesearch12 · 2026-09-11
- Developer Building a Unified Leaderboard of All Model Benchmark Scores — airesearch12 · 2026-09-11
- Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals — realsohamparekh · 2026-09-11
- GPT-5.6 writes well but is instantly forgettable, user complains — BasedRaddka · 2026-09-11
- Opus Refuses Protein Research Codebase Over 'Safety' Concerns, Dev Considers Rolling His Own — josephdviviano · 2026-09-11
- User Hails Unconfirmed 'DeepSeek 4.1 Flash' as an Inflection Point in LLMs — himanshustwts · 2026-09-11