H-CDLMs: continuous diffusion LMs jointly denoise tokens and semantic clusters for better, cheaper generation
abursuc · x · 2026-10-24
Researchers introduce Hierarchical Continuous Diffusion LMs (H-CDLMs), extending the ReDi idea of jointly modeling images and semantic features to continuous diffusion LMs: the model denoises tokens and coarse semantic clusters simultaneously. The authors report better generation quality for less compute, with a technical thread detailing the method. ReDi authors Spyros Gidaris and ThKouz endorsed the work.
More from Research
- EMNLP oral paper analyzes 16,940 interview responses to define what makes a good qualitative answer — windx0303 · 2026-10-24
- kalomaze shares hilariously incoherent early outputs of tiny byte-level code models — kalomaze · 2026-10-24
- A 17-state Beeping Busy Beaver machine that quasihalts iff the Collatz conjecture is false — rickasaurus · 2026-10-24
- NeurIPS paper OAT: attribute agent failures by learning only from success flows — SharonYixuanLi · 2026-10-24
- Redditor proposes crowdsourced multilingual agent benchmarks to escape English-only evals — Prestigiouspite · 2026-10-24
- Muon optimizer may nearly halve LLM training compute, challenging a decade of AdamW — bycloud · 2026-10-24