Orthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/2605.12825/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMTOrthrus: Memory-Efficient Parallel Token Generation via Dual-View Diffusion — mentioned by FranckDernoncouhttps://news.ycombinator.com/item?id=48154866urn:hackerpapers:mention:2605.12825:48154866Fri, 15 May 2026 22:38:58 GMT<p>Paper: https://arxiv.org/abs/2605.12825 ; Code+models: https://github.com/chiennv2000/orthrus ; Disclosure: co-author. Idea: Inject a trainable diffusion attention module into each layer of a frozen AR Transformer. Both heads share one KV cache. Diffusion head projects K=32 tokens in parallel; AR head verifies in a second pass and accepts the longest matching prefix. Output distribution is provably identical to the base model. Results: - Up to 7.8x TPF, ~6x wall-clock on MATH-500. - 16% of params trained, &lt;1B tokens, 24h on 8xH200. - vs. diffusion LMs (Dream, Fast-dLLM-v2, SDAR, Mercury…</p><p><a href="https://hackerpaper.kaipfister.de/p/2605.12825/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.LG