Tag
#text-generation
Reports tagged text-generation.
Reports
2 reportsNov 2026
11-01 speculative-decoding How speculative decoding checks ahead—and when it saves time A cheaper model supplies candidate text for the larger model to check together; speedup depends on how much of the draft survives and how much the shared check costs. Oct 2026
10-31 diffusion-language-models How parallel text generation coordinates a passage Diffusion language models build text through repeated recovery steps. What matters for agreement is which choices become shared context, when they become fixed, and whether later information can change them.