A Sampler’s Blog

The blogger is interested in inference-time scaling for LLMs. He also shares research ideas he finds insightful and occasional updates on his own work.
His recent works include mdm reasoning (GDPO), test-time alignment (TRI-TSMC), discrete diffusion (SCDD, DiDi Instruct), continuous diffusion (VSDM, SB-TS), and sampling (DeepPT, Contour).
🔥 Welcome! Parallel Power Tempering — A 9B model can be comparable to GPT5/ Opus 4.5/ GLM4.7.
Contact me at: firstnamelastname056@gmail.com. Also looking for tennis enthusiast in NJ.
La pensée n’est qu’un écliar au milieu d’une longue nuit. Mais c’est cet éclair qui est tout. - Poincaré