Model updates

Official news, first rumor reports and substantive progress. Each entry keeps the information published at that time.

• Tencent open-weights DSpine, a parallel speculative-decoding drafter for Qwen3-8B

  Tencent released Qwen3-8B-DSpine-b16 on Hugging Face with inference code on GitHub: a parallel speculative-decoding drafter that injects each position's predecessor feature through network depth so all positions draft in parallel while keeping a causal chain. On the authors' Qwen3-8B acceptance-length table, DSpine averages 7.83 at T=0 versus 7.55 for DFlash2 and 7.44 for DSpark. Paper is arXiv 2609.36173; Apache-2.0 weights and SGLang serve flags are documented on the model card.

  OfficialOfficial updateOct 16Source: Tencent on Hugging FacePart of DSpine

RSS for this view →