Radar ·
Qwen3.8-27B: all 64 layers converted to Gated DeltaNet
Radar records the Qwen3.8-27B release, with all 64 layers converted to Gated DeltaNet through per-layer distillation and LoRA r32, in English and Japanese, plus a recipe for turning full attention into linear attention.
Radar records the release of Qwen3.8-27B, with all 64 layers converted to Gated DeltaNet, a form of linear attention. According to the source, the 16 original full attention layers were recovered through per-layer distillation, with LoRA adaptation at rank 32. The model is described as available in English and Japanese. The source also points to a concrete recipe for converting a pretrained transformer with full attention into linear attention via distillation. The original publication is dated October 22, 2026.
The relevance lies in the architecture conversion technique, not in a Rota Nacional feature. The platform offers an API compatible with OpenAI and Anthropic and applies personal data detection before any model runs, but it does not train or convert models. To verify the material, consult the publication indicated in the Radar source and check the layer count, the distillation method and the languages before citing it. If your team uses AI to study the article, do not paste excerpts with customer data or credentials into the chat.
Original source: Lançamento do Qwen3.8-27B com todas as camadas convertidas para Gated DeltaNet