FreeLLMTencent launches Qwen3.8-27B-DSpine-b16 draft model on Hugging Face
The new five‑layer draft model adds parallel token generation for Qwen3.8‑27B with improved benchmarks and open‑source tooling.
Tencent released Qwen3.8-27B-DSpine-b16 on Hugging Face, a five‑layer draft model for the Qwen3.8‑27B base.
The model uses a block size of 16, an 8,192‑token sliding window and supports up to 4,096 new tokens per request.
Benchmarks show DSpine outperforms DFlash2 and autoregressive decoding, e.g., GSM8K scores rise from 5.69 to 7.20 and throughput gains reach 2.71× over AR.
Installation requires cloning the DSpine repository and running a pip install, then serving with SGLang using the provided command line options.
No inference provider currently hosts the model, and it is available under an Apache 2.0 license with 1,464 downloads last month.
O que isso significa para você
The open release lets developers experiment with speculative decoding without waiting for commercial services. For users seeking free or locally run models, the draft model offers a way to speed up generation on compatible hardware while remaining fully open source.
Fontes
- Tencent releases Qwen3.8-27B-DSpine-b16 on Hugging Face, Hugging Face trending (official), 7 Nov 2026
Escrito automaticamente a partir das fontes acima, depois verificado pelo nosso código: todo número e nome deve aparecer nelas, e nenhuma frase pode copiar uma fonte. Encontrou um erro? Reporte‑o.
Qual IA você pode usar grátis hoje?
Todo modelo de IA gratuito, classificado por votos humanos cegos, com a memória e velocidade que ele precisa na sua própria GPU.