Tencent · Speculative decoding draft
DSpine for Qwen3.8-27B
DSpine is a five-layer, block-size-16 draft model for speculative decoding with Qwen3.8-27B, released by Tencent under Apache 2.0 and designed to run on SGLang. On GSM8K at greedy decoding it reaches an acceptance length of 7.20, against 5.69 for the official DFlash2 draft model. On agent workloads it reports end-to-end throughput of up to 3.84× over autoregressive decoding (WorkBuddy Office, low reasoning), and the page lists the SGLang launch flags needed to use it.
Open the source ↗Similar Developer Tools releases
- TPU 8tx and TPU 8i Max
Google · TPU Agent Infrastructure · Week 47, 2026
- NVIDIA Dynamo Recipes for Korean Open Models
NVIDIA · Dynamo Deployment Recipes · Week 47, 2026
- SageMaker Inference GPU partitioning (MIG)
Amazon Web Services (AWS Machine Learning Blog) · GPU Partitioning · Week 46, 2026
- Kinference
Buoyant · Inference Control Plane · Week 46, 2026
- Lutheos
Lutheos · Private inference API · Week 46, 2026
- ZYFA
Zyphra · FP8 Attention Kernels · Week 46, 2026
More from Tencent
- Hy4-preview
Tencent · MoE Coding Preview · Week 35, 2026
- HyCreator
Tencent · Text-to-Video · Week 34, 2026
- Hunyuan Hy3
Tencent · Text Generation · Week 28, 2026
- Hy-MT2-30B-A3B
Tencent · Multilingual · Week 21, 2026
- Hunyuan Hy3 Preview
Tencent · Reasoning · Week 17, 2026
- AngelSlim
Tencent · AI Infrastructure · Week 11, 2026
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.