inclusionAI releases Ling3-DSpark, a DSpark speculative decoding model for Ling-3.0-flash
Original title:inclusionAI/Ling-3.0-flash-dspark
AI summary
inclusionAI has released Ling3-DSpark, a DSpark speculative decoding model designed for Ling-3.0-flash with 1.36B parameters that dynamically selects the number of draft tokens via a confidence head. Across nine benchmarks including GSM8K, the average accepted length is 5.29, reaching 6.40 on GSM8K. The model was trained with SpecForge and can be deployed via SGLang.
Key Takeaways
- inclusionAI has released Ling3-DSpark,a DSpark speculative decoding model designed for Ling-3.
- 0-flash with 1.
- 36B parameters that dynamically selects the number of draft tokens via a confidence head.
- Across nine benchmarks including GSM8K,the average accepted length is 5.
- 29,reaching 6.
Quick Facts
- Published
- Updated
- Category
- Model Releases
- AI score
- 60
该模型把投机解码的接受长度按工作负载分开报告,数学与代码任务明显高于对话类,部署时可据此对不同场景的加速收益有更实际预期。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: inclusionAI releases Ling3-DSpark, a DSpark speculative decoding model for Ling-3.0-flash. August 9, 2026. https://new.nat6.net/en/news/cmt333krj0fv5ro6tjw6bmtb2/