@charles_irl: it’s hot spec summer
Summary
DeepSeek has open-sourced DeepSpec, a full-stack codebase for training and evaluating speculative decoding models.
View Cached Full Text
Cached at: 06/28/26, 08:04 AM
it’s hot spec summer
Lisan al Gaib (@scaling01): DeepSeek just open-sources another piece of their training stack.
DeepSpec: a full-stack codebase for training and evaluating speculative decoding models
Similar Articles
DeepSeek open-sources inference optimizations with 60–85% faster generation [pdf]
DeepSeek open-sourced DeepSpec, a full-stack codebase for training and evaluating draft models for speculative decoding, enabling 60-85% faster generation. It includes data preparation, training, and evaluation scripts with support for multiple draft model algorithms (DSpark, DFlash, Eagle3).
DeepSpec - a deepseek-ai Collection
DeepSeek AI released the DeepSpec collection on Hugging Face, featuring speculative decoding models (dspark, dflash, eagle3) based on Qwen3 and Gemma4 in various sizes (1B-3B).
@danielhanchen: DeepSeek just released DSpark for V4 Flash & Pro, a new speculative decoding method boosting throughput by 51% to 400%!…
DeepSeek released DSpark, a speculative decoding method that boosts throughput by 51% to 400% for V4 Flash & Pro, along with the open-source DeepSpec codebase for training and evaluating draft models.
@DeRonin_: DeepSeek just dropped a 5-page paper + free GitHub repo that makes any LLM respond 80% faster it's called speculative d…
DeepSeek released a paper and MIT-licensed open-source implementation of speculative decoding (DSpark) that speeds up LLM responses by up to 80% by using a small 'guess' model and a large 'check' model, achieving both speed and accuracy without tradeoffs.
@charles_irl: Speculation Is All You Need. In this blog post, we announce the co-release (w/ Z Lab) of six more state-of-the-art DFla…
Modal and Z Lab release six new DFlash speculative decoding draft models for Qwen 3.x, achieving over 1000 tokens per second on a B200 and arguing that speculative decoding is the most impactful inference optimization.