SpecBundle (Phase 1), released by the SpecForge team alongside partners like Ant and EigenAI, introduces a set of production-ready EAGLE-3 model checkpoints designed to enhance speculative decoding's performance and accessibility. This release targets instruct-tuned models and aims to bridge the gap in production-ready tooling for speculative decoding, which has struggled to gain traction in the open-source community. The lack of high-quality draft models and limited training datasets has hindered the adoption of state-of-the-art methods like EAGLE3, which requires robust draft models to function effectively. SpecForge v0.2 brings significant system upgrades, focusing on usability and scalability. Key enhancements include streamlined data processing pipelines that boost efficiency—data regeneration is now ten times faster than in earlier versions. The integration of multiple execution backends, such as SGLang and Hugging Face Transformers, simplifies model support and allows users to select the best option for their needs. This flexibility should encourage broader use of speculative decoding techniques across various applications. The release aims to address three main issues: the scarcity of production-ready tools, the lack of quality draft models, and the limited scope of existing models that often fail to generalize effectively. By offering standardized training frameworks and high-performance draft models, the SpecForge team hopes to foster research and development in speculative decoding, making it easier for developers and enterprises to implement these techniques in real-world scenarios.