AI & ML interests

A Family of Dynamic UltraFast Small Language Models Ready for Embodied Artificial General Intelligence!

Recent Activity

JingzeShi  updated a dataset 1 day ago
SmallDoge/SmallThoughts
JingzeShi  updated a Space 1 day ago
SmallDoge/README
View all activity

Doge

SmallDoge

Welcome to SmallDoge, where we pioneer the development of compact, high-performance small language models. Our focus is on creating ultra-fast SLMs using innovative dynamic algorithms. Committed to transparency and collaboration, all our training details and code are openly accessible on the SmallDoge GitHub repository.

Our Mission: To democratize access to advanced AI by developing efficient, open-source small language models that empower a wide range of applications and research.

Join our community on Discord!

Explore Our Projects

We offer a suite of resources and models:

  • Small-Doges: A versatile series of SLMs, including pre-trained base models, supervised fine-tuned models, and models enhanced with reinforcement learning.
  • Doge-CheckPoints: A collection of model checkpoints designed for seamless continued training on new datasets, ensuring smoother adaptation and minimizing training instability.
  • Small-Datasets: Curated, multi-stage, high-quality datasets specifically engineered to effectively train small language models, boosting their capabilities and helpfulness.
  • Doge-Downstream-Applications: A selection of SLMs optimized for various downstream tasks and real-world applications.