DASLab
Overview
DASLab refers to the Distributed Algorithms and Systems lab at ist-austria. The lab is recognized for its research into efficient machine learning algorithms, particularly in the domain of model quantization and optimization for local deployment.
Key Research & Innovations
- GSQ+RCO Quantization Framework: Developed by DASLab, this technique combines Gumbel Softmax Quantization (GSQ) and Riemannian Constrained Optimization (RCO).
- Objective: Enable high-accuracy local deployment of large language models without significant performance degradation.
- Case Study: Successfully applied to Qwen3.8-27B Quantization: GSQ+RCO for Local, Accurate LLM Deployment.
- Results: Reduced model size to ~11.8GB while maintaining zero accuracy loss compared to the full precision model.
- Parameters: Optimized for models with approximately 27 billion parameters.