ClinchWorks
Explore a subject, work through the curriculum, and choose your next step.
Reduce memory and latency while preserving useful model quality. Actual quantization, INT4 packing, calibration and capacity calculations plus optional model comparison. Includes detailed lessons, worked solutions, failure injection, independent challenges and instructor assessment.
Build reliable RAG systems from ingestion through grounded generation and evaluation. Expanded engineering lessons with worked examples, runnable local practice, failure investigation and evidence-based assessment.
Implement and debug attention, multi-head layout, rotary positions, and transformer blocks using transparent NumPy experiments plus a complete optional PyTorch implementation. Includes causality tests, worked calculations, and an architecture capstone.
