Compressed-Sensing-Guided, Inference-Aware Structured Reduction for Large Language Models

Develop a framework that bridges token-level pruning and model-level structured sparsity. Optimize LLMs for real-time inference by merging these two disparate fields.

Suggested repo: compress-infer

"Dynamic model compression that works in real-time."

Estimated effort: 80h