Reproduce “Context Tuning for In-Context Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jack Lu, Ryan Teehan, Zhenbang Yang, Mengye Ren
Context Tuning adapts a model to few-shot tasks by converting demonstrations into a trainable memory, then refining that memory with gradients while freezing model weights. Its variant CT-KV tunes layer-specific keys and values. On NLP-LR, it reports 44.2% accuracy and 145 seconds of training per task, versus TTT’s 44.1% and 342 seconds; results are comparable, not universally more accurate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.