Organized by: Alignment Research Center (ARC) , AIcrowd WhestBench 2026: ARC White Box Estimation Challenge WhestBench is a benchmark for white box activation estimation : given the weights of a small ReLU multi layer perceptron (MLP) and a strict floating point operation (FLOP) budget, predict the average post activation value of every neuron when the network is fed standard Gaussian inputs. This is the train dataset for WhestBench 2026 — pre baked MLPs paired with their ground truth activation statistics. Quick start The pure HuggingFace path (no whestbench install required): The whestbench convenience wrapper (adds schema validation + metadata.json access): Run an estimator end to end via the CLI: ➡️ New to the challenge? Head over to the WhestBench starter kit for a worked example estimator, the recommended project layout, FLOP tracking patterns with flopscope , local testing tips, and the submission workflow. What's in this dataset Each row is one MLP, paired with the Monte Carlo–computed ground truth statistics of its post activation outputs. Four things travel together per row: 1. The MLP weights (the network you'll analyse). 2. The per layer ground truth means — what your e…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy