MAttr open-sourced: one training run yields the full attribution curve across all sparsity levels
nateliason · x · 2026-09-24
Thread post for the MAttr paper: built with Noah Goodman, Christopher Potts, Dan Jurafsky and others, with code open-sourced on GitHub.
- One training run produces the entire CPR-vs-sparsity curve—the learned scores form a ranking, not a circuit at a fixed budget
- Universal API: learnscores takes any lossfn(mask) and node count; model adapters support nodes, edges, neurons, SAE latents, ViT patches
- A sibling repo covers parameter attribution with RL/SFT training code and configs, plus a HuggingFace MIB circuit-track submission
More from Research
- BFL shows FLUX 3 Action flying a simulated drone, hinting at broader uses — bfl_ai · 2026-09-24
- BFL open-sources 7B world action model FLUX 3 Action, tops RoboLab 6.1pp ahead — bfl_ai · 2026-09-24
- IKEA Assembly Benchmark: Top Model Score Jumped From 28% to 80% in 10 Months — emollick · 2026-09-24
- Juan Perdomo, researcher on performative prediction, joins NYU as assistant professor — thegautamkamath · 2026-09-24
- Claude Opus 5.5 and GPT-6 Sol land on Biomni Lab research platform — KexinHuang5 · 2026-09-24
- JevK5 open decision model ranks #2 of 76 on JevBench, outputs calibrated probabilities in ~13ms — airesearch12 · 2026-09-24