[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-transformer-blocks-are-not-equally-nonlinear":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},1702,"transformer-blocks-are-not-equally-nonlinear","Transformer Blocks Are Not Equally Nonlinear","A new study finds that how linear a transformer's feed-forward block is depends on training, not architecture - with real implications for model compression.","Not every layer of a transformer is doing the hard nonlinear work you might assume.\n\nResearchers tested feed-forward network blocks across GPT-2, Pythia-160m, and Llama-160m, measuring how well a simple linear approximation could reconstruct each block's output. They called this metric \"linear recoverability\" and found it varies wildly - not just across models, but between adjacent layers within the same model. Some blocks scored above 0.99 (nearly linear); others fell below 0.3 (strongly nonlinear). Critically, GPT-2 and Pythia-160m share the same activation function and width, yet show sharply different recoverability profiles. That means the linearity of a block is a product of training, not a fixed architectural property.\n\nThis matters for anyone trying to compress or prune large language models. The study shows that highly recoverable blocks can be replaced with far simpler single-layer approximations - GPT-2's early feed-forward block was compressed to one-eighth the parameters with only a 0.77 perplexity increase. Blocks with low recoverability, by contrast, resist this treatment and flag where aggressive compression will hurt. That is a more targeted signal than the blunt per-layer heuristics most pruning pipelines use today.\n\nThe researchers also flag a quiet methodological trap: trained linear baselines often under-converge on transformer activations because the inputs are ill-conditioned, making past linearity estimates unreliable. The closed-form least-squares approach sidesteps that problem entirely - which suggests some prior work on transformer internals may need revisiting.","[\"ai\",\"machine-learning\",\"transformers\",\"model-compression\"]","2026-06-19T04:00:00.000Z","2026-06-19T10:15:43.905Z","2026-06-19T14:21:37.472Z","published",null,[],"ai",[24,26,27,28],"machine-learning","transformers","model-compression",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.19379",0,{"sections":35},[36,40,44,49,54,59,64,68,72,77,82,87,92,97],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",491,"2026-06-19T14:59:11.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",132,{"name":45,"slug":46,"count":47,"latest_published_at":48},"Policy","policy",88,"2026-06-16T09:26:09.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Consumer Tech","consumer-tech",78,"2026-06-16T17:58:24.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",62,"2026-06-18T15:24:16.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Deals","deals",58,"2026-06-19T14:43:50.000Z",{"name":65,"slug":66,"count":62,"latest_published_at":67},"Software","software","2026-06-16T20:00:00.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":18},"Dev Tools","dev-tools",50,{"name":73,"slug":74,"count":75,"latest_published_at":76},"Science","science",38,"2026-06-18T04:00:00.000Z",{"name":78,"slug":79,"count":80,"latest_published_at":81},"Gaming","gaming",31,"2026-06-16T15:25:13.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"General","general",26,"2026-06-13T18:35:15.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"Startups","startups",23,"2026-06-16T15:00:00.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",19,"2026-06-14T08:00:00.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]