{"name":"NextTransformer","repo":"Interactive-Active-AI/NextTransformer","url":"https://midorreal.com/project/interactive-active-ai-nexttransformer","repository":"https://github.com/Interactive-Active-AI/NextTransformer","description":":sunrise: A comprehensive, cutting-edge open-source framework for next-generation Transformers/LLM, with compute-matched benchmarking and a platform for autonomous architecture research.","facts":{"license":{"value":"MIT","source":"https://github.com/Interactive-Active-AI/NextTransformer","checked":"22 Nov 2026"},"language":{"value":"Python","source":"https://github.com/Interactive-Active-AI/NextTransformer","checked":"22 Nov 2026"},"homepage":{"value":"https://interactive-active-ai.github.io/NextTransformer/","source":"https://interactive-active-ai.github.io/NextTransformer/","checked":"22 Nov 2026"},"releases":{"value":"0","source":"https://github.com/Interactive-Active-AI/NextTransformer/releases","checked":"20 Nov 2026"},"contributors":{"value":"13","source":"https://github.com/Interactive-Active-AI/NextTransformer/graphs/contributors","checked":"20 Nov 2026"},"archived":{"value":"no","source":"https://github.com/Interactive-Active-AI/NextTransformer","checked":"22 Nov 2026"}},"stars":490.0,"writeup":{"what_it_is":"NextTransformer is an open-source framework for training and benchmarking Transformer and language model architectures. It trains recurrent, sparse, linear, residual and hybrid variants on the same data and token budget, with matched training compute, and records results as audited files.","audience":"Researchers and engineers studying language model architectures under compute-matched training.","claims":[{"kind":"specific","claim":"Trains multiple architecture families on the same data, tokens and training FLOPs per token","excerpt":"trains recurrent depth, sparse and linear attention, residual-stream variants and hybrids on the same data, for the same 49.152B tokens","status":"not_checked"},{"kind":"specific","claim":"Training compute is matched to within 1% at admission","excerpt":"training compute matched to within 1% and audited over the whole run","status":"not_checked"},{"kind":"specific","claim":"Uses a 2,048-token context and a 32,768-token tokenizer","excerpt":"one 32,768-token tokenizer · 2,048-token context","status":"not_checked"},{"kind":"specific","claim":"Provides a model zoo of many architecture families","excerpt":"Many architectures, one framework: attention, sparse, linear, looped, residual and hybrid models","status":"not_checked"},{"kind":"broad","claim":"Described as a comprehensive, cutting-edge framework","excerpt":"The comprehensive, cutting-edge open-source framework, benchmark","status":"too_broad"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-20"},"why_now":null,"owner_supplied":null,"score":{"verdict":"real","hype":63,"reality":76,"gap":-14,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-10-31T07:46:13+00:00","text":"Repository created","source":"https://github.com/Interactive-Active-AI/NextTransformer"}],"cite":"NextTransformer is a GitHub project at https://github.com/Interactive-Active-AI/NextTransformer written mainly in Python, described there as \":sunrise: A comprehensive, cutting-edge open-source framework for next-generation Transformers/LLM, with compute-matched benchmarking and a platform for autonomous architecture research\".","attribution":"Data from Mid or Real, https://midorreal.com/project/interactive-active-ai-nexttransformer, checked 22 November 2026. Attribution with a link is required.","still_real":null}