{"name":"TokenGate","repo":"token-gate-llms/TokenGate","url":"https://midorreal.com/project/token-gate-llms-tokengate","repository":"https://github.com/token-gate-llms/TokenGate","description":"Process all your context faster with LLMs that can skim and selectively read what matters","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/token-gate-llms/TokenGate","checked":"20 Nov 2026"},"language":{"value":"Python","source":"https://github.com/token-gate-llms/TokenGate","checked":"20 Nov 2026"},"homepage":{"value":"https://arxiv.org/abs/2610.26754","source":"https://arxiv.org/abs/2610.26754","checked":"20 Nov 2026"},"last_release":{"value":"v0.1.0, 21 Oct 2026","source":"https://github.com/token-gate-llms/TokenGate/releases","checked":"20 Nov 2026"},"releases":{"value":"1","source":"https://github.com/token-gate-llms/TokenGate/releases","checked":"20 Nov 2026"},"contributors":{"value":"4","source":"https://github.com/token-gate-llms/TokenGate/graphs/contributors","checked":"20 Nov 2026"},"archived":{"value":"no","source":"https://github.com/token-gate-llms/TokenGate","checked":"20 Nov 2026"}},"stars":118.0,"writeup":{"what_it_is":"TokenGate is a drop-in neural network component that extends existing trained language models with learnable token selection. A gate module at an early layer scores context tokens and keeps the top-K, so that only the kept tokens are processed by the layers above it. The project reports up to 11.3x faster prefill and 3.4x lower memory with no accuracy loss.","audience":"Researchers and developers who want faster prefill on long-context language model inputs","claims":[{"kind":"specific","claim":"Reports up to 11.3x faster prefill with 3.4x lower memory at no accuracy degradation","excerpt":"up to **11.3x faster** prefill with **3.4x lower memory** at **no accuracy degradation**","status":"not_checked"},{"kind":"specific","claim":"Keeps only the top-K scored context tokens for processing above the gate layer","excerpt":"keeps the top-*K*. The layers above *L* process only","status":"not_checked"},{"kind":"broad","claim":"Trains with standard PEFT-style recipes using a config object and trainer","excerpt":"Training with TokenGate is as simple and efficient as standard PEFT training.","status":"too_broad"},{"kind":"specific","claim":"Runs on Python 3.10 or later with CUDA PyTorch 2.9 or later","excerpt":"Python ≥ 3.10, CUDA PyTorch ≥ 2.9","status":"not_checked"},{"kind":"specific","claim":"Released checkpoints for Qwen3 and Qwen3.5 models on Hugging Face","excerpt":"liana313/tokengate-niah-qwen3-1.7b","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-14"},"why_now":null,"owner_supplied":null,"score":{"verdict":"real","hype":80,"reality":65,"gap":16,"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":"2026-10-21T00:23:21+00:00","text":"Latest release v0.1.0","source":"https://github.com/token-gate-llms/TokenGate/releases"},{"date":"2026-10-19T17:03:09+00:00","text":"Repository created","source":"https://github.com/token-gate-llms/TokenGate"}],"cite":"TokenGate is a GitHub project at https://github.com/token-gate-llms/TokenGate written mainly in Python, described there as \"Process all your context faster with LLMs that can skim and selectively read what matters\".","attribution":"Data from Mid or Real, https://midorreal.com/project/token-gate-llms-tokengate, checked 22 November 2026. Attribution with a link is required.","still_real":null}