How to read relevance and efficiency numbers
- Relevance 1–5 rates how central an item is to the workshop's question, not paper quality. 5 = core, 4 = directly relevant, 3 = relevant or foundational technique, 2 = context (major models, datasets, benchmarks). Scores come from the curating agent's reading of the abstract; treat one-point differences as noise.
- Efficiency numbers are only comparable within the same measure.
closed-loop control(full sense→act loop),open-loop chunk rate(model output rate in isolation),throughput(often batched),relative latency(vs. each paper's own baseline),relative memoryandenergyare different quantities. Many entries omit hardware; don't compare those at all. - Numbers are as the authors state them; nothing was re-measured.
Topic overview
Runs a small open model inside your browser (WebGPU; recent Chrome, Edge or Safari). Nothing is sent to a server. The model is downloaded once and cached. It only sees the top results below, and it can still make mistakes, so check the cited items.