mobility-model-zoo/scout-large

Token Classification • 0.6B • Updated about 19 hours ago • 1

Note Finds jobs-to-be-done, pains and gains in German and English mobility texts and quotes them verbatim, with actor type, evidence type and evidence scope. On the frozen benchmark pilot-v2 (guideline v2) it reaches a comparison composite of 0.72, against 0.67 for the best zero-shot small model, measured as agreement with two frontier reference models (Claude and the GPT mini tier). It reads a 9,240-character interview in 8.1 s on 4 CPU cores with 2.5 GB of memory, in 0.84 s on the GPU of a MacBook