GREP · Software mentions — f14 (CPU)

Reads scientific PDFs and pulls out every software name, tool and library the authors used, with its location in the text, a type and a purpose. This leaf runs f14, the strongest of the three models on its own. GREP runs three independently trained models over the same papers; a mention enters the final record only when at least two of the three find the same span. A work unit is about ten papers, roughly 5 MB in and one small JSON out. Every unit goes to three different volunteers and the head accepts it once two produce matching output. Runs on any machine with 6 GB of free memory and 15 GB of free disk; a single core is enough, though more cores finish a unit faster. The model ships inside a one-time 5.6 GB image download, after which the work runs entirely offline.

Measured accuracy (character-level F1; higher is better):

Model                dev    test   Softcite   SoFAIR clean
----------------------------------------------------------
f14                0.823   0.819     0.8653         0.8909   <- this leaf
f13                0.814   0.807     0.8637         0.8854
V1                 0.800   0.781     0.8096         0.7950
crowd (2-of-3)     0.833   0.828     0.8648         0.8940