JevForge: decision-data synthesis, training and web-action scoring

An end-to-end research toolkit builds decision records, trains and calibrates candidate scorers, and serves a Jev-compatible API. Its released Qwen3.5-0.8B model ranks supplied web actions, while the public demo presents recorded Mind2Web decisions.

Open demo ↗

Notes

The model card reports Choice top-1 of 0.5787 on 2,400 test records and 0.6373 on 1,158 OOD records using website-disjoint splits and a saved temperature of 0.9717.

source-backed — Public repository, docs or live artifact. About this label

Destinations

Also linked

• JevForge-0.8B model ↗
• JevForge-Mind2Web dataset ↗

Filed under

• #local
• #training
• #browser
• #evaluation

In collections

• Benchmarks & independent tests →
• Browser & computer use →
• Local & replica models →