Commit History
2 days ago
-
Held-out results on the audited task set (62 tasks)
Verified signatureMiniWeb held-out task success after the task audit: MiniWebAgent 58.1, untrained single agent 38.7, untrained planner 45.2, task-centric 47.8, specialists 52.7, no-done 34.4. The adapters are unchanged.0064939
12 days ago
-
The paper's other executors (task-centric, family specialists, no-done) and a fuller card
Verified signature811681b -
MiniWebAgent executor: Qwen3.5-4B LoRA trained on MiniWebGen
Verified signatureb834cdb