LOREL
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
LOReL Sawyer (unseen multi-step composition)
25.21Success Rate
4
LOReL Sawyer human provided rephrasals
40.16Success Rate
3
LOREL Images (partial obs.) (test)
40Success Rate
3
LOREL States (fully obs.) (test)
66.7Success Rate
3