RoboJev-evals / benchmark.jsonl
josefchen's picture
Normalize success labels consistently across the three evaluation sets
64c786b verified
Raw History Contribute Delete
1.21 MB
File too large to display, you can check the raw version instead.