500 original multimodal single-decision inputs from 193 trajectories. The complete available source conversation prefix is retained, including prior assistant turns and omitted-image markers. The selected current assistant answer and all future turns are excluded. The current screenshot retains its original PNG bytes.

The 500 nominal references are unmodified cases sampled from original normal-task rollouts, without trigger, original success, or model-output filtering. Missing auxiliary trigger annotations do not remove otherwise valid observations. Training/source-validation and runtime episodes/images are excluded. Fully black observations and malformed histories are excluded; original source errors in otherwise valid task histories remain.

Every input has only input_id, task_id, and messages. Source filenames, private strata/pair bindings, training labels and checkpoint-specific output judgments are not included. IDs are opaque; record order is shuffled independently of outputs. Resolve relative image paths against the snapshot directory and apply the model-native chat/image processor. Planned Qwen3.5 use is enable_thinking=False, with image processor size shortest_edge=65536, longest_edge=1048576.

These are offline next-action inputs; no generated action is executed. One action does not establish simulator task success or persistent internal behavior. This input release does not claim a trained checkpoint or runtime GT.

Source: BEAT dataset, revision bb75922979748334916b67e79b3232d25f12595d. Original licensing and attribution apply. Input and image checksums are in input_manifest.json; detailed source mapping remains in private audit records.

Downloads last month: 15

Total file size:

418 MB