Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

Provable VLA captured artifact archive

This public dataset archives captured tensors and historical large research intermediates from the provable-vla workspace so local copies can be removed after upload verification. They are captured or generated data, not cryptographic proofs and do not establish that the deployed model executed or that the reference action was computed correctly.

The files may contain model inputs, intermediate activations, and output tensors for one captured SO100/SmolVLM2 inference. These artifacts have not been exhaustively scanned for embedded credentials; treat all contents as public. The source checkpoint is the public Hugging Face HuggingFaceTB/SmolVLM2-500M-Video-Instruct revision pinned by the workspace; no license beyond the original sources is asserted here.

MANIFEST.json records each source path, size, and SHA-256 from the local workspace. The archived bytes must be checked against those hashes before reuse.

Historical DeepProve candidate archive (2026-11-18)

Added 42 experimental ONNX and integer-graph candidate artifacts (78.10 GiB total) under captures/deepprove_full_onnx_fp32_candidate_cpu/. These are alternate exporter/calibration/probe iterations. They do not prove deployed-model arithmetic, caller policy, full VLA inference, or reference-output correctness. The per-file source path, size, and SHA-256 are listed in MANIFEST.json. The artifacts derive from a public SmolVLA checkpoint and the signed public-dataset capture; no license beyond original source licenses is asserted.

Downloads last month
75