Reproducibility
Trace the result to its source.
This gallery is a read-only, in-progress snapshot containing only validated, committed task shards. It is not a final campaign result.
Campaign
Evaluation identity
- Campaign ID
- molmoact2-libero-full-20260815-20d5f33
- Created
- Aug 15, 2026, 11:34 PM UTC
- Repository commit
- 20d5f33ede…02c7b80c
- LeRobot commit
- 28b4f721f9…8023705f
- Clean checkout
- Yes
- Catalog
- hf-libero-0.1.3
- Source catalog SHA-256
- c59a79b7fa…2c4fdba3
Policy
Checkpoint
- Model
- standard
- Repository ID
- allenai/MolmoAct2-LIBERO
- Revision
- 0d24a92bd1…97c65aa2
- Seed
- 1000
- Episodes per task
- 50
- Cameras
- agentview, eye-in-hand
Public projection
Publication manifest
The exporter allowlists public files and binds them into a SHA-256 inventory before deployment.
- Format
- molmoact-public-gallery-preview
- Scope
- In-progress preview
- Captured
- Aug 16, 2026, 5:10 PM UTC
- Completed tasks included
- 111
- Completed rollouts included
- 5,550
- Files
- 17,029
- Total bytes
- 1.0 GB
- MP4 objects
- 11,100
- Manifest SHA-256
- 8369990b03…28902216
Bound source attestations
- Campaign plan SHA-256
- 06d1d8f727…c792a4f9
- Catalog SHA-256
- c59a79b7fa…2c4fdba3
- Reproduction lock SHA-256
- 65d4726177…63f7959b
- Committed task set SHA-256
- e6abbc13a8…65933336
Task pages expose embedded digest fields, but never link to raw manifests, evaluator logs, host paths, or attempt directories.
Validation boundary
Complete means content-bound.
A task is publishable only after its task-level COMPLETE marker and valid validation report bind the aggregate, episodes, video inventory, and provenance. The campaign-root COMPLETE attestation does not exist yet, so this snapshot cannot be recorded as a benchmark result.