Spaces:
Running
Running
Commit History
3 days ago
-
Add selectable model cards with saved per-panel preferences
Verified signatureccaf22c -
Restore top model summary cards across Easy, Hard and Plus
Verified signatureb8b8501
4 days ago
-
Consolidate result summaries and add selectable task comparison columns across Easy, Hard and Plus
Verified signatureaff3034 -
Show concise agent version labels
Verified signature4adcba2 -
Simplify Easy result notices and usage captions
Verified signaturefd21180
6 days ago
-
Add current RLE instruction audits for Easy, Hard and Plus
Verified signatureCompare the dated RLE working-tree snapshot against pinned native descriptions and runtime language pools. Preserve all published episode instructions, results, costs, videos, history and Dataset pointers. Validate all 500 audit rows and desktop/mobile navigation for each panel.555be04 -
Add OFT All 140k to Easy, Hard and Plus with verified completion counts
Verified signatured43af66 -
Add OFT All 140k to Easy, Hard and Plus with verified completion counts
Verified signature3114d7b -
Add OFT All 140k to Easy, Hard and Plus with verified completion counts
Verified signature5d6df64 -
Add OFT All 140k to Easy, Hard and Plus with verified completion counts
Verified signature9918a19
8 days ago
-
Publish RoboTwin 2.0 Hard: 200 native-instruction episodes; retain Easy results
Verified signatureb844444
10 days ago
-
Restore cost totals and per-task accounting after Dataset migration
Verified signatureResolve cost-record evidence paths using the same pinned Dataset root as episode data in both current and archived views. Restore totals, task costs, per-task means and CSV exports while preserving signature validation, experiment results and Dataset evidence.1450770 -
Show Kinex Astra/high + VLAct 50k results and complete model conditioning inputs
Verified signature53042a5
11 days ago
-
Publish complete VLAct Clean 50K result (37/50), retain all comparisons
Verified signature9b54b48 -
Use pinned Dataset evidence, keep all views, and publish VLAct All 47/50 results
Verified signature7b1e0ec -
Select the three Kinex VLA reruns and preserve previous valid attempts
Verified signature050b961 -
Remove instruction audit pages and retain verified native recording evidence
Verified signature91f9b77 -
Remove cost breakdown panels and recover standalone episode wall times
Verified signature2bfe8a9 -
Publish 50-task official RoboTwin VLA results, videos and complete inference evidence
Verified signature59e6c88
12 days ago
-
Align benchmark presentation and four-times three-view videos; audit VLA model and agent I/O
Verified signature000e043 -
Clarify success counts and completed episode denominators
Verified signaturedd87d49 -
Publish eight native episodes with policy failure and recovery traces
Verified signature518ecbf -
Add complete standalone baseline and detailed token accounting
Verified signature8e48ad6 -
Add first matched RoboTwin outcomes with full sessions and policy traces
Verified signature3b30eee -
Rebuild portal for matched RoboTwin native policy experiments
Verified signaturef150553 -
Refresh research portal with paired videos and complete execution records
Verified signatureReference-style layout, native-frame-preserving dual-view videos, instruction/session/tool/artifact inspection, and audited token usage with explicitly estimated API-equivalent costs. Existing experiment outcomes and source evidence are preserved.37db166 -
Publish native LIBERO agent and VLA integration results
Verified signatureThree valid B0/B1/B2 episodes, paired native recordings, observable sessions, policy I/O, and evidence checksums. Single-task integration validation only.87846ef