SynthRL: Scaling Visual Reasoning with Verifiable Data Synthesis
Paper
•
2506.02096
•
Published
•
51
SynthRL
Models & Datasets of SynthRL
-
-
Jakumetsu/SynthRL-A-MMK12-8K-7B
Updated • 35 • 1
Jakumetsu/A-MMK12-8K
Viewer • Updated • 11.5k • 115Note The main dataset we used in SynthRL. 8K seed data plus 3.3K synthesised data.
-
Jakumetsu/A-MMK12-4K
Viewer • Updated • 5.71k • 20 -
Jakumetsu/A-MMK12-2K
Viewer • Updated • 2.86k • 19
Jakumetsu/SynthRL_evaluation_data
Preview • Updated • 22Note Our evaluation data, includes raw difficulty battle records, and difficulty elo ratings.
-
Jakumetsu/MMK12-8K
Viewer • Updated • 8.07k • 20 -
Jakumetsu/MMK12-4K
Viewer • Updated • 4.1k • 19 -
Jakumetsu/MMK12-2K
Viewer • Updated • 2.05k • 18
Jakumetsu/K12-Freeform-8K
Preview • Updated • 27Note Processed data, removed any YoN questions and convert all MCQ questions into freeform. The input of SynthRL data engine.