Zhanli-Li/DataWorld-35B-A3B
Reinforcement Learning • 35B • Updated • 69 • 2
DataWorld
Agentic RL fine-tune of Qwen3.6-35B-A3B for data analysis & cowork , plus its 100-task held-out evaluation benchmark.
Note The model — an agentic RL fine-tune of Qwen3.6-35B-A3B (35B total / 3B activated). Apache-2.0.
Zhanli-Li/DataWorld-Bench
Updated • 2.04k • 1Note The benchmark — 100 held-out evaluation tasks over 50 worlds, in harbor task format. Apache-2.0.
Zhanli-Li/DataWorld-Train-2k-part1
RL Environment • Updated • 426 • 2Note Training data (part 1 of 2) in harbor task format — 597 of the 2,135-task training slice, worlds p000000–p000499.
Zhanli-Li/DataWorld-Train-2k-part2
RL Environment • Updated • 425 • 2Note Training data (part 2 of 2) in harbor task format — the other 1,538 tasks, worlds p000500–p009920.
Data World: Scaling Data Environments from Scratch
Paper • 2610.21656 • Published • 36Note Official Paper of Data World