OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
Paper
•
2506.20512
•
Published
•
30
🐙 OctoThinker
Mid-training Incentivizes Reinforcement Learning Scaling
-
-
OctoThinker/MegaMath-Web-Pro-Max
Viewer • Updated • 69.2M • 12 -
OctoThinker/OctoThinker-8B-Long-Base
8B • Updated • 10 -
OctoThinker/OctoThinker-8B-Hybrid-Base
8B • Updated • 12 • 2 -
OctoThinker/OctoThinker-8B-Short-Base
8B • Updated • 46 -
OctoThinker/OctoThinker-3B-Short-Zero
4B • Updated • 8 -
OctoThinker/OctoThinker-3B-Hybrid-Zero
4B • Updated • 32 -
OctoThinker/OctoThinker-1B-Long-Zero
1B • Updated • 7 -
OctoThinker/OctoThinker-1B-Hybrid-Zero
1B • Updated • 7 -
OctoThinker/OctoThinker-1B-Short-Zero
1B • Updated • 41 -
OctoThinker/Llama3.2-3B-Zero
4B • Updated • 10 -
OctoThinker/OctoThinker-3B-Long-Zero
4B • Updated • 7 -
OctoThinker/OctoThinker-1B-Long-Base
1B • Updated • 19 -
OctoThinker/OctoThinker-1B-Short-Base
1B • Updated • 23 -
OctoThinker/OctoThinker-1B-Hybrid-Base
1B • Updated • 79 -
OctoThinker/OctoThinker-3B-Long-Base
3B • Updated • 22 -
OctoThinker/OctoThinker-3B-Hybrid-Base
3B • Updated • 108 -
OctoThinker/OctoThinker-3B-Short-Base
3B • Updated • 9