Hugging Face's logo

OctoThinker
community
https://github.com/GAIR-NLP/OctoThinker
GAIR-NLP
Activity Feed

AI & ML interests

None defined yet.

Recent Activity

SinclairWangĀ  authored a paper 6 days ago
MegaScience: Pushing the Frontiers of Post-Training Datasets for Science Reasoning
SinclairWangĀ  updated a model 17 days ago
OctoThinker/OctoThinker-3B-Hybrid-Zero
SinclairWangĀ  updated a model 17 days ago
OctoThinker/OctoThinker-3B-Hybrid-Base
View all activity

Organization Card
Community About org cards

šŸ™ OctoThinker is led by GAIR

šŸŽÆ Our Goal: To reshape the pre-training trajectory so models scale better under RL.

Check our technical report for more details!

image/png

Collections 4

Mid-training Analysis Checkpoints (Llama-3.2-3B)
What makes a base language model suitable for RL? Through controlled experiments, we identify key factors then leverage them to scale up mid-training.
Mid-training Analysis Checkpoints (Llama-3.2-3B)
What makes a base language model suitable for RL? Through controlled experiments, we identify key factors then leverage them to scale up mid-training.
View 4 collections