Matched OpenR1 SFT checkpoints and subsequent GRPO models for studying token-level behavior during the transition from SFT to RL.
Zbib M PRO
zbeeb
P(doom) Rejects the premise
AI & ML interests
KAUST - AUB
Recent Activity
authored a paper about 17 hours ago
Arabic Safety Alignment as Selective Refusal: An Empirical Study of SFT, DPO, and Guard Calibration updated a collection 1 day ago
Easy-Tune: cleaned Math Data updated a dataset 1 day ago
zbeeb/OpenThoughts3-Qwen3-4B-Math-21kOrganizations
models 38
zbeeb/Qwen2.5-3B-Instruct-OpenR1-SFT-GRPO
Text Generation • 3B • Updated • 14
zbeeb/Qwen2.5-3B-Base-GRPO
Text Generation • 3B • Updated • 15
zbeeb/Qwen2.5-1.5B-Base-GRPO
Text Generation • 2B • Updated • 417
zbeeb/Qwen2.5-Math-1.5B-Base-GRPO
Text Generation • 2B • Updated • 424 • 1
zbeeb/Qwen2.5-3B-OpenR1-SFT-GRPO
Text Generation • 3B • Updated • 415
zbeeb/Qwen2.5-Math-1.5B-OpenR1-SFT-GRPO
Text Generation • 2B • Updated • 176
zbeeb/Qwen2.5-1.5B-OpenR1-SFT-GRPO
Text Generation • 2B • Updated • 177
zbeeb/Qwen2.5-3B-Instruct-OpenR1-SFT
Text Generation • 3B • Updated • 170
zbeeb/Qwen2.5-3B-OpenR1-SFT
Text Generation • 3B • Updated • 152
zbeeb/Qwen2.5-Math-1.5B-OpenR1-SFT
Text Generation • 2B • Updated • 157
datasets 15
zbeeb/OpenThoughts3-Qwen3-4B-Math-21k
Viewer • Updated • 21.2k • 37
zbeeb/openr1-token-level-training-logs-20261007
Preview • Updated • 1.6k
zbeeb/OpenR1-SFT-Math-20k
Viewer • Updated • 60.8k • 442
zbeeb/Skywork-DeepScaleR-Merged-Verifiable-Dedup
Viewer • Updated • 98.9k • 43
zbeeb/DeepScaleR-Verifiable-Dedup
Viewer • Updated • 37.7k • 48
zbeeb/Skywork-OR1-Math-Verifiable-Dedup
Viewer • Updated • 97.8k • 57
zbeeb/Staleness-GRPO-DAPO-Math-17k
Viewer • Updated • 17k • 234
zbeeb/dapo-math-17k-qwen3-clean
Viewer • Updated • 17k • 34
zbeeb/beneficial-dpo-dataset
Viewer • Updated • 10k • 29
zbeeb/TAPS-Datasets
Viewer • Updated • 210k • 45 • 4