Total
Olmo
€2.26M
Fable
€1.51B
Total
Olmo
5.81T+
Fable
200T+
Total
Olmo
1.2 × 1024
Fable
2 × 1027
Pre-trainingPost-training
AspectPre-trainingSFTRL
Data source
Existing text
Commissioned text
Text produced by the model itself
Signal
Token prediction
Token prediction (answer only)
Reward for the whole rollout
Whose judgement?
Everyone who wrote, whoever filtered
The writer (human or AI)
HumansRLHFAIRLAIFCodeRLVRConstitutionCAI