Data and models for 'Constitutional Midtraining: Content Presence Drives Alignment Gains'. Post-BFT models are LoRA adapters on post-SFT.
-
cho-ai/constitutional-mt-data
Viewer • Updated • 2.35M • 338 • 5 -
geodesic-research/sft-warm-start-200k
Viewer • Updated • 400k • 137 -
geodesic-research/nemotron-super-120b-cc-mt-replay-only-1b-200k-sft
121B • Updated • 30 -
geodesic-research/nemotron-super-120b-cc-mt-replay-only-1b-200k-sft-gsm8k-lora
Updated