Constitutional Midtraining - a cho-ai Collection

cho-ai 's Collections

updated 2 days ago

Data and models for 'Constitutional Midtraining: Content Presence Drives Alignment Gains'. Post-BFT models are LoRA adapters on post-SFT.

Upvote

cho-ai/constitutional-mt-data

Viewer • Updated Jul 31 • 2.35M • 344 • 5

Note Constitutional corpus, generated from values in Anthropic's Constitution. Documents are tagged by cluster and value. Reasoning variants and assembled training mixes available.
geodesic-research/sft-warm-start-200k

Viewer • Updated Apr 13 • 400k • 138

Note Shared instruction-tuning mix (200K examples) used for the value-neutral SFT stage.
geodesic-research/nemotron-super-120b-cc-mt-replay-only-1b-200k-sft

121B • Updated 3 days ago • 29

Note Control, post-SFT (original weights, re-uploaded Oct 2026)
geodesic-research/nemotron-super-120b-cc-mt-replay-only-1b-200k-sft-gsm8k-lora

Updated Jul 14

Note Control, post-BFT (LoRA adapter; base: Control, post-SFT)
geodesic-research/nemotron-super-120b-cc-mt-curriculum-dr

124B • Updated Jul 9 • 16

Note Curriculum-DR, post-MT
geodesic-research/nemotron-super-120b-cc-mt-curriculum-dr-200k-sft

124B • Updated Jul 9 • 16

Note Curriculum-DR, post-SFT
geodesic-research/nemotron-super-120b-cc-mt-curriculum-dr-200k-sft-gsm8k-lora

Updated Jul 14

Note Curriculum-DR, post-BFT (LoRA adapter; base: Curriculum-DR, post-SFT)
geodesic-research/nemotron-super-120b-cc-mt-uniform-dr-200k-sft

124B • Updated Jul 9 • 8

Note Uniform-DR, post-SFT
geodesic-research/nemotron-super-120b-cc-mt-uniform-dr-200k-sft-gsm8k-lora

Updated Jul 14

Note Uniform-DR, post-BFT (LoRA adapter; base: Uniform-DR, post-SFT)
geodesic-research/nemotron-super-120b-cc-mt-curriculum-nodr

124B • Updated Jul 9 • 11

Note Curriculum-noDR, post-MT
geodesic-research/nemotron-super-120b-cc-mt-curriculum-nodr-200k-sft

124B • Updated Jul 9 • 13

Note Curriculum-noDR, post-SFT
geodesic-research/nemotron-super-120b-cc-mt-curriculum-nodr-200k-sft-gsm8k-lora

Updated Jul 14

Note Curriculum-noDR, post-BFT (LoRA adapter; base: Curriculum-noDR, post-SFT)
geodesic-research/nemotron-super-120b-cc-mt-uniform-nodr

124B • Updated Jul 9 • 12

Note Uniform-noDR, post-MT
geodesic-research/nemotron-super-120b-cc-mt-uniform-nodr-200k-sft

124B • Updated Jul 9 • 15

Note Uniform-noDR, post-SFT
geodesic-research/nemotron-super-120b-cc-mt-uniform-nodr-200k-sft-gsm8k-lora

Updated Jul 14

Note Uniform-noDR, post-BFT (LoRA adapter; base: Uniform-noDR, post-SFT)

Upvote