All Tools
O
Fine-tuningFreeOpen Source
OPENINSTRUCT
AllenAI post-training stack for SFT, DPO, and RLVR
Apache-2.0
ABOUT
Open instruction-tuning papers often ship one-off scripts that do not cover SFT, DPO, and RL with verifiable rewards in one repo. OpenInstruct unifies those stages on public data so labs can reproduce Tulu 3 / OLMo instruct recipes instead of stitching trainers together.
INTEGRATION GUIDE
1. Supervised-finetune Llama 3.1 or OLMo on Tulu public instruction mixes
2. Run DPO preference tuning from OpenInstruct configs on 8 GPUs
3. Train RLVR with GRPO on code or math verifiable-reward tasks
4. Reproduce Tulu 3 post-training stages from the published scripts
TAGS
pythonsftdporlvrtuluolmoallenaipost-training