Olmo3-B3PO_MC-7B
📄 Paper: Parameter Exploration for RLVR via Variational Learning · arXiv:2608.09805
📦 Code: insait-institute/c3po
Olmo-3 7B fine-tuned with the decoupled Monte-Carlo variant of B3PO, from the paper "Parameter Exploration for RLVR via Variational Learning".
3PO is a family of parameter-space exploration strategies for Reinforcement Learning with Verifiable Rewards (RLVR). Instead of relying only on action-space heuristics (temperature, clipping, entropy bonuses), 3PO samples model weights from an approximate posterior learned with the variational optimizer IVON; the amount of weight noise becomes an extra control lever for exploration.
B3PO_MC combines B3PO rollouts with the decoupled variance-reduction scheme described in Appendix C of the paper: the IVON gradient estimate uses several Monte-Carlo weight samples that are decoupled from the rollout perturbation, reducing gradient variance without changing the rollout budget.
If you intend to reproduce this run, this corresponds to setting
DECOUPLED_MC_SAMPLES > 1(ivon_config.decoupled_mc_samples) on top of the B3PO configuration.
Training
| Base / warm-start | BayesRL/Olmo3-IVON-SFT-7B |
| Foundation model | allenai/Olmo-3-1025-7B |
| Algorithm | B3PO + decoupled MC variance reduction (Appendix C) |
| RL data | DAPO-Math-17k |
| Optimizer | IVON, lr 1.0, ESS (λ) 1e9 |
| Hardware | 8× NVIDIA H200 (144 GB) |
Evaluation
Evaluated on AIME 2024–2026, MATH-500, AMC 2023, and Minerva. See the paper for full results.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("BayesRL/Olmo3-B3PO_MC-7B")
tok = AutoTokenizer.from_pretrained("BayesRL/Olmo3-B3PO_MC-7B")
Citation
@misc{venkatkrishna2026parameterexploration,
title={Parameter Exploration for RLVR via Variational Learning},
author={Vatsal Venkatkrishna and Nico Daheim and Iryna Gurevych},
year={2026},
eprint={2608.09805},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2608.09805},
}