espressovi
/

BODHI-qwen-3-math-8b-rlvr

Text Generation

Model card Files Files and versions

BODHI

Artifact for BODHI project. This model was RL-trained starting from espressovi/BODHI-qwen-3-8b-distil.

Performance

The following results are at T=1.0, 16K tokens and pass@8.

AIME25: 32.73 %

Downloads last month: 95

Safetensors

Model size

8B params

Tensor type

F32

·

Model tree for espressovi/BODHI-qwen-3-math-8b-rlvr

Base model

Qwen/Qwen3-8B-Base

Finetuned

(507)

this model

Dataset used to train espressovi/BODHI-qwen-3-math-8b-rlvr

Collection including espressovi/BODHI-qwen-3-math-8b-rlvr

BODHI

Artifacts for the paper titled "BODHI: Can LLMs Branch Out and Discover Heterogeneous Inferences?" • 18 items • Updated Jul 17