espressovi
/

BODHI-qwen-2.5-32b-distil

Text Generation

Model card Files Files and versions

BODHI

Artifact for the BODHI project. This math model is created by Long-CoT distillation starting from Qwen/Qwen2.5-32B.

Performance

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

AIME25: 59.23 %

Downloads last month: 62

Safetensors

Model size

33B params

Tensor type

BF16

·

Model tree for espressovi/BODHI-qwen-2.5-32b-distil

Base model

Qwen/Qwen2.5-32B

Finetuned

(128)

this model

Dataset used to train espressovi/BODHI-qwen-2.5-32b-distil

Collection including espressovi/BODHI-qwen-2.5-32b-distil

BODHI

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