Moonlight556/sokrates-qwen3-8b-prontoqa-oak-dpo-iter1

SOKRATES: Qwen3-8B PrOntoQA OaK-DPO Iteration 1

First DPO iteration achieving 96.8% accuracy on PrOntoQA.

Performance

Stage	Accuracy
SFT	93.3%
DPO Iter 1	96.8%

Usage

from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained(
    "Moonlight556/sokrates-qwen3-8b-prontoqa-oak-dpo-iter1",
    torch_dtype="bfloat16"
)