seibergwitten/dpo-qwen-cot-merged.ver0
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 24, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

seibergwitten/dpo-qwen-cot-merged.ver0 is a 4 billion parameter language model fine-tuned from Qwen/Qwen3-4B-Instruct-2507. This model utilizes Direct Preference Optimization (DPO) via Unsloth to enhance reasoning capabilities, specifically Chain-of-Thought (CoT), and improve structured response quality. It is designed for applications requiring aligned and coherent outputs, particularly in complex reasoning tasks.

Loading preview...