BugSum-Qwen-7B-DPO

BugSum-Qwen-7B-DPO

This is an experimental Direct Preference Optimization LoRA variant of BugSum-Qwen-7B for concise English bug-report summarization. It was trained by Md. Abid Shahriar during thesis research.

The repository contains an adapter, not a standalone model. Load it on Qwen/Qwen2.5-Coder-7B-Instruct with PEFT. The training corpus is not included because its source datasets have separate access and licensing conditions.

Method

The experiment used 256 preference records, 4-bit loading, and three training epochs. It started from the supervised BugSum adapter.

The thesis evaluation did not establish a statistically significant improvement over the supervised baseline, so this artifact should be treated as an experimental research checkpoint rather than a proven upgrade.

Intended use and limitations

Use it for research, educational comparison, or human-reviewed bug-summary drafting. It may omit important facts or hallucinate details and should not drive automated production decisions.

Licensing

The base model is distributed separately under Apache-2.0. This adapter is currently shared privately for research review while a project-level release license is finalized. Dataset rights are not transferred.