LoRAPERTVS

Inspiration

In earlier work, we tested whether SKILL.md files (instruction files that steer AI agents) make Apertus better at software engineering, using benchmarks for merge conflict resolution, secure code generation and program repair. Skills mostly changed the form of Apertus's answers, not its ability. Only Apertus-8B on ConGra improved; everything else stayed flat or got worse. So we asked: if prompts only reshape output, can a lightweight LoRA adapter change what Apertus can actually do?

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for LoRAPERTVS

Built With

• python