ConfigGuard SLM
This model is a LoRA adapter fine-tuned on MutConf-Bench v2 for configuration error detection and repair.
What it does
It takes a short mutated configuration snippet and returns strict JSON with:
- config_type
- mutation_type
- line_number
- explanation
- minimal_fix
Training setup
- Base model: Qwen/Qwen2.5-1.5B-Instruct
- Dataset: MutConf-Bench v2
- Training run: qlora-full
- Train loss: 0.1896
- Eval loss: 0.0419
- Train samples: 650
- Validation samples: 80
Usage
Load the adapter with the base model and prompt the model to return valid JSON only.
Notes
The repository contains the adapter weights and tokenizer files. The model is optimized for short, structured configuration-debugging tasks rather than long-form generation.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support