alpha7845/lora-lens

LoRA Lens

LoRA Lens is a small, transparent audit tool for parameter-efficient fine-tuning releases. Upload an adapter_config.json or paste one into the editor to get an immediate reviewer-facing readout.

What it checks

• Base model and PEFT type
• Rank, alpha, and the effective scaling factor
• Target modules and missing metadata
• Dropout and modules_to_save review notes
• Conservative adapter-size estimates when hidden_size is available
• Optional trainable-parameter share against a supplied base-model size

Why it exists

Adapter repositories are often technically valid but difficult to review. LoRA Lens makes the important release metadata visible before a model card is written, without uploading weights or requiring an API key.

Scope

This is a metadata and sanity-check tool, not a substitute for evaluation. Parameter estimates depend on the base architecture; verify claims against the actual model before publishing.

Local run

pip install -r requirements.txt
python app.py

Built by Praveen as part of an open model tooling portfolio.