asim-salim/gemma-4-e4b-doc-classifier-lora

Gemma 4 E4B — Document Classifier (LoRA)

LoRA adapter finetuned from google/gemma-4-E4B-it for single-page document classification and identity-field extraction from scanned/faxed financial documents.

Classes

bank_statement, drivers_license, monthly_bill, paystub, state_id, unknown

Output format

{"schema_object_wrapper": [{"cls": "", "confidence": 0.0, "issuer": "", "holder": "", "page_marker": ""}]}

Usage

from unsloth import FastVisionModel

model, processor = FastVisionModel.from_pretrained("asim-salim/gemma-4-e4b-doc-classifier-lora")
FastVisionModel.for_inference(model)

Adapter: 147 MB LoRA over google/gemma-4-E4B-it. Trained with unsloth + TRL (PEFT 0.19.1).

License

Governed by the Gemma Terms of Use and the Gemma Prohibited Use Policy.

Limitations

Trained on a specific document corpus; not evaluated for general-purpose document understanding. Outputs must not be treated as authoritative for financial decisions.