You can deploy this model and start earning money today!

alexpin03/
gemma4-31b-it-lora-fact-check-v31

Dataset: 2000 human-annotated US/Italian political claims
Method: QLoRA (4-bit NF4, bitsandbytes) + LoRA r=16, alpha=32
Epochs: 3 | Final loss : ~0.39 | Token accuracy : ~91.7%
Framework: TRL 1.8.0, PEFT 0.19.1, Transformers 5.12.1
Hardware: 4× NVIDIA A100 80GB (CINECA Leonardo)

Name: gemma4-31b-it-lora-fact-check-v31
Availability: InStock

Views

No views yet

gemma4-31b-it-lora-fact-check-v31

QLoRA adapter (r=16, alpha=32) fine-tuned on top of google/gemma-4-31b-it for political claim fact-checkability classification.

Given a political statement (and optional context), the model outputs a JSON object:

json

{"reason": "one sentence naming the decisive rule",
  "label": "FACT-CHECKABLE" | "NOT-FACT-CHECKABLE",
  "factual_form_but_vague": true | false}