Load annotation data
Upload a FIMI Data Labeler JSON report or paste its contents above, then click Load pasted JSON.
How to use this data to improve model performance
1. Few-shot prompting (Labeler)
Use "Copy few-shot examples" and paste them into the Data Labeler’s optional few-shot field. When you click "Suggest labels (BYOT AI)", the model will see your human-labeled examples first, which improves consistency and accuracy.
2. Fine-tuning
Use "Export for fine-tuning (JSONL)" to download a JSONL file. Each line is a training example in chat format (user message = text, assistant message = JSON labels). You can use this with:
- OpenAI fine-tuning (GPT) or custom training pipelines
- Local models (e.g. Ollama) via frameworks that accept chat-style JSONL
3. Evaluation
Keep a held-out set of annotations (e.g. 20%) and run the Labeler’s "Suggest labels" on the same text. Compare AI suggestions to your labels to measure precision and drift over time.