Information Extraction on Campaign-finance filings
98.71Mean per field AccuracyFine-tuning
Evaluation results
| Method | Links | |
|---|---|---|
| Fine-tuningTrain Tokens (Base Model)=502968 / 11178, Inference Tokens (Base Model)=62099 / 13772026.04 | 98.71 | |
| KNNInference Tokens (Base Model)=219450 / 13712026.04 | 55.12 | |
| DSPy GEPATrain Tokens (Base Model)=2142493 / 20824, Train Tokens (Reflection Model)=206378 / 91251, Inference Tokens (Base Model)=265211 / 18472026.04 | 53.21 | |
| DSPy BootstrapFewShot (#12)Train Tokens (Base Model)=55479 / 144, Inference Tokens (Base Model)=736331 / 18292026.04 | 42.95 | |
| DSPy MIPROv2Train Tokens (Base Model)=6399076 / 40216, Inference Tokens (Base Model)=358187 / 18382026.04 | 42.31 | |
| Initial promptInference Tokens (Base Model)=62099 / 16512026.04 | 36.54 | |
| AIRTrain Tokens (Embed Model)=200111 / -, Train Tokens (Base Model)=176632 / 3559, Train Tokens (Reflection Model)=102743 / 52487, Inference Tokens (Base Model)=172235 / 16582026.04 | 35.9 | |
| TextGradTrain Tokens (Base Model)=284301 / 1723, Train Tokens (Reflection Model)=218746 / 194031, Inference Tokens (Base Model)=187991 / 14042026.04 | 25.64 |