Fine-tuned Zero-Shot Classification Model

This model was fine-tuned using the SmartShot approach on a synthetic dataset derived from LLaMa to improve zero-shot classification performance.

Model Details

Base Model: MoritzLaurer/roberta-base-zeroshot-v2.0-c
Training Data: Synthetic data created for natural language inference tasks
Fine-tuning Method: SmartShot approach with NLI framing

Usage

from transformers import pipeline

classifier = pipeline("zero-shot-classification", model="gincioks/smartshot-zeroshot-finetuned-v0.1.2")
text = "Shares of Hyundai Motor jumped nearly 8% on Wednesday, a day after South Korea announced a 'green new deal' to spur use of environmentally friendly vehicles."
labels = ["Shares that rise due to the 'green new deal'", "Shares that fall due to the 'green new deal'"]
results = classifier(text, labels)
print(results)

Training Procedure

This model was fine-tuned with the following parameters:

Learning rate: 2e-05
Epochs: 1
Batch size: 1
Warmup ratio: 0.06

Performance and Limitations

The model achieves improved performance on zero-shot classification tasks but may still have limitations in domains not covered by the training data.