Gemma 2 2b fine tuned for sentiment analysis of financial articles, give it an article -> outputs a specific JSON format with confidence and reasoning values:

{
  "reasoning": "The article reports that Qualcomm's CEO Cristiano Amon expressed concerns about the phone market recovering more slowly than expected, which led to a decline in stock value after initially rising due to strong earnings projections. The company projected flattish or single-digit percentage growth for units this year, indicating a slow recovery from a prolonged slump.",
  "sentiment": -0.15,
  "confidence": 0.75
}

Confidence and variance get shakier the lower down the quants, but overall similar performance.