NOSIBLE/prediction-v1.1-base: Specialized Prediction Classification
NOSIBLE/prediction-v1.1-base is a compact 0.8 billion parameter model, fine-tuned from Qwen3-0.6B, engineered for the precise task of classifying short text snippets as containing a prediction or not. Its training on 100,000 real-world financial search results from Nosible Search Feeds provides it with robust capabilities for handling messy, unstructured financial and web data.
Key Capabilities
- Accurate Predictive Statement Detection: Reliably distinguishes predictive intent from descriptive or historical text, even with subtle cues.
- Robustness on Real-World Data: Effectively processes noisy, naturally occurring financial content without extensive preprocessing.
- Scalable Prediction Mining: Its lightweight architecture enables fast, cost-effective extraction of forecasts and estimates from large text corpora.
- Outperforms Larger LLMs: Demonstrates superior accuracy compared to larger state-of-the-art LLMs for this specific classification task, at a significantly lower operational cost.
Good For
- Identifying predictions within financial news, reports, or web content.
- Large-scale analysis of financial text for forward-looking statements.
- Applications requiring efficient and precise classification of predictive language in English financial contexts.
Important Usage Requirements
To ensure optimal performance, users must adhere to strict usage guidelines, including disabling reasoning tokens (enable_thinking=False), using a specific system prompt ("Classify whether it contains a prediction or does not contain a prediction."), and constraining output to ["prediction", "_prediction"] using grammars or regex.