NOSIBLE/financial-sentiment-v1.2-base

Overview

NOSIBLE/financial-sentiment-v1.2-base is a specialized financial sentiment classification model, fine-tuned from Qwen3-0.6B-Base. It classifies short text snippets into "positive", "neutral", or "negative" financial impact categories. This version, v1.2, significantly expands on its predecessor by offering multilingual coverage across 94 languages, including English and 93 additional languages, while maintaining English performance.

Key Capabilities

• Multilingual Financial Sentiment: Classifies financial sentiment in 94 languages, improving global news and search feed analysis.
• Wider Topic Coverage: Enhanced classification for currency and G10-geography related texts, beyond just company news.
• Performance Improvement: Achieves a +6.47pp accuracy increase on multilingual validation sets and +8.87pp on currency/geography feeds compared to v1.1.
• Instruction Following: Reframes sentiment classification as instruction following, producing a single label token per input.

Strict Usage Requirements

To ensure optimal performance, users must adhere to specific requirements:

• Disable thinking tokens (enable_thinking=False).
• Use the exact system prompt: "Classify the financial sentiment as positive, neutral, or negative."
• Constrain output to ["positive", "neutral", "negative"] using grammars, regex, or guided decoding.

Limitations

As a small model (0.8B parameters), it is optimized for specific classification tasks and may struggle with highly nuanced text. It is domain-specific to financial contexts and not suitable for general sentiment analysis or aspect-based sentiment. The model analyzes sentiment but does not verify factual accuracy.