Ayansk11/FinSenti-Qwen3-0.6B
Ayansk11/FinSenti-Qwen3-0.6B is a 0.6 billion parameter Qwen3-based model developed by Ayansk11, specifically fine-tuned for financial sentiment analysis. It excels at classifying short English financial texts (headlines, earnings snippets) into positive, negative, or neutral, providing a short reasoning chain. This model is optimized for deployment on resource-constrained hardware like Raspberry Pi or older laptops due to its small size and efficient inference requirements.
Loading preview...
FinSenti-Qwen3-0.6B: Financial Sentiment Analysis Model
FinSenti-Qwen3-0.6B is a compact 0.6 billion parameter model from the Qwen3 family, developed by Ayansk11 as part of the FinSenti collection. It is specifically fine-tuned for financial sentiment analysis, designed to process short financial texts such as headlines and earnings snippets. The model provides a sentiment classification (positive, negative, or neutral) along with a concise reasoning chain, adhering to a strict <reasoning>...</reasoning><answer>...</answer> output format for easy downstream parsing.
Key Capabilities
- Financial Text Classification: Accurately classifies short English financial texts (1-3 sentences) into positive, negative, or neutral sentiment.
- Reasoning Chain Generation: Produces a brief, human-readable reasoning chain explaining its sentiment decision.
- Structured Output: Guarantees a consistent XML-like output format, simplifying integration into automated workflows.
- Resource Efficient: Its small size (0.6B parameters) and low memory footprint (approx. 2GB GPU memory for inference) make it suitable for deployment on edge devices or older hardware.
Training and Performance
The model underwent a two-stage training process: Supervised Fine-Tuning (SFT) on ~15.2K samples from the FinSenti Dataset, followed by GRPO (Generative Reinforcement Learning with Policy Optimization) using four equally weighted reward functions (sentiment correctness, format compliance, reasoning quality, output consistency). It achieved a mean reward of approximately 3.59 / 4.0 on the validation set, indicating high performance in sentiment accuracy and format adherence. The model is trained exclusively on English financial news and performs best within this domain.
Limitations
This model is not designed for long documents (context capped at 2048 tokens), multi-asset or numerical reasoning, languages other than English, or tasks requiring extensive background knowledge. It provides a hard cutoff of three sentiment labels (positive/negative/neutral).