Ayansk11/FinSenti-Qwen3-1.7B

TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kPublished:Apr 8, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Ayansk11/FinSenti-Qwen3-1.7B is a 1.7 billion parameter Qwen3-based model developed by Ayansk11, specifically fine-tuned for financial sentiment analysis. It excels at classifying short financial texts (headlines, earnings snippets) into positive, negative, or neutral, providing a reasoning chain for its decision. This model is optimized for deployment on devices with limited VRAM, such as 6GB laptop GPUs, while maintaining coherent reasoning.

Loading preview...

FinSenti-Qwen3-1.7B Overview

FinSenti-Qwen3-1.7B is a 1.7 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, providing both a classification (positive, negative, neutral) and a short reasoning chain. This model is designed to be efficient, capable of running on a 6 GB laptop GPU, making it a useful middle-sized option for financial text processing.

Key Capabilities

  • Financial Sentiment Classification: Accurately classifies short financial texts (1-3 sentences) like headlines and earnings snippets.
  • Reasoning Chain Generation: Produces a concise explanation for its sentiment classification, enhancing interpretability.
  • Structured Output: Adheres to a strict <reasoning>...</reasoning><answer>...</answer> format, simplifying downstream parsing.
  • Efficient Deployment: Its 1.7B parameters allow for inference on hardware with limited VRAM (e.g., ~4GB for batch=1).

Training Details

The model was trained using a two-stage process: Supervised Fine-Tuning (SFT) on the FinSenti dataset (~15.2K samples) followed by GRPO (Generative Reinforcement Learning from Policy Optimization). GRPO utilized four reward functions (sentiment correctness, format compliance, reasoning quality, output consistency), achieving a mean reward of approximately 3.71 / 4.0 on the validation set. It was built using Unsloth and TRL on the Qwen/Qwen3-1.7B base.

Limitations

  • Short Text Focus: Not designed for long documents; context capped at 2048 tokens.
  • Single-Asset Reasoning: Classifies sentiment for individual texts, not aggregated across multiple sources.
  • No Numerical Reasoning: Cannot perform complex mathematical forecasts.
  • English Only: Training data was exclusively in English.
  • Limited Background Knowledge: Relies on its base pretraining for general knowledge, cannot look up external information.
  • Three-Label Output: Provides only positive, negative, or neutral classifications.