jhon53/Llama3_1_8B_Finance_QLoRA-merged-16bit

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:May 27, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

jhon53/Llama3_1_8B_Finance_QLoRA-merged-16bit is an 8 billion parameter Llama 3.1 model fine-tuned by jhon53 using QLoRA for 3-class financial sentiment analysis (negative, neutral, positive). This model excels at classifying financial sentiment, achieving significantly improved accuracy and Macro-F1 scores on both in-domain (FPB) and out-of-domain (FiQA-SA) datasets compared to its base model. It is specifically optimized for financial text classification tasks, leveraging a 32768 token context length.

Loading preview...

Llama 3.1 8B Financial Sentiment Model

This model, developed by jhon53, is a fine-tuned version of the Meta-Llama-3.1-8B-Instruct base model, specifically optimized for financial sentiment analysis. It utilizes QLoRA (4-bit NF4 frozen base with bf16 LoRA adapters) to efficiently achieve high performance on a specialized task.

Key Capabilities

  • 3-Class Financial Sentiment Analysis: Accurately classifies financial text into negative, neutral, or positive sentiment categories.
  • Enhanced Performance: Demonstrates substantial improvements over the base Llama 3.1 8B model in financial sentiment tasks. For instance, it achieves 0.9748 accuracy on FPB in-domain data (up from 0.8908) and 0.9402 accuracy on FiQA-SA out-of-domain data (up from 0.8120).
  • Efficient Fine-tuning: Fine-tuned using QLoRA on the FinGPT/fingpt-sentiment-train dataset, making it a specialized and efficient solution for financial NLP.

Good for

  • Financial Text Classification: Ideal for applications requiring precise sentiment analysis of financial news, reports, social media, or other textual data.
  • Research and Development: Provides a strong baseline for further research into financial NLP and fine-tuning techniques on large language models.
  • Resource-Efficient Deployment: The QLoRA fine-tuning approach allows for more efficient deployment and inference compared to full fine-tuning.