Kuvera 8B

Model Description

Intended Uses & Limitations

• Answering user questions about personal finance topics such as budgeting, saving, investing, debt management, and basic financial planning.
• Powering chatbots or virtual assistants focused on financial guidance.
• Providing initial information and suggestions for common financial dilemmas.
• Use in research settings to understand how language models can address nuanced financial queries.

• Not Financial Advice: The model's responses are for informational and educational purposes only and should not be considered professional financial advice. Users should always consult with a qualified financial advisor before making financial decisions.
• Synthetic Data: While the responses are curated to be helpful, they are synthetically generated. The model might not always capture the full complexity or the most up-to-date information for every individual situation.
• Potential for Bias: The training data, although curated, may contain inherent biases present in the original Reddit queries or in the synthetic response generation process.
• Knowledge Cutoff: The model's knowledge is limited to the information present in its training data and the knowledge cutoff of its base model. It may not be aware of very recent financial events or changes in regulations.
• Non-Reasoning Base: The base model is described as "Non-reasoning." While fine-tuning on a specialized dataset can imbue some domain-specific reasoning capabilities, complex multi-step financial planning or deep inferential reasoning might be beyond its current scope.

Training Data

• Dataset: Akhil-Theerthala/Personal-Finance-Queries
• Size: Approximately 20,000 real Reddit queries.
• Responses: Synthetically curated in multiple phases.
• Key Feature: The dataset generation process paid specific attention to the basic financial necessities and the psychological conditions of the recipient when crafting the responses.

Training Procedure

• Base Model: Qwen/Qwen3-8B
• Finetuning Approach: Full Finetuning
• Hyperparameters:
  • Number of Epochs: 5
  • Learning Rate: 5e-5
  • Learning Rate Scheduler: Cosine (Min LR: 5e-6)
  • Batch Size: 24

Further Information & Collaboration

• Contact: akhiltvsn@gmail.com
• Future Work:
  • Exploring Mixture of Experts (MoE) methods for further model development, where each "expert" focuses on evaluating a core aspect of the query like, psychological intent, general social biases, region-sepecific contexts, query breakdown, possible choices and concequences etc.
• Call for Collaboration: I am a solo dude just randomly working on this project during my free time. If you are interested in this project, and want to expand the scope, then do ping me here, on Linkedin or just send me a mail.

Citation