MoonFinance โ Version 3 (March 2026)
MoonFinance is a specialized financial reasoning model suite developed as part of an independent startup research initiative. This release represents Version 3 (March 2026) of the MoonFinance model line, incorporating improved domain adaptation, reasoning stability, and updated financial datasets.
Model Overview
- Model Name: MoonFinance
- Version: v3.1
- Developed by: SoumilB7
- Base Model:
unsloth/llama-3-8b-bnb-4bit - Architecture: Quantized LLaMA-3 with Low-Rank Adapter (LoRA) finetuning
- Primary Domain: Financial analysis, trading signals, macro reasoning, retrieval-augmented financial QA
- License: CC-BY-4.0
This model is designed to assist in:
- Fundamental equity analysis
- Technical indicator interpretation
- Financial news reasoning
- Strategy hypothesis generation
- Structured financial Q&A pipelines
- Retrieval-augmented generation (RAG) workflows
Version 3 Improvements (March 2026)
This release introduces:
- Expanded financial training corpus updated to March 2026 market context
- Improved multi-step reasoning consistency
- Better handling of noisy market narratives and speculative signals
- Stability improvements in long-context inference
- Enhanced alignment for structured financial outputs
- Optimized LoRA adaptation layers for reduced hallucination in domain queries
Weights are optimized for efficient inference on consumer GPUs via 4-bit quantization.
Model tree for SoumilB7/Moonfinance_Fundamental
Base model
meta-llama/Meta-Llama-3-8B Quantized
unsloth/llama-3-8b-bnb-4bit