A financial intelligence model fine-tuned on 527 historical market analysis training pairs spanning 2016–2026, covering market movements, economic indicators, geopolitical events, and their outcomes.

Model Details

Property	Value
Base Model	Qwen/Qwen2.5-14B-Instruct
Fine-tuning	LoRA (rank 64, alpha 128) via Unsloth
Training Data	527 weekly market analysis pairs (2016–2026)
Quantization	Q4_K_M via llama.cpp
File Size	~8.4 GB
Context Length	4096 tokens
Format	GGUF (compatible with Ollama, llama.cpp, LM Studio)

Training Data Sources

Yahoo Finance: Daily OHLCV for 25 major assets (S&P 500, NASDAQ, Gold, Oil, BTC, etc.)
FRED: Economic indicators (GDP, CPI, unemployment, fed funds rate, yield curve)
Wikipedia Current Events: 90,745 curated events across 122 months
GDELT 2.0: Global news coverage from 75 weeks

Usage with Ollama

bash

# Download the GGUF file
# Create a Modelfile:
cat > Modelfile << 'EOF'
FROM ./finsight_q4km.gguf
PARAMETER temperature 0.3
PARAMETER num_ctx 4096
SYSTEM "You are FinSight, an expert AI financial analyst..."
EOF

ollama create finsight -f Modelfile
ollama run finsight "What are the market implications of rising inflation?"

Capabilities

Market Analysis: Interprets market movements with historical context
Historical Pattern Matching: References similar past periods when analyzing current events
Trend Prediction: Directional calls with confidence levels for major asset classes
Risk Assessment: Identifies risk factors based on historical parallels

Part of FinSight AI

This model powers FinSight AI, a real-time financial intelligence platform with RAG, vector search (Qdrant), and a live dashboard.

License

Apache 2.0 (same as Qwen 2.5 base model)

vivek-nalliah/finsight-qwen2.5-14b-gguf

FinSight AI — Fine-tuned Qwen 2.5 14B (Q4_K_M GGUF)