financial-gpt-oss-20b-q8

Financial GPT-OSS 20B Q8 - Quantized Financial Analysis Model

Model Description

This is a quantized Q8_0 GGUF version of a fine-tuned financial analysis model based on GPT-OSS 20B. The model has been specialized for financial market analysis, providing technical analysis, risk assessments, trading signals, and price forecasts for various securities.

Model Details

• Base Model: GPT-OSS 20B (unsloth/gpt-oss-20b-unsloth-bnb-4bit)
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Quantization: Q8_0 GGUF format for efficient inference
• Model Size: 12.1 GB (quantized from ~40GB original)
• Context Length: 131,072 tokens
• Architecture: Mixture of Experts (MoE) with 32 experts, 4 active

Training Data

The model was fine-tuned on 22,250 financial conversation pairs covering:

Supported Assets

• Major Tech Stocks: AAPL, AMZN, NVDA, META, TSLA, NFLX, ADBE, ORCL, INTC, CSCO
• Healthcare: UNH, ABBV, GILD, TMO
• Energy: CVX, XOM, VLO
• Financial Services: WFC, USB, TFC, SCHW, SPGI
• Consumer: WMT, COST, TGT, SBUX, UPS
• And many more S&P 500 companies

Analysis Types

1. Risk Assessments - Comprehensive risk analysis including volatility metrics, technical risks, and mitigation strategies
2. Trading Signals - Buy/Sell/Hold recommendations with entry points, stop losses, and price targets
3. Price Forecasts - Short-term (1 week, 1 month, 3 month) price projections with scenario analysis
4. Technical Analysis - MACD, RSI, moving averages, support/resistance levels
5. Market Context - Sector performance, relative positioning, institutional activity indicators

Capabilities

Technical Analysis

• Momentum Indicators: RSI, MACD, signal line crossovers
• Trend Analysis: Moving averages (SMA20, SMA50), trend strength assessment
• Support/Resistance: Key price levels identification
• Volume Analysis: Volume trends and patterns

Risk Management

• Volatility Assessment: Daily volatility calculations and sector comparisons
• Position Sizing: Portfolio allocation recommendations (typically 2-3% limits)
• Stop-Loss Strategy: Risk-based exit points
• Risk/Reward Ratios: Systematic 1:2 risk/reward targeting

Market Intelligence

• Sector Rotation: Technology sector focus with broader market context
• Relative Performance: Security vs sector performance comparison
• Market Phase: Trending, consolidation, reversal pattern recognition
• Institutional Signals: Volume and momentum-based institutional activity indicators

Usage Examples

Risk Assessment Query

User: "Can you assess the risk level for AAPL?"

Model Response:
Risk Assessment for AAPL:

**Overall Risk Level: Low**

**Volatility Analysis:**
- Daily volatility: 0.27%
- Sector: tech (High volatility)

**Technical Risks:**
- RSI at 44.53: Neutral positioning
- Trend strength: Moderate

**Key Risk Factors:**
1. Market correlation risk
2. Sector-specific headwinds
3. Technical resistance at $146.89
4. Support breakdown below $132.90

**Risk Mitigation:**
- Position sizing: Limit to 2-3% of portfolio
- Use stop-loss orders
- Consider hedging strategies

Trading Signal Query

User: "Should I buy or sell AAPL?"

Model Response:
Trading Signal for AAPL: **BUY**

Current Price: $147.79
Daily Change: -0.35%

**Signal Rationale:**
- MACD crossed above signal line
- RSI at 59.12 indicates neutral momentum
- Price is above 20-day SMA ($144.40)

**Risk Management:**
- Entry: $147.79
- Stop Loss: $140.40 (-5%)
- Target: $162.56 (+10%)
- Risk/Reward Ratio: 1:2

Note: This is technical analysis only. Consider fundamental factors and your risk tolerance.

Model Performance

• Training Epochs: 1000 steps with checkpoints at 250, 500, 750, and 1000
• Architecture: Leverages MoE efficiency with 32 experts, 4 active per token
• Optimization: LoRA fine-tuning preserves base model capabilities while adding financial expertise
• Quantization: Q8_0 maintains model quality while reducing inference requirements

Technical Specifications

• Model Type: Causal Language Model (Financial Analysis Specialist)
• Embedding Dimension: 2,880
• Feed Forward Dimension: 2,880
• Attention Heads: 64 (8 key-value heads)
• RoPE Theta: 150,000
• Vocabulary Size: 201,088 tokens
• Quantization Method: Q8_0 with MXFP4 expert weights

Limitations and Disclaimers

⚠️ Important Financial Disclaimer

1. Not Financial Advice: This model provides technical analysis and educational content only. It is not licensed financial advice.

2. Risk Warning:

   • All trading involves risk of loss
   • Past performance does not guarantee future results
   • Consider your risk tolerance and investment objectives

3. Technical Limitations:

   • Based on technical analysis patterns only
   • Does not incorporate fundamental analysis
   • May not reflect real-time market conditions
   • Training data has a specific time cutoff

4. Model Limitations:

   • Responses are generated, not real-time analysis
   • No access to current market data during inference
   • Should be used as one input among many in decision-making

Usage Recommendations

1. Combine with Other Analysis: Use alongside fundamental analysis, news, and current market conditions
2. Risk Management: Always implement proper risk management techniques
3. Paper Trading: Test strategies with paper trading before live implementation
4. Professional Consultation: Consult with licensed financial advisors for personalized advice
5. Continuous Learning: Stay updated with market conditions and economic indicators

Installation and Usage

For GGUF-compatible Software

• Ollama: ollama run beenyb/financial-gpt-oss-20b-q8
• llama.cpp: Compatible with Q8_0 quantization
• Jan AI: Import as local model
• Open WebUI: Add as custom model

System Requirements

• RAM: Minimum 16GB, recommended 32GB
• Storage: 15GB free space
• CPU: Modern multi-core processor
• GPU: Optional but recommended for faster inference

License

This model is released under the Apache 2.0 License. Please ensure compliance with your intended use case.

Citation

If you use this model in your research or applications, please cite:

@misc{financial-gpt-oss-20b-q8,
  title={Financial GPT-OSS 20B Q8: Fine-tuned Financial Analysis Model},
  author={beenyb},
  year={2025},
  publisher={Hugging Face Hub},
  url={https://huggingface.co/beenyb/financial-gpt-oss-20b-q8}
}

---

Generated with Claude Code | Training completed: January 2025