darkknight25/trading_dataset_v2

📈 LLM Trading Instruction Dataset – V2 (2023–2025)

Dataset Version: 2
Filename: llm_trading_dataset_20250629_115817.jsonl
Entries: 157k
Period Covered: 2023–2025
Format: JSON Lines (.jsonl)
Task: Instruction Tuning for Financial Signal Classification
Target Models: LLaMA, Mistral, GPT-J, Falcon, Zephyr, DeepSeek, Qwen

🧠 Overview

This second version of the dataset expands the time horizon and depth of training data for instruction-tuned LLMs by covering real-world market indicators from 2023 through 2025. It enables financial models to learn patterns, sentiment, and timing in Buy/Sell signal generation.

📁 Dataset Format

Each entry follows the instruction tuning schema:

{
  "instruction": "Given technical indicators, predict if it's a Buy or Sell signal.",
  "input": "AAPL on 2025-03-20 with indicators: EMA20=224.55, EMA50=232.09, BB_upper=254.87, BB_lower=203.31, MACD=-6.81, MACD_signal=-4.88, RSI=33.61, CCI=-85.31, STOCH_K=15.95, STOCH_D=15.7",
  "output": "Buy"
}

📌 Fields:

    instruction – Prompt for LLM task (uniform for all entries)

    input – Date, stock symbol, and associated technical indicators

    output – Predicted trading signal: "Buy" or "Sell"

🔍 Technical Indicators Used

Indicator	Description
EMA20 / EMA50	Short and medium-term exponential MA
BB_upper/lower	Bollinger Bands – price volatility zones
MACD, MACD_sig	Momentum crossover indicators
RSI	Overbought/Oversold indicator (0–100)
CCI	Momentum-based deviation indicator
STOCH_K / D	Stochastic oscillator %K/%D lines

🔧 Example Usage

import json

with open("llm_trading_dataset_20250629_115817.jsonl") as f:
    for line in f:
        ex = json.loads(line)
        print("Prompt:", ex["instruction"])
        print("Indicators:", ex["input"])
        print("Decision:", ex["output"])

🧪 Use Cases

Finetune instruction-tuned LLMs for trading automation

Evaluate transformer models for financial decision tasks

Build explainable AI advisors using LLM-based logic

Backtest models on realistic multi-year indicators

Create copilot assistants for traders & hedge funds

📌 Version Info Version Range Covered Notes v1 2025 only Initial dataset release v2 2023–2025 Extended multi-year training set

📜 License

MIT License – Open for use, distribution, and modification.

Attribution recommended for research and commercial tools.

🤝 Contact

📧 [email protected]

🧠 AI/Trading Collab: DM for finetuning support or strategy model help