darkknight25/TRADING_DATASET_v1

LLM Trading Instruction Dataset – 2025 Edition

Filename: llm_trading_dataset_20250629_115540.jsonl
Size: 20,306 entries
Format: JSON Lines (.jsonl)
Created: June 29, 2025
Task Type: Instruction Tuning for Financial Signal Prediction
Target Use: Finetuning Instruction-Following LLMs (e.g., LLaMA, Mistral, GPT-NeoX)

🧠 Dataset Overview

This dataset is crafted to train and evaluate large language models for predicting Buy/Sell signals based on technical indicators in real-world market conditions.

Each entry follows a structured instruction format, making it suitable for instruction-tuned or reinforcement-aligned LLMs.

📁 Format

Each line contains a JSON object with the following fields:

{
  "instruction": "Given technical indicators, predict if it's a Buy or Sell signal.",
  "input": "AAPL on 2025-03-14 with indicators: EMA20=229.85, EMA50=235.28, BB_upper=257.98, BB_lower=212.55, MACD=-5.63, MACD_signal=-2.33, RSI=30.81, CCI=-176.99, STOCH_K=12.19, STOCH_D=7.04",
  "output": "Buy"
}

🔍 Fields:

instruction: Standardized prompt for LLMs

input: Stock ticker, date, and key technical indicators

output: Target label – "Buy" or "Sell"

🛠️ Use Cases

Finetuning instruction-following models for financial forecasting

Training agents to act as trading advisors

Simulating AI portfolio decision-making

Building autonomous trading copilots

Researching interpretable signal generation

📉 Indicators Included

EMA20, EMA50 (Exponential Moving Averages)

BB_upper, BB_lower (Bollinger Bands)

MACD, MACD_signal (Trend Momentum)

RSI (Relative Strength Index)

CCI (Commodity Channel Index)

STOCH_K, STOCH_D (Stochastic Oscillator)

import json

with open("llm_trading_dataset_20250629_115540.jsonl") as f:
    for line in f:
        sample = json.loads(line)
        print("Prompt:", sample["instruction"])
        print("Input:", sample["input"])
        print("Target:", sample["output"])

📜 License

MIT License (customizable) Attribution recommended for academic or commercial use. 🤝 Contact

For collab, model finetuning, or tool integrations: 📧 [email protected]