VerticalIntelligence/aihedged-trading-datasets

AIHedged Trading Datasets — Bybit Tick Data

Tick-level trade data from Bybit for ETH/BTC/SOL/DOGE perpetuals, captured 2025-04-20 to 2026-04-19 (1 year).

Coverage

Symbol	Trades
ETH/USDT	784,357,803
BTC/USDT	594,109,922
SOL/USDT	209,092,277
DOGE/USDT	128,124,272
Total	~1.72 billion

Schema

Column	Type	Description
ts	timestamp[us, tz=UTC]	Trade timestamp (microsecond precision)
symbol	string	Trading pair (e.g. "BTC/USDT")
side	string	"buy" or "sell" (taker side)
price	float64	Trade price (quote asset per base)
qty	float64	Trade quantity (base asset)
exchange	string	Exchange identifier ("bybit")
trade_id	string	Exchange-assigned unique trade ID

Layout

Partitioned by symbol and month, ordered by ts within each file:

trades_tick/
├── BTCUSDT/
│   ├── 2025-04.parquet
│   ├── 2025-05.parquet
│   └── ...
├── ETHUSDT/
├── SOLUSDT/
└── DOGEUSDT/

Compression: zstd. Row group size: 100k rows.

Loading

datasets library

from datasets import load_dataset
ds = load_dataset("VerticalIntelligence/aihedged-trading-datasets", split="train")

DuckDB (recommended for SQL queries)

import duckdb
con = duckdb.connect()
con.sql("""
    SELECT * FROM 'hf://datasets/VerticalIntelligence/aihedged-trading-datasets/trades_tick/BTCUSDT/2025-04.parquet'
    WHERE side = 'buy'
    LIMIT 100
""")

Polars / pandas

import polars as pl
df = pl.read_parquet("hf://datasets/VerticalIntelligence/aihedged-trading-datasets/trades_tick/BTCUSDT/2025-04.parquet")

Source

Captured directly from Bybit public WebSocket trade feed via the regime-trader research stack. Trade IDs are exchange-assigned and globally unique per symbol.

License

CC BY 4.0 — free to use with attribution.