saidurga001301/mathmetics-dataset-float-long

Transformer Math Dataset (250,000,000 Samples Sharded)

High-precision synthetic mathematical expression dataset generated for training sequence-to-sequence math Transformers in JAX/Flax.

Dataset Structure

• —Total Samples: 250,000,000
• —Shard Format: JSONL sharded files (100,000 samples per shard)
• —Supported Operations: +, -, *, /, ^, sin, cos, tan, log, ln, exp, sqrt, abs
• —Expression Depth Range: Depth 4 to 6
• —Integer Operand Ratio: 20%

Data Fields

Each line in the .jsonl shard files is a JSON object with the following fields:

• —expr (str): Syntactically valid mathematical expression (e.g. "sin((3.5))+cos((1.2))")
• —val (str): Target evaluated numerical result formatted to precision (e.g. "0.6")
• —category (str): Operation-depth category bucket (e.g. "sin_d4")
• —ops_used (list[str]): List of mathematical functions/operators present in the expression (e.g. ["sin", "+", "cos"])

Usage Example

from datasets import load_dataset

dataset = load_dataset("saidurga001301/mathmetics-dataset-float-long", streaming=True)
for sample in dataset["train"]:
    print(sample["expr"], "->", sample["val"], "ops:", sample["ops_used"])
    break