THULab/gsmm_ai

GSMM AI Green-Software Measurement Traces (TsFile)

This dataset is the Apache TsFile conversion of ykgmfq/gsmm_ai, released with the GSMM AI paper. It records power, voltage, current, and experiment bookkeeping for embedded-AI measurements.

Modalities: Time-series.

Overview

• Source revision: 1f437393d9cee80f826341e245e290fe29e18a4a
• Source file: data.parquet
• Rows/observations: 371,800
• (System, Model) combinations: six (0/0, 0/1, 1/0, 1/1, 2/0, 2/1)
• Epoch values: 0-4,499 (4,500 epochs)
• Chunk sizes: 25, 250, and 500
• Source license: apache-2.0

The Start Time [ns] and End Time [ns] columns are an elapsed nanosecond clock, not Unix timestamps. The source contains no wall-clock timezone.

TsFile schema

The converted table is gsmm_ai in gsmm_ai.tsfile (371,800 rows; 3,908,396 bytes).

Column	Role	Type	Source / meaning
Time	TIME	INT64 (ms)	Floor division of Start Time [ns] by 1,000,000; elapsed time
system	TAG	STRING	Source numeric System code, preserved as text
model	TAG	STRING	Source numeric Model code, preserved as text
start_time_ns	FIELD	INT64	Original Start Time [ns]
end_time_ns	FIELD	INT64	Original End Time [ns]
power_w	FIELD	FLOAT	Source Power [W]
voltage_v	FIELD	FLOAT	Source Voltage [V]
current_a	FIELD	FLOAT	Source Current [A]
epoch	FIELD	INT64	Source epoch number
chunk_size	FIELD	INT64	Source chunk-size value

Conversion notes

• Unit-labelled source names are normalized to safe names (Power [W] to power_w, etc.). The original start/end nanosecond values are retained as fields for auditability.
• Time uses integer milliseconds from the elapsed nanosecond start clock; it must not be interpreted as a Unix date.
• System and Model values are string TAGs so each combination is a device dimension. Numeric codes are not decoded or remapped.
• All 371,800 source rows and all nine source columns are retained; no null values or measurements are imputed. The only added column is Time, and the only naming changes are the safe field names shown above.

Read example

from tsfile import TsFileReader

path = "gsmm_ai.tsfile"
with TsFileReader(path) as reader:
    with reader.query_table(
        "gsmm_ai",
        ["power_w", "voltage_v", "current_a"],
        batch_size=4096,
    ) as result:
        batch = result.read_arrow_batch()
        if batch is not None:
            print(batch.to_pandas().head())

Source & license

• Original dataset: https://huggingface.co/datasets/ykgmfq/gsmm_ai
• Author / publisher: ykgmfq
• Paper: https://doi.org/10.57967/hf/3477
• License: Apache-2.0

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("gsmm_ai.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())