ztensor 1.2.3

Unified, zero-copy, and safe I/O for deep learning formats
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139

import os
import argparse

import config
from datagen import generate_tensor_dict
from runners import run_single_benchmark
from sweep import run_sweep
from plot import draw_plot


def main():
    parser = argparse.ArgumentParser(
        description="Benchmark tensor serialization formats (ztensor, safetensors, pytorch, pickle, hdf5, gguf)"
    )
    parser.add_argument(
        "mode",
        nargs="?",
        choices=["run", "sweep", "plot"],
        default="run",
        help="Run mode: single 'run', matrix 'sweep', or 'plot' to regenerate charts",
    )
    parser.add_argument(
        "--size",
        type=int,
        default=512,
        help="Data size in MB for single run (default: 512)",
    )
    parser.add_argument(
        "--runs",
        type=int,
        default=3,
        help="Number of timed runs per data point (default: 3)",
    )
    parser.add_argument(
        "--warmup", type=int, default=1, help="Number of warmup runs (default: 1)"
    )
    parser.add_argument(
        "--csv",
        type=str,
        default="bench_out/sweep_results.csv",
        help="CSV path for plot mode",
    )
    parser.add_argument(
        "--backend",
        type=str,
        choices=["numpy", "torch"],
        default="numpy",
        help="Tensor backend: 'numpy' (default) or 'torch'",
    )
    parser.add_argument(
        "--formats",
        type=str,
        nargs="+",
        default=None,
        help="Specific formats to benchmark (default: all native)",
    )
    parser.add_argument(
        "--data-style",
        type=str,
        choices=["random", "structured", "mixed_dtype"],
        default="random",
        help="Data generation style (default: random)",
    )
    parser.add_argument(
        "--scenario",
        type=str,
        choices=["full_load", "zero_copy", "selective", "fastest"],
        default="full_load",
        help="Benchmark scenario (default: full_load)",
    )
    parser.add_argument(
        "--dist",
        type=str,
        choices=["mixed", "small", "large", "llama-1b"],
        default="mixed",
        help="Tensor distribution (default: mixed)",
    )
    args = parser.parse_args()

    config.BACKEND = args.backend
    config.BENCH_RUNS = args.runs
    config.WARMUP_RUNS = args.warmup

    if args.mode == "sweep":
        run_sweep()
        return

    if args.mode == "plot":
        draw_plot(args.csv)
        return

    # Default: single run mode
    size_mb = args.size
    data_style = args.data_style
    scenario = args.scenario
    distribution = args.dist
    tensors = generate_tensor_dict(size_mb, distribution, data_style)
    formats = args.formats or config.NATIVE_FORMATS

    os.makedirs("bench_out", exist_ok=True)

    scenario_label = f", {scenario}" if scenario != "full_load" else ""
    style_label = f", {data_style}" if data_style != "random" else ""
    size_label = f"{size_mb}MB " if distribution != "llama-1b" else ""
    print(
        f"\n=== Single Run: {size_label}{distribution}{style_label}{scenario_label}, "
        f"{config.BENCH_RUNS} runs + {config.WARMUP_RUNS} warmup ===\n"
    )
    print(
        f"{'Format':<20} | {'Write (GB/s)':>12} | {'Read (s)':>12} | {'Read (GB/s)':>12} | {'File (MB)':>10}"
    )
    print("-" * 80)

    for fmt in formats:
        filepath = config.filepath_for_format(fmt)

        result = run_single_benchmark(fmt, tensors, filepath, scenario=scenario)

        if result is None:
            continue

        w_speed = result["write_median"]
        r_lat = result["read_median"]
        file_mb = result["file_size_gb"] * 1024
        data_size_gb = result["data_size_gb"]
        r_tp = data_size_gb / r_lat if r_lat > 0 else 0

        print(
            f"{fmt:<20} | {w_speed:>12.2f} | {r_lat:>12.4f} | {r_tp:>12.2f} | {file_mb:>10.1f}"
        )

        if os.path.exists(filepath):
            os.remove(filepath)

    print()


if __name__ == "__main__":
    main()