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

# ztensor Benchmarks

This directory contains the scripts used to benchmark `ztensor` against other formats and to generate the performance charts found in the main `README.md` and the website.

## Prerequisites

Install the required Python dependencies:

```bash
pip install -r requirements.txt
```

## Reproducing README/Website Figures

The figures in the [README](../README.md) and [website](../website/) are generated by a standalone script that ensures consistent styling and output:

```bash
python scripts/generate_readme_charts.py
```

This reads the hardcoded benchmark data in the script and produces SVGs in `website/static/charts/`.

## Running the Benchmarks

To run the actual benchmarks and generate fresh performance data on your machine:

```bash
python bench.py
```

### Modes

- **Single Run** (default): Runs a quick benchmark with default settings.
  ```bash
  python bench.py --size 512 --runs 5
  ```

- **Full Sweep**: Runs the complete suite of benchmarks (all formats, sizes, and distributions) used to gather the data for the README.
  ```bash
  python bench.py sweep
  ```
  This will:
  1. Generate synthetic data (random and structured).
  2. Benchmark write and read performance for all supported formats.
  3. Save results to `bench_out/sweep_results.csv`.
  4. Generate a set of analysis plots in `bench_out/plots/`.

### Updating the Charts

If you run a sweep and want to update the official README/website charts with your new data:
1. Run `python bench.py sweep` to get new numbers.
2. Update the `CROSS_FORMAT_DATA` and `WRITE_DIST_DATA` lists in `scripts/generate_readme_charts.py` with the median values from your results.
3. Run `python scripts/generate_readme_charts.py` to regenerate the SVGs.