Method ExportAsync

Namespace: LMKit.Finetuning.Export

Assembly: LM-Kit.NET.dll

ExportAsync(IEnumerable<ChatTrainingSample>, string, DatasetBuilderOptions, IProgress<ExportProgress>, CancellationToken)

Exports chat training samples to a ShareGPT-style JSON dataset using streaming I/O (asynchronous).

public static Task<ExportResult> ExportAsync(IEnumerable<ChatTrainingSample> samples, string outputFilePath, DatasetBuilderOptions options = null, IProgress<ExportProgress> progress = null, CancellationToken cancellationToken = default)

Parameters

samples IEnumerable<ChatTrainingSample>: The chat training samples to export. Must not be null.
outputFilePath string: Destination path for the ShareGPT JSON file (for example, ./data/sharegpt.json). Parent directories are created if needed.
options DatasetBuilderOptions: Optional exporter configuration. When null, sensible defaults are used.
progress IProgress<ExportProgress>: Optional progress reporter. Receives periodic ExportProgress updates after each sample is written.
cancellationToken CancellationToken: A token observed for cooperative cancellation.

Returns

Task<ExportResult>: An ExportResult describing the outcome, including counts and resolved output paths.

Remarks

The exporter writes one JSON object per sample directly to the output stream, avoiding large in-memory buffers for big datasets. Images embedded in the samples are emitted to a sibling folder.

The output JSON schema:

[
  {
    "id": "sample001",
    "images": ["images/sample001_1.png", "..."],
    "messages": [
      { "role": "system", "content": "..." },
      { "role": "user",   "content": "..." },
      { "role": "assistant", "content": "..." }
    ]
  },
  ...
]

Exceptions