Description
Elastic.Clients.Elasticsearch version: 8.14.6
Elasticsearch version: 8.14.3
.NET runtime version: 8.0
Operating system version: Windows
Description of the problem including expected versus actual behavior:
I'm using the MachineLearning.GetTrainedModelsStatsAsync call to get the stats of a trained model, this request always results in System.Text.Json.JsonException: The JSON value could not be converted to System.Int32. Path: $.trained_model_stats[0].model_size_stats.required_native_memory_bytes. Invoking the route manually http://localhost:9200/_ml/trained_models/intfloat__e5-small-v2/_stats works.
Upon inspection of the Model class the response is deserialized into, I found the required_native_memory_bytes is defined as an int rather than ByteSize as the other fields of the same type:
namespace Elastic.Clients.Elasticsearch.MachineLearning;
public sealed partial class TrainedModelSizeStats
{
/// <summary>
/// <para>The size of the model in bytes.</para>
/// </summary>
[JsonInclude, JsonPropertyName("model_size_bytes")]
public Elastic.Clients.Elasticsearch.ByteSize ModelSizeBytes { get; init; }
/// <summary>
/// <para>The amount of memory required to load the model in bytes.</para>
/// </summary>
[JsonInclude, JsonPropertyName("required_native_memory_bytes")]
public int RequiredNativeMemoryBytes { get; init; }
}
The required_native_memory_bytes is much bigger than an int.
Steps to reproduce:
- Deploy an ML model
- Use the .NET elastic client to invoke the model like
await client.MachineLearning.GetTrainedModelsStatsAsync(new Ids("intfloat__e5-small-v2"), cancellationToken); - The code throws a deserialization exception
Expected behavior
The TrainedModelSizeStats class should type the RequiredNativeMemoryBytes field as Elastic.Clients.Elasticsearch.ByteSize, and the GetTrainedModelsStatsAsync call should succeed.