AWS cli documentation change
Summary
Added support for incremental training data channels, S3 data distribution types, and training input mode parameters. Updated documentation for model versioning and data channel limits.
Security assessment
The changes introduce new training configuration options (incremental training, data distribution strategies, input modes) but contain no explicit security vulnerability fixes or security feature documentation. The S3 data distribution type parameter relates to performance/optimization rather than access control or encryption.
Diff
diff --git a/cli/latest/reference/cleanroomsml/create-trained-model.md b/cli/latest/reference/cleanroomsml/create-trained-model.md index 252913c78..c72ebbfaf 100644 --- a//cli/latest/reference/cleanroomsml/create-trained-model.md +++ b//cli/latest/reference/cleanroomsml/create-trained-model.md @@ -15 +15 @@ - * [AWS CLI 2.27.46 Command Reference](../../index.html) » + * [AWS CLI 2.27.49 Command Reference](../../index.html) » @@ -73,0 +74 @@ See also: [AWS API Documentation](https://docs.aws.amazon.com/goto/WebAPI/cleanr + [--incremental-training-data-channels <value>] @@ -74,0 +76 @@ See also: [AWS API Documentation](https://docs.aws.amazon.com/goto/WebAPI/cleanr + [--training-input-mode <value>] @@ -209,0 +212,43 @@ JSON Syntax: +`--incremental-training-data-channels` (list) + +> Specifies the incremental training data channels for the trained model. +> +> Incremental training allows you to create a new trained model with updates without retraining from scratch. You can specify up to one incremental training data channel that references a previously trained model and its version. +> +> Limit: Maximum of 20 channels total (including both `incrementalTrainingDataChannels` and `dataChannels` ). +> +> (structure) +> +>> Defines an incremental training data channel that references a previously trained model. Incremental training allows you to update an existing trained model with new data, building upon the knowledge from a base model rather than training from scratch. This can significantly reduce training time and computational costs while improving model performance with additional data. +>> +>> trainedModelArn -> (string) +>> +>>> The Amazon Resource Name (ARN) of the base trained model to use for incremental training. This model serves as the starting point for the incremental training process. +>> +>> versionIdentifier -> (string) +>> +>>> The version identifier of the base trained model to use for incremental training. If not specified, the latest version of the trained model is used. +>> +>> channelName -> (string) +>> +>>> The name of the incremental training data channel. This name is used to identify the channel during the training process and must be unique within the training job. + +Shorthand Syntax: + + + trainedModelArn=string,versionIdentifier=string,channelName=string ... + + +JSON Syntax: + + + [ + { + "trainedModelArn": "string", + "versionIdentifier": "string", + "channelName": "string" + } + ... + ] + + @@ -213,0 +259,2 @@ JSON Syntax: +> Limit: Maximum of 20 channels total (including both `dataChannels` and `incrementalTrainingDataChannels` ). +> @@ -224,0 +272,9 @@ JSON Syntax: +>> +>> s3DataDistributionType -> (string) +>> +>>> Specifies how the training data stored in Amazon S3 should be distributed to training instances. This parameter controls the data distribution strategy for the training job: +>>> +>>> * `FullyReplicated` \- The entire dataset is replicated on each training instance. This is suitable for smaller datasets and algorithms that require access to the complete dataset. +>>> * `ShardedByS3Key` \- The dataset is distributed across training instances based on Amazon S3 key names. This is suitable for larger datasets and distributed training scenarios where each instance processes a subset of the data. +>>> + @@ -229 +285 @@ Shorthand Syntax: - mlInputChannelArn=string,channelName=string ... + mlInputChannelArn=string,channelName=string,s3DataDistributionType=string ... @@ -238 +294,2 @@ JSON Syntax: - "channelName": "string" + "channelName": "string", + "s3DataDistributionType": "FullyReplicated"|"ShardedByS3Key" @@ -243,0 +301,18 @@ JSON Syntax: +`--training-input-mode` (string) + +> The input mode for accessing the training data. This parameter determines how the training data is made available to the training algorithm. Valid values are: +> +> * `File` \- The training data is downloaded to the training instance and made available as files. +> * `FastFile` \- The training data is streamed directly from Amazon S3 to the training algorithm, providing faster access for large datasets. +> * `Pipe` \- The training data is streamed to the training algorithm using named pipes, which can improve performance for certain algorithms. +> + +> +> Possible values: +> +> * `File` +> * `FastFile` +> * `Pipe` +> + + @@ -387,0 +463,6 @@ trainedModelArn -> (string) +versionIdentifier -> (string) + +> The unique version identifier assigned to the newly created trained model. This identifier can be used to reference this specific version of the trained model in subsequent operations such as inference jobs or incremental training. +> +> The initial version identifier for the base version of the trained model is “NULL”. + @@ -398 +479 @@ trainedModelArn -> (string) - * [AWS CLI 2.27.46 Command Reference](../../index.html) » + * [AWS CLI 2.27.49 Command Reference](../../index.html) »