AWS sagemaker documentation change
Summary
Updated data format requirements from 'Converse format' to 'format shown below' for SFT and DPO. Removed monitoring instructions for training jobs.
Security assessment
Changes are clarifications of documentation formatting and removal of redundant monitoring steps. No security vulnerabilities, incidents, or security features are mentioned or implied.
Diff
diff --git a/sagemaker/latest/dg/nova-fine-tune-1.md b/sagemaker/latest/dg/nova-fine-tune-1.md index c751a9252..954a6d01a 100644 --- a//sagemaker/latest/dg/nova-fine-tune-1.md +++ b//sagemaker/latest/dg/nova-fine-tune-1.md @@ -28 +28 @@ Preparing high-quality, properly formatted data is a critical first step in the -SFT data format requirements - For both full-rank SFT and LoRA SFT, data should follow the Converse format. For examples and constraints of this format, see [Preparing data for fine-tuning Understanding models](https://docs.aws.amazon.com//nova/latest/userguide/fine-tune-prepare-data-understanding.html). +SFT data format requirements - For both full-rank SFT and LoRA SFT, data should follow the format shown below. For examples and constraints of this format, see [Preparing data for fine-tuning Understanding models](https://docs.aws.amazon.com//nova/latest/userguide/fine-tune-prepare-data-understanding.html). @@ -34 +34 @@ SFT data validation - To validate your dataset format before submission, we reco -DPO data format requirements - For both DPO in full-rank and DPO with LoRA, data should follow the Converse format. The dataset also needs to be in the similar format as SFT except the last turn needs to have preference pairs. +DPO data format requirements - For both DPO in full-rank and DPO with LoRA, data should follow the format shown below. The dataset also needs to be in the similar format as SFT except the last turn needs to have preference pairs. @@ -432,6 +431,0 @@ The following sample notebook demonstrates how to run a training job. For additi - # 5. Monitor your training job - # To monitor your training job, you can either: - # * Go to your AWS console -> Amazon Sagemaker AI -> Training -> Training Jobs - # * run the following command - - # sm.describe_training_job(TrainingJobName="<complete training job name>")