AWS bedrock documentation change
Summary
Updated documentation to include distillation as a customization option, added details about Amazon Nova model requirements, and restructured the model support table to include distillation capabilities.
Security assessment
The changes focus on adding and clarifying model customization options, particularly around distillation and Amazon Nova models. There is no mention of security vulnerabilities, incidents, or security-specific features in the changes.
Diff
diff --git a/bedrock/latest/userguide/model-customization-prepare.md index 443c8b9a7..978be3ef9 100644 --- a/bedrock/latest/userguide/model-customization-prepare.md +++ b/bedrock/latest/userguide/model-customization-prepare.md @@ -5 +5 @@ -Model support for fine-tuning and continued pre-training data formatPrepare training and validation datasets for your custom model +Model support for distillation, fine-tuning and continued pre-training Prepare datasets for your custom model @@ -11 +11,3 @@ Before you can begin a model customization job, you need to minimally prepare a - * The type of customization job (fine-tuning or Continued Pre-training). + * The type of customization job (Distillation, Fine-tuning, or Continued Pre-training). + +If you are planning on using Distillation, see [Prerequisites for Amazon Bedrock Model Distillation](./prequisites-model-distillation.html) for more information. @@ -18 +20 @@ Before you can begin a model customization job, you need to minimally prepare a -## Model support for fine-tuning and continued pre-training data format +## Model support for distillation, fine-tuning and continued pre-training @@ -20 +22 @@ Before you can begin a model customization job, you need to minimally prepare a -The following table shows details of the fine-tuning and continued pre-training data format supported for each respective model: +The following table shows the input and output modalities for the distillation, fine-tuning and continued pre-training supported for each respective model: @@ -22,15 +24,15 @@ The following table shows details of the fine-tuning and continued pre-training -Model name | Fine-tuning:Text-to-text | Fine-tuning: Text-to-image & Image-to-embeddings | Fine-tuning: Text+Image-to-Text & Text+Video-to-Text | Continued Pre-training:Text-to-text | Fine-tuning: Single-turn messaging | Fine-tuning: Multi-turn messaging ----|---|---|---|---|---|--- -Amazon Nova Pro | Yes | Yes | Yes | No | Yes | Yes -Amazon Nova Lite | Yes | Yes | Yes | No | Yes | Yes -Amazon Nova Micro | Yes | No | No | No | Yes | Yes -Amazon Titan Text G1 - Express | Yes | No | No | Yes | No | No -Amazon Titan Text G1 - Lite | Yes | No | No | Yes | No | No -Amazon Titan Text Premier | Yes | No | No | No | No | No -Amazon Titan Image Generator G1 V1 | Yes | Yes | No | No | No | No -Amazon Titan Multimodal Embeddings G1 G1 | Yes | Yes | No | No | No | No -Anthropic Claude 3 Haiku | No | No | No | No | Yes | Yes -Cohere Command | Yes | No | No | No | No | No -Cohere Command Light | Yes | No | No | No | No | No -Meta Llama 2 13B | Yes | No | No | No | No | No -Meta Llama 2 70B | Yes | No | No | No | No | No +Model name | Distillation: Text-to-text | Fine-tuning:Text-to-text | Fine-tuning: Text-to-image & Image-to-embeddings | Fine-tuning: Text+Image-to-Text & Text+Video-to-Text | Continued Pre-training:Text-to-text | Fine-tuning: Single-turn messaging | Fine-tuning: Multi-turn messaging +---|---|---|---|---|---|---|--- +Amazon Nova Pro | Yes | Yes | Yes | Yes | No | Yes | Yes +Amazon Nova Lite | Yes | Yes | Yes | Yes | No | Yes | Yes +Amazon Nova Micro | Yes | Yes | No | No | No | Yes | Yes +Amazon Titan Text G1 - Express | No | Yes | No | No | Yes | No | No +Amazon Titan Text G1 - Lite | No | Yes | No | No | Yes | No | No +Amazon Titan Text Premier | No | Yes | No | No | No | No | No +Amazon Titan Image Generator G1 V1 | No | Yes | Yes | No | No | No | No +Amazon Titan Multimodal Embeddings G1 G1 | No | Yes | Yes | No | No | No | No +Anthropic Claude 3 Haiku | Yes | No | No | No | No | Yes | Yes +Cohere Command | No | Yes | No | No | No | No | No +Cohere Command Light | No | Yes | No | No | No | No | No +Meta Llama 2 13B | No | Yes | No | No | No | No | No +Meta Llama 2 70B | No | Yes | No | No | No | No | No @@ -40 +42 @@ To see the default quotas that apply for training and validation datasets used f -## Prepare training and validation datasets for your custom model +## Prepare datasets for your custom model @@ -49 +51 @@ Fine-tuning: Text-to-text -For text-to-text models, prepare a training and optional validation dataset. Each JSON object is a sample containing both a `prompt` and `completion` field. Use 6 characters per token as an approximation for the number of tokens. The format is as follows. +For Titan, Cohere, and Llama text-to-text models, prepare a training and optional validation dataset. Each JSON object is a sample containing both a `prompt` and `completion` field. Use 6 characters per token as an approximation for the number of tokens. The format is as follows: @@ -62 +64,32 @@ The following is an example item for a question-answer task: -Amazon Nova models have different fine-tuning requirements. To fine-tune these models, follow the instructions at [Guidelines for preparing your data for Amazon Nova](https://docs.aws.amazon.com/nova/latest/userguide/customize-fine-tune-prepare.html). +Amazon Nova models expect the training data in a different JSON structure. These models use a system prompt along with `"role": "user"` and `"role": "assistant"` messages to fine tune the model. The format is as follows: + + + // train.jsonl + { + "schemaVersion": "bedrock-conversation-2024", + "system": [ + { + "text": "You are a digital assistant with a friendly personality" + } + ], + "messages": [ + { + "role": "user", + "content": [ + { + "text": "What is the capital of Mars?" + } + ] + }, + { + "role": "assistant", + "content": [ + { + "text": "Mars does not have a capital. Perhaps it will one day." + } + ] + } + ] + } + +For more information, follow the instructions at [Guidelines for preparing your data for Amazon Nova](https://docs.aws.amazon.com/nova/latest/userguide/customize-fine-tune-prepare.html). @@ -66,0 +100,4 @@ Fine-tuning: Text-to-image & Image-to-embeddings +###### Note + +Amazon Nova models have different fine-tuning requirements. To fine-tune these models, follow the instructions at [Guidelines for preparing your data for Amazon Nova](https://docs.aws.amazon.com/nova/latest/userguide/customize-fine-tune-prepare.html). + @@ -101,0 +139,4 @@ Fine-tuning: Single-turn messaging +###### Note + +Amazon Nova models have different fine-tuning requirements. To fine-tune these models, follow the instructions at [Guidelines for preparing your data for Amazon Nova](https://docs.aws.amazon.com/nova/latest/userguide/customize-fine-tune-prepare.html). + @@ -138,0 +180,4 @@ Fine-tuning: Multi-turn messaging +###### Note + +Amazon Nova models have different fine-tuning requirements. To fine-tune these models, follow the instructions at [Guidelines for preparing your data for Amazon Nova](https://docs.aws.amazon.com/nova/latest/userguide/customize-fine-tune-prepare.html). + @@ -177,0 +223,4 @@ Distillation +###### Note + +Amazon Nova models have different requirements. To distill these models, follow the instructions at [Distilling Amazon Nova models](https://docs.aws.amazon.com/nova/latest/userguide/customize-distill.html). +