Kiến thức cần nắm:
Giải thích chi tiết:
Model Cards là tài liệu mô tả:
SageMaker Model Cards:
import boto3
sm_client = boto3.client('sagemaker')
response = sm_client.create_model_card(
ModelCardName='genai-chatbot-model-card',
Content=json.dumps({
"model_overview": {
"model_description": "Customer support chatbot using Claude 3",
"model_creator": "ML Team",
"problem_type": "Text Generation"
},
"intended_uses": {
"purpose_of_model": "Answer customer queries",
"intended_users": ["Customer support team"],
"out_of_scope_uses": ["Medical advice", "Legal advice"]
}
}),
ModelCardStatus='Draft'
)
Kiến thức cần nắm:
Giải thích chi tiết:
Data Lineage cho GenAI:
Source Documents (S3)
↓ [Glue Catalog tracks source]
Embedding Pipeline (Bedrock/Lambda)
↓ [Metadata tags preserved]
Vector Store (OpenSearch)
↓ [Source attribution in metadata]
RAG Response (Bedrock)
↓ [Citations include source docs]
User Response
CloudTrail logging cho Bedrock:
Kiến thức cần nắm:
Giải thích chi tiết:
AI Governance Framework:
| Component | Implementation |
|---|---|
| Policy definition | Organizational AI usage policies |
| Access control | IAM, SCPs, resource policies |
| Monitoring | CloudWatch, CloudTrail |
| Compliance checks | Lambda automated checks |
| Reporting | CloudWatch dashboards, model cards |
| Review process | Human review workflows |
Kiến thức cần nắm:
Giải thích chi tiết:
Monitoring Pipeline:
Model Invocations → CloudWatch Logs
↓
CloudWatch Metrics (token usage, error rates)
↓
CloudWatch Alarms (anomaly detection)
↓
SNS → Lambda (automated remediation)
Drift Detection: