SageMaker Training <=> S3 bucket (Official)
.
├── opt
│   └── ml
│       ├── input
│       │   ├── data
│       │   │   ├── channel 1 <-----------r----------- s3://bucket-data1
│       │   │   └── channel N <-----------r----------- s3://bucket-dataN
│       │   └── config
│       │       ├── hyperparameters.json
│       │       ├── inputdataconfig.json
│       │       └── resourceconfig.json
│       ├── output
│       │   ├── data          -----------w-----------> s3://output-path/<job-name>-<timestamp>/output/output.tar.gz
│       │   └── failure
│       ├── model             -----------w-----------> s3://output-path/<job-name>-<timestamp>/output/model.tar.gz
│       ├── checkpoints       <-----------r/w-----------> s3://checkpoint-dest
│       └── code
└── tmp

estimator = Estimator(
  checkpoint_s3_uri='s3://checkpoint-dest',
  output_path='s3://output-path',
  base_job_name='job-name',
  input_mode='File',
)

estimator.fit(inputs={
  'channel1': 's3://bucket-data1',
  ...
  'channelN': 's3://bucket-dataN',})


More info refer to 
- https://docs.aws.amazon.com/sagemaker/latest/dg/model-train-storage.html#model-train-storage-env-var-summary
- https://docs.aws.amazon.com/sagemaker/latest/dg/your-algorithms-training-algo-output.html

======================================================================================================================================

SageMaker Training <=> S3 bucket (Current)
.
└── opt
    └── ml
        └── stable-diffusion-webui
            ├── <dataset>             <-----------r----------- s3://aigc-bucket/dataset & s3://aigc-bucket/Stable-diffusion/train/<model-name>/<request-id>/input
            ├── extensions
            ├── ...
            └── model
                ├── dreambooth        <-----------r----------- s3://aigc-bucket/Stable-diffusion/model/<model-name>/<request-id>/output
                │   └── model-name
                │       └── db_config.json
                ├── stable-diffusion  -----------w-----------> s3://aigc-bucket/Stable-diffusion/train/<model-name>/<request-id>/output
                │   └── model-name
                └── Lora

SageMaker Inference <=> S3 bucket
.
└── opt
    └── ml
        └── model                     <-----------r----------- s3://aigc-bucket/checkpoint/custom & s3://aigc-bucket/<model-type>/checkpoint/<model-name>/<request-id> & s3://aigc-bucket/Stable-diffusion/train/<model-name>/<request-id>/output

Create Model <=> S3 bucket
.
└── opt
    └── ml
        └── model                     <-----------r----------- s3://aigc-bucket/checkpoint/custom & s3://aigc-bucket/Stable-diffusion/checkpoint/<model-name>/<request-id> & s3://aigc-bucket/Stable-diffusion/train/<model-name>/<request-id>/output
                                      -----------w-----------> s3://aigc-bucket/Stable-diffusion/model/<model-name>/<request-id>/output

Mapping Relationship:
- output:model = 1:1
- output:<job-name>-<timestamp> = 1:1