# --- MAIN EXPERIMENT SETTINGS
name: blog-benchmarks-all
description: 'Benchmarks -- Blog: Choose the best data source for your Amazon SageMaker training job'
role: SageMakerRoleBenchmark
region: us-east-1
bucket: null                # 'null' for Default Bucket
output_prefix: experiments
parallelism: 6              # 0 is sync execution, >=1 is async execution
repeat: 3
disable_profiler: True
order: interleave           # 'default | 'interleave' | 'random' - order to run the 'repeat'ed trials (not to create hot-spots)
clean: False                # 'True' - delete all previous trials before start



# --- DATASET DEFINITIONS
datasets: 'datasets/datasets_caltech.yml'


# --- FSx FILESYSTEM SETUP (is required only if 'fsx' input mode is used)
fsx:
  fsx_file_system_id: fs-XXXXXXXXXXXXXXXXX   # <--- specify your FSX filesystem ID
  fsx_file_system_directory_path: /fsx
  security_group_ids:
      - sg-XXXXXXXXXXXXXXXXX                 # <--- specify your Security Group IDs
  subnets:
      - subnet-XXXXXXXXXXXXXXXXX             # <--- specify your Subnet IDs


# --- TRIAL DEFINITIONS
# Default values for trials (if not overriden by specific trial)
base_trial:
  script: script-model.py
  source_dir: scripts
  framework: tf
  framework_version: '2.4.1'
  py_version: py37
  volume_size: 500
  instance_type: ml.p3.2xlarge
  instance_count: 1
  max_run: 36000
  hyperparameters:
    prefetch_first: False
    prefetch_size: -1
    prefetch_to_gpu: False
    num_parallel_calls: -1    # ('None' or int) goes as arg to tf.map (and tf.batch) 
    num_parallel_reads: -1    # ('None' or int) goes as arg to TFRecordDataset
    private_pool: 8           # ('None' or int) private workers thread pool
    epochs: 1
    backbone_model: resnet50
    batch_size: 64
    input_dim: 224
    cache: None               # None, 'mem', or <path> for on-disk cache (e.g. 'tmp/cache')


# Trial definitions
trials:

# ==== JPG-File: 1x ====
  - inputs:
      train:
        dataset: Caltech-jpg-1x
        input_mode: file
    
  - inputs:
      train:
        dataset: Caltech-jpg-1x
        input_mode: fsx
    
  - inputs:
      train:
        dataset: Caltech-jpg-1x
        input_mode: ffm

# ==== JPG-File: 50x ====
  - inputs:
      train:
        dataset: Caltech-jpg-50x
        input_mode: file
    
  - inputs:
      train:
        dataset: Caltech-jpg-50x
        input_mode: fsx
    
  - inputs:
      train:
        dataset: Caltech-jpg-50x
        input_mode: ffm
    
# ==== TFR-JPG-File: 1x ====
  - inputs:
      train:
        dataset: Caltech-tfr-jpg-1x
        input_mode: file
    
  - inputs:
      train:
        dataset: Caltech-tfr-jpg-1x
        input_mode: fsx
    
  - inputs:
      train:
        dataset: Caltech-tfr-jpg-1x
        input_mode: ffm

# ==== TFR-JPG-File: 50x ====
  - inputs:
      train:
        dataset: Caltech-tfr-jpg-50x
        input_mode: file
    
  - inputs:
      train:
        dataset: Caltech-tfr-jpg-50x
        input_mode: fsx
    
  - inputs:
      train:
        dataset: Caltech-tfr-jpg-50x
        input_mode: ffm