example_data/info.txt

=== Training Example Data Info ===

This example is saved from the raw training dataset (before data_processor).
Data is saved in unnormalized form to match testing phase format.

Sample keys:
  image: shape=torch.Size([2, 3, 256, 256]), dtype=torch.uint8
  state: shape=torch.Size([7]), dtype=torch.float32
  action: shape=torch.Size([16, 7]), dtype=torch.float32
  is_pad: shape=torch.Size([16]), dtype=torch.bool
  raw_lang: str
  reasoning: dict
  timestamp: int
  episode_id: int
  action_raw: shape=torch.Size([16, 7]), dtype=torch.float32
  dataset_id: str

Files saved:
  - camera_{i}.png: raw images from dataset (unnormalized, original resolution)
  - state_raw.csv: raw state values (unnormalized)
  - action_raw.csv: raw action values (unnormalized)
  - raw_lang.txt: language instruction (if available)
  - reasoning.json: reasoning data (if available)
  - info.txt: this file

Note: These raw values can be directly compared with testing phase examples.