synthetic_vlm

Small deterministic image/text dataset for global-KD integration runs.

Functions

make_synthetic_llm_dataset

make_synthetic_vlm_dataset

make_synthetic_llm_dataset(tokenizer, count=16, seq_length=128, split=None, **_)
Parameters:
  • count (int)

  • seq_length (int)

  • split (str | None)

  • _ (object)

Return type:

list[dict]

make_synthetic_vlm_dataset(count=16, split=None, **_)
Parameters:
  • count (int)

  • split (str | None)

  • _ (object)

Return type:

list[dict]