bridge.data.sources.hf_adapters#

Schema adapters for already-loaded Hugging Face datasets.

Module Contents#

Functions#

_prompt_completion_example

_native_conversation_adapter

_squad_adapter

_gsm8k_adapter

_openmathinstruct2_adapter

_strip_intermediate_boxed

_openmathinstruct2_thinking_adapter

_rdr_adapter

_cord_v2_adapter

_medpix_adapter

_raven_adapter

_llava_video_adapter

_valor32k_avqa_adapter

_decode_audio

_audio_adapter

_cv17_adapter

prepare_hf_dataset_for_adapter

Apply dataset-level preparation required by a schema adapter.

validate_hf_dataset_adapter

Validate an optional registered Hugging Face schema adapter name.

adapt_hf_dataset

Normalize an already-loaded dataset into canonical SFT examples.

Data#

HFDatasetAdapter

_ADAPTERS

API#

bridge.data.sources.hf_adapters.HFDatasetAdapter#

None

bridge.data.sources.hf_adapters._prompt_completion_example(
prompt: str,
completion: str,
original_answers: list[str] | None = None,
) → dict[str, Any]#
bridge.data.sources.hf_adapters._native_conversation_adapter(
example: collections.abc.Mapping[str, Any],
kwargs: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._squad_adapter(
example: collections.abc.Mapping[str, Any],
_: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._gsm8k_adapter(
example: collections.abc.Mapping[str, Any],
_: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._openmathinstruct2_adapter(
example: collections.abc.Mapping[str, Any],
_: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._strip_intermediate_boxed(text: str) → str#
bridge.data.sources.hf_adapters._openmathinstruct2_thinking_adapter(
example: collections.abc.Mapping[str, Any],
_: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._rdr_adapter(
example: collections.abc.Mapping[str, Any],
kwargs: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._cord_v2_adapter(
example: collections.abc.Mapping[str, Any],
kwargs: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._medpix_adapter(
example: collections.abc.Mapping[str, Any],
_: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._raven_adapter(
example: collections.abc.Mapping[str, Any],
_: collections.abc.Mapping[str, Any],
) → dict[str, Any] | None#
bridge.data.sources.hf_adapters._llava_video_adapter(
example: collections.abc.Mapping[str, Any],
kwargs: collections.abc.Mapping[str, Any],
) → dict[str, Any] | None#
bridge.data.sources.hf_adapters._valor32k_avqa_adapter(
example: collections.abc.Mapping[str, Any],
kwargs: collections.abc.Mapping[str, Any],
) → dict[str, Any] | None#
bridge.data.sources.hf_adapters._decode_audio(audio: Any) → tuple[Any, int]#
bridge.data.sources.hf_adapters._audio_adapter(
example: collections.abc.Mapping[str, Any],
kwargs: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters._cv17_adapter(
example: collections.abc.Mapping[str, Any],
kwargs: collections.abc.Mapping[str, Any],
) → dict[str, Any]#
bridge.data.sources.hf_adapters.prepare_hf_dataset_for_adapter(
dataset: Any,
*,
adapter_name: str | None,
adapter_kwargs: collections.abc.Mapping[str, Any] | None = None,
) → Any#

Apply dataset-level preparation required by a schema adapter.

bridge.data.sources.hf_adapters._ADAPTERS: dict[str, bridge.data.sources.hf_adapters.HFDatasetAdapter]#

None

bridge.data.sources.hf_adapters.validate_hf_dataset_adapter(adapter_name: str | None) → None#

Validate an optional registered Hugging Face schema adapter name.

bridge.data.sources.hf_adapters.adapt_hf_dataset(
dataset: collections.abc.Iterable[collections.abc.Mapping[str, Any]],
*,
adapter_name: str | None,
adapter_kwargs: collections.abc.Mapping[str, Any] | None = None,
) → list[dict[str, Any]]#

Normalize an already-loaded dataset into canonical SFT examples.

Parameters:

  • dataset – Loaded Hugging Face rows.

  • adapter_name – Optional registered schema adapter. Native conversation rows require no adapter.

  • adapter_kwargs – Declarative adapter-specific options.

Returns:

Canonical text or multimodal conversation examples.