lightrft.datasets.audio_alpaca¶

class lightrft.datasets.audio_alpaca.AudioAlpacaHandler[source]¶

Data Handler for Audio Alpaca dataset.

get_media_info(item: Dict[str, Any]) → Dict[str, Dict[str, Any]][source]¶

Extract audio bytes info for chosen and rejected audios.

Parameters:: item (Dict[str, Any]) – A data item from load_data
Returns:: Dict containing audio bytes for ‘chosen_audio’ and ‘rejected_audio’
Return type:: Dict[str, Dict[str, Any]]

Example:

info = handler.get_media_info(item)

load_data(path: str) → List[Dict[str, Any]][source]¶

Load and validate Audio Alpaca data from parquet files.

Parameters:: path (str) – Path to the directory containing parquet files
Returns:: A list of data items
Return type:: List[Dict[str, Any]]

Example:

data = handler.load_data("path/to/audio-alpaca")

parse_item(item: Dict[str, Any], media_content: Dict[str, Any], config: Dict[str, Any]) → Tuple[List[Dict], List[Dict], Dict][source]¶

Parse a single Audio Alpaca item into message pairs for ranking.

Parameters:

item (Dict[str, Any]) – Raw data item from Audio Alpaca dataset.
media_content (Dict[str, Any]) – Loaded audio content with ‘chosen_audio’ and ‘rejected_audio’ keys.
config (Dict[str, Any]) – Configuration dict with task_instruction template.

Returns:

A tuple of (messages0, messages1, metadata)

Return type:

Tuple[List[Dict], List[Dict], Dict]

Example:

msg0, msg1, other = handler.parse_item(item, media_content, config)