mstar.model.whisper.config

Config for Whisper encoder-decoder ASR models (large-v3 and friends).

Values are read from the HF checkpoint’s config.json + generation_config.json so the same class serves any Whisper size.

Classes

WhisperModelConfig(d_model, decoder_layers, ...)

class mstar.model.whisper.config.WhisperModelConfig(d_model: 'int' = 1280, decoder_layers: 'int' = 32, decoder_attention_heads: 'int' = 20, decoder_ffn_dim: 'int' = 5120, encoder_layers: 'int' = 32, num_mel_bins: 'int' = 128, vocab_size: 'int' = 51866, max_target_positions: 'int' = 448, max_source_positions: 'int' = 1500, activation_function: 'str' = 'gelu', scale_embedding: 'bool' = False, decoder_start_token_id: 'int' = 50258, eos_token_id: 'int' = 50257, no_timestamps_token_id: 'int' = 50364, lang_to_id: 'dict[str, int]'=<factory>, task_to_id: 'dict[str, int]'=<factory>, suppress_tokens: 'list[int]' = <factory>, begin_suppress_tokens: 'list[int]' = <factory>)

Bases: object

Parameters:

• d_model (int)

• decoder_layers (int)

• decoder_attention_heads (int)

• decoder_ffn_dim (int)

• encoder_layers (int)

• num_mel_bins (int)

• vocab_size (int)

• max_target_positions (int)

• max_source_positions (int)

• activation_function (str)

• scale_embedding (bool)

• decoder_start_token_id (int)

• eos_token_id (int)

• no_timestamps_token_id (int)

• lang_to_id (dict[str, int])

• task_to_id (dict[str, int])

• suppress_tokens (list[int])

• begin_suppress_tokens (list[int])

activation_function: str = 'gelu'

begin_suppress_tokens: list[int]

d_model: int = 1280

decoder_attention_heads: int = 20

decoder_ffn_dim: int = 5120

decoder_layers: int = 32

decoder_prompt_ids(language='en', task='transcribe')

<|startoftranscript|><|{lang}|><|{task}|><|notimestamps|>.

Parameters:

• language (str)

• task (str)

Return type:

list[int]

decoder_start_token_id: int = 50258

encoder_layers: int = 32

eos_token_id: int = 50257

classmethod from_pretrained(local_dir)

Parameters:

local_dir (str | Path)

Return type:

WhisperModelConfig

property head_dim: int

lang_to_id: dict[str, int]

max_source_positions: int = 1500

max_target_positions: int = 448

no_timestamps_token_id: int = 50364

num_mel_bins: int = 128

scale_embedding: bool = False

suppress_tokens: list[int]

task_to_id: dict[str, int]

vocab_size: int = 51866