def build_encoder_layer(
self, args, positional_embedding: Optional[RelativePositionalEmbedding] = None
):
return super().build_encoder_layer(
SpeechTransformerConfig.from_namespace(args),
positional_embedding=positional_embedding,
)
def add_args(parser):
"""Add model-specific arguments to the parser."""
# we want to build the args recursively in this case.
gen_parser_from_dataclass(parser,
SpeechTransformerConfig(),
delete_default=False,
with_prefix="")
def add_args(cls, parser):
"""Add model-specific arguments to the parser."""
# we want to build the args recursively in this case.
# do not set defaults so that settings defaults from various architectures still works
gen_parser_from_dataclass(parser,
SpeechTransformerConfig(),
delete_default=True,
with_prefix="")
def build_decoder(cls,
args,
tgt_dict,
embed_tokens,
scheduled_sampling_rate_scheduler=None):
return super().build_decoder(
SpeechTransformerConfig.from_namespace(args),
tgt_dict,
embed_tokens,
scheduled_sampling_rate_scheduler=scheduled_sampling_rate_scheduler,
)
def build_encoder(cls,
args,
conv_layers_before=None,
input_size=83,
transformer_context=None):
return super().build_encoder(
SpeechTransformerConfig.from_namespace(args),
conv_layers_before=conv_layers_before,
input_size=input_size,
transformer_context=transformer_context,
)
def __init__(
self,
args,
return_fc=False,
pre_encoder=None,
input_size=83,
transformer_context=None,
):
self.args = args
super().__init__(
SpeechTransformerConfig.from_namespace(args),
return_fc=return_fc,
pre_encoder=pre_encoder,
input_size=input_size,
transformer_context=transformer_context,
)
def __init__(
self,
args,
dictionary,
embed_tokens,
no_encoder_attn=False,
output_projection=None,
scheduled_sampling_rate_scheduler=None,
):
self.args = args
super().__init__(
SpeechTransformerConfig.from_namespace(args),
dictionary,
embed_tokens,
no_encoder_attn=no_encoder_attn,
output_projection=output_projection,
scheduled_sampling_rate_scheduler=scheduled_sampling_rate_scheduler,
)
def build_model(cls, args, task):
"""Build a new model instance."""
# make sure all arguments are present in older models
base_architecture(args)
if args.encoder_layers_to_keep:
args.encoder_layers = len(args.encoder_layers_to_keep.split(","))
if args.decoder_layers_to_keep:
args.decoder_layers = len(args.decoder_layers_to_keep.split(","))
if getattr(args, "max_source_positions", None) is None:
args.max_source_positions = DEFAULT_MAX_SOURCE_POSITIONS
if getattr(args, "max_target_positions", None) is None:
args.max_target_positions = DEFAULT_MAX_TARGET_POSITIONS
if getattr(args, "offload_activations", False):
args.checkpoint_activations = True # offloading implies checkpointing
args.min_params_to_wrap = getattr(args, "min_params_to_wrap",
DEFAULT_MIN_PARAMS_TO_WRAP)
cfg = SpeechTransformerConfig.from_namespace(args)
return super().build_model(cfg, task)
def build_output_projection(self, args, dictionary, embed_tokens):
super().build_output_projection(
SpeechTransformerConfig.from_namespace(args), dictionary,
embed_tokens)
def build_embedding(cls, args, dictionary, embed_dim, path=None):
return super().build_embedding(
SpeechTransformerConfig.from_namespace(args), dictionary,
embed_dim, path)
def __init__(self, args, encoder, decoder):
cfg = SpeechTransformerConfig.from_namespace(args)
super().__init__(cfg, encoder, decoder)
self.args = args