def build_model(cls, args, task):
"""Build a new model instance."""
# make sure all arguments are present in older models
base_lm_architecture(args)
if getattr(args, "max_target_positions", None) is not None:
max_target_positions = args.max_target_positions
else:
max_target_positions = getattr(args, "tokens_per_sample",
DEFAULT_MAX_TARGET_POSITIONS)
def load_pretrained_embedding_from_file(embed_path, dictionary,
embed_dim):
num_embeddings = len(dictionary)
padding_idx = dictionary.pad()
embed_tokens = Embedding(num_embeddings, embed_dim, padding_idx)
embed_dict = utils.parse_embedding(embed_path)
utils.print_embed_overlap(embed_dict, dictionary)
return utils.load_embedding(embed_dict, dictionary, embed_tokens)
if args.is_wordlm and hasattr(task, "word_dictionary"):
dictionary = task.word_dictionary
elif isinstance(task, SpeechRecognitionEspressoTask):
dictionary = task.target_dictionary
else:
dictionary = task.source_dictionary
# separate decoder input embeddings
pretrained_decoder_embed = None
if args.decoder_embed_path:
pretrained_decoder_embed = load_pretrained_embedding_from_file(
args.decoder_embed_path, dictionary, args.decoder_embed_dim)
# one last double check of parameter combinations
if args.share_embed and (args.decoder_embed_dim !=
args.decoder_out_embed_dim):
raise ValueError(
"--share-embed requires "
"--decoder-embed-dim to match --decoder-out-embed-dim")
if args.decoder_freeze_embed:
pretrained_decoder_embed.weight.requires_grad = False
decoder = SpeechLSTMDecoder(
dictionary=dictionary,
embed_dim=args.decoder_embed_dim,
hidden_size=args.decoder_hidden_size,
out_embed_dim=args.decoder_out_embed_dim,
num_layers=args.decoder_layers,
dropout_in=args.decoder_dropout_in,
dropout_out=args.decoder_dropout_out,
attn_type=None,
encoder_output_units=0,
pretrained_embed=pretrained_decoder_embed,
share_input_output_embed=args.share_embed,
adaptive_softmax_cutoff=(utils.eval_str_list(
args.adaptive_softmax_cutoff, type=int) if args.criterion
== "adaptive_loss" else None),
max_target_positions=max_target_positions,
)
return cls(decoder, args)
def build_decoder(cls, cfg, tgt_dict, embed_tokens):
return SpeechLSTMDecoder(
tgt_dict,
embed_dim=cfg.decoder.embed_dim,
hidden_size=cfg.decoder.hidden_size,
out_embed_dim=cfg.decoder.hidden_size,
num_layers=cfg.decoder.layers,
dropout_in=cfg.decoder.dropout_in,
dropout_out=cfg.decoder.dropout_out,
residual=cfg.decoder.residual,
pretrained_embed=embed_tokens,
share_input_output_embed=True, # disallow fc_out in decoder
max_target_positions=DEFAULT_MAX_TARGET_POSITIONS,
)