mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-04 14:44:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Minor fixes.

Browse Source
This commit is contained in:
Fangjun Kuang 2022-04-28 06:51:11 +08:00
parent e9f0975868
commit af20922320
2 changed files with 42 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

58
egs/librispeech/ASR/pruned_transducer_stateless3/decode.py
View File

@ -99,27 +99,28 @@ def get_parser():
"--epoch",
type=int,
default=28,
help="It specifies the checkpoint to use for decoding."
"Note: Epoch counts from 0.",
help="""It specifies the checkpoint to use for decoding.
Note: Epoch counts from 0.
You can specify --avg to use more checkpoints for model averaging.""",
)
parser.add_argument(
"--iter",
type=int,
default=0,
help="""If positive, --epoch is ignored and it
will use the checkpoint exp_dir/checkpoint-iter.pt.
You can specify --avg to use more checkpoints for model averaging.
""",
)
parser.add_argument(
"--avg",
type=int,
default=15,
help="Number of checkpoints to average. Automatically select "
"consecutive checkpoints before the checkpoint specified by "
"'--epoch'. ",
)
parser.add_argument(
"--avg-last-n",
type=int,
default=0,
help="""If positive, --epoch and --avg are ignored and it
will use the last n checkpoints exp_dir/checkpoint-xxx.pt
where xxx is the number of processed batches while
saving that checkpoint.
""",
"'--epoch' and '--iter'",
)
parser.add_argument(
@ -454,13 +455,19 @@ def main():
)
params.res_dir = params.exp_dir / params.decoding_method
params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
if params.iter > 0:
params.suffix = f"iter-{params.iter}-avg-{params.avg}"
else:
params.suffix = f"epoch-{params.epoch}-avg-{params.avg}"
if "fast_beam_search" in params.decoding_method:
params.suffix += f"-beam-{params.beam}"
params.suffix += f"-max-contexts-{params.max_contexts}"
params.suffix += f"-max-states-{params.max_states}"
elif "beam_search" in params.decoding_method:
params.suffix += f"-beam-{params.beam_size}"
params.suffix += (
f"-{params.decoding_method}-beam-size-{params.beam_size}"
)
else:
params.suffix += f"-context-{params.context_size}"
params.suffix += f"-max-sym-per-frame-{params.max_sym_per_frame}"
@ -477,8 +484,9 @@ def main():
sp = spm.SentencePieceProcessor()
sp.load(params.bpe_model)
# <blk> is defined in local/train_bpe_model.py
# <blk> and <unk> is defined in local/train_bpe_model.py
params.blank_id = sp.piece_to_id("<blk>")
params.unk_id = sp.piece_to_id("<unk>")
params.vocab_size = sp.get_piece_size()
logging.info(params)
@ -486,8 +494,20 @@ def main():
logging.info("About to create model")
model = get_transducer_model(params)
if params.avg_last_n > 0:
filenames = find_checkpoints(params.exp_dir)[: params.avg_last_n]
if params.iter > 0:
filenames = find_checkpoints(params.exp_dir, iteration=-params.iter)[
: params.avg
]
if len(filenames) == 0:
raise ValueError(
f"No checkpoints found for"
f" --iter {params.iter}, --avg {params.avg}"
)
elif len(filenames) < params.avg:
raise ValueError(
f"Not enough checkpoints ({len(filenames)}) found for"
f" --iter {params.iter}, --avg {params.avg}"
)
logging.info(f"averaging {filenames}")
model.to(device)
model.load_state_dict(average_checkpoints(filenames, device=device))

3
egs/librispeech/ASR/pruned_transducer_stateless3/train.py
View File

@ -725,6 +725,8 @@ def train_one_epoch(
try:
batch = next(dl)
except StopIteration:
name = "libri" if idx == 0 else "giga"
logging.info(f"{name} reaches end of dataloader")
break
batch_idx += 1
@ -966,6 +968,7 @@ def run(rank, world_size, args):
train_giga_cuts = gigaspeech.train_S_cuts()
train_giga_cuts = filter_short_and_long_utterances(train_giga_cuts)
train_giga_cuts = train_giga_cuts.repeat(times=None)
if args.enable_musan:
cuts_musan = load_manifest(