mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-09 10:02:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support specifying iteration number of checkpoints for decoding. (#336)

Browse Source 
See also #289
This commit is contained in:
Fangjun Kuang 2022-04-28 14:09:22 +08:00 committed by GitHub
parent 9aeea3e1af
commit caab6cfd92
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 40 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
egs/librispeech/ASR/pruned_transducer_stateless2/decode.py
View File

@ -98,27 +98,28 @@ 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(
@ -151,7 +152,7 @@ def get_parser():
"--beam-size",
type=int,
default=4,
help="""An interger indicating how many candidates we will keep for each
help="""An integer indicating how many candidates we will keep for each
frame. Used only when --decoding-method is beam_search or
modified_beam_search.""",
)
@ -453,13 +454,19 @@ 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}"
@ -476,8 +483,9 @@ 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)
@ -485,8 +493,20 @@ 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))