"model_dir": model_dir,
"model_blueprint": model_blueprint,
"exist_model": exist_model,
"start_epoch": train_stage,
"epochs": epochs,
"use_gpu": use_gpu,
"gpu_id": gpu_id,
"max_change": 10.,
"benchmark": benchmark,
"suffix": suffix,
"report_times_every_epoch": report_times_every_epoch,
"report_interval_iters": report_interval_iters,
"record_file": "train.csv"
})
trainer = trainer.SimpleTrainer(package)
if run_lr_finder and utils.is_main_training():
trainer.run_lr_finder("lr_finder.csv",
init_lr=1e-8,
final_lr=10.,
num_iters=2000,
beta=0.98)
endstage = 3 # Do not start extractor.
else:
trainer.run()
#### Extract xvector
if stage <= 4 <= endstage and utils.is_main_training():
# There are some params for xvector extracting.
data_root = "data" # It contains all dataset just like Kaldi recipe.
}, {
"model_dir": model_dir,
"model_blueprint": model_blueprint,
"exist_model": exist_model,
"start_epoch": train_stage,
"epochs": epochs,
"use_gpu": use_gpu,
"gpu_id": gpu_id,
"benchmark": benchmark,
"suffix": suffix,
"report_times_every_epoch": report_times_every_epoch,
"report_interval_iters": report_interval_iters,
"record_file": "train.csv"
})
trainer = trainer.SimpleTrainer(package, stop_early=stop_early)
if run_lr_finder and utils.is_main_training():
trainer.run_lr_finder("lr_finder.csv",
init_lr=1e-8,
final_lr=10.,
num_iters=2000,
beta=0.98)
endstage = 3 # Do not start extractor.
else:
trainer.run()
#### Extract xvector
if stage <= 4 <= endstage and utils.is_main_training():
# There are some params for xvector extracting.
data_root = "data" # It contains all dataset just like Kaldi recipe.