embedding / hydra.yaml
Hervé Bredin
feat: initial import 0faaae7
hydra:
run:
dir: ${protocol}/${task._target_}/${now:%Y-%m-%d}/${now:%H-%M-%S}
sweep:
dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}/${protocol}/${task._target_}
subdir: ${hydra.job.num}
hydra_logging:
version: 1
formatters:
simple:
format: '[%(asctime)s][HYDRA] %(message)s'
handlers:
console:
class: logging.StreamHandler
formatter: simple
stream: ext://sys.stdout
root:
level: INFO
handlers:
- console
loggers:
logging_example:
level: DEBUG
disable_existing_loggers: false
job_logging:
version: 1
formatters:
simple:
format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
handlers:
console:
class: logging.StreamHandler
formatter: simple
stream: ext://sys.stdout
file:
class: logging.FileHandler
formatter: simple
filename: ${hydra.job.name}.log
root:
level: INFO
handlers:
- console
- file
disable_existing_loggers: false
sweeper:
_target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
max_batch_size: null
launcher:
_target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
help:
app_name: pyannote-audio-train
header: == ${hydra.help.app_name} ==
footer: 'Powered by Hydra (https://hydra.cc)
Use --hydra-help to view Hydra specific help'
template: "${hydra.help.header}\n\npyannote-audio-train protocol={protocol_name}\
\ task={task} model={model}\n\n{task} can be any of the following:\n* vad (default)\
\ = voice activity detection\n* scd = speaker change detection\n* osd = overlapped\
\ speech detection\n* xseg = multi-task segmentation\n\n{model} can be any of\
\ the following:\n* debug (default) = simple segmentation model for debugging\
\ purposes\n\n{optimizer} can be any of the following\n* adam (default) = Adam\
\ optimizer\n\n{trainer} can be any of the following\n* fast_dev_run for debugging\n\
* default (default) for training the model\n\nOptions\n=======\n\nHere, we describe\
\ the most common options: use \"--cfg job\" option to get a complete list.\n\
\n* task.duration: audio chunk duration (in seconds)\n* task.batch_size: number\
\ of audio chunks per batch\n* task.num_workers: number of workers used for\
\ generating training chunks\n\n* optimizer.lr: learning rate\n* trainer.auto_lr_find:\
\ use pytorch-lightning AutoLR\n\nHyper-parameter optimization\n============================\n\
\nBecause it is powered by Hydra (https://hydra.cc), one can run grid search\
\ using the --multirun option.\n\nFor instance, the following command will run\
\ the same job three times, with three different learning rates:\n pyannote-audio-train\
\ --multirun protocol={protocol_name} task={task} optimizer.lr=1e-3,1e-2,1e-1\n\
\nEven better, one can use Ax (https://ax.dev) sweeper to optimize learning\
\ rate directly:\n pyannote-audio-train --multirun hydra/sweeper=ax protocol={protocol_name}\
\ task={task} optimizer.lr=\"interval(1e-3, 1e-1)\"\n\nSee https://hydra.cc/docs/plugins/ax_sweeper\
\ for more details.\n\nUser-defined task or model\n==========================\n\
\n1. define your_package.YourTask (or your_package.YourModel) class\n2. create\
\ file /path/to/your_config/task/your_task.yaml (or /path/to/your_config/model/your_model.yaml)\n\
\ # @package _group_\n _target_: your_package.YourTask # or YourModel\n\
\ param1: value1\n param2: value2\n3. call pyannote-audio-train --config-dir\
\ /path/to/your_config task=your_task task.param1=modified_value1 model=your_model\
\ ...\n\n${hydra.help.footer}"
hydra_help:
hydra_help: ???
template: 'Hydra (${hydra.runtime.version})
See https://hydra.cc for more info.
== Flags ==
$FLAGS_HELP
== Configuration groups ==
Compose your configuration from those groups (For example, append hydra/job_logging=disabled
to command line)
$HYDRA_CONFIG_GROUPS
Use ''--cfg hydra'' to Show the Hydra config.
'
output_subdir: ''
overrides:
hydra: []
task:
- protocol=VoxCeleb.SpeakerVerification.VoxCeleb_X
- task=SpeakerEmbedding
- task.num_workers=20
- task.min_duration=2
- task.duration=5.
- task.num_classes_per_batch=64
- task.num_chunks_per_class=4
- task.margin=10.0
- task.scale=50.
- model=XVectorSincNet
- trainer.gpus=1
- +augmentation=background_then_reverb
job:
name: train
override_dirname: +augmentation=background_then_reverb,model=XVectorSincNet,protocol=VoxCeleb.SpeakerVerification.VoxCeleb_X,task.duration=5.,task.margin=10.0,task.min_duration=2,task.num_chunks_per_class=4,task.num_classes_per_batch=64,task.num_workers=20,task.scale=50.,task=SpeakerEmbedding,trainer.gpus=1
id: ???
num: ???
config_name: config
env_set: {}
env_copy: []
config:
override_dirname:
kv_sep: '='
item_sep: ','
exclude_keys: []
runtime:
version: 1.0.4
cwd: /gpfsdswork/projects/rech/eie/uno46kl/xvectors/debug
verbose: false