SEML: Slurm Experiment Management Library
seml wouldn't output anything to the console if SSH forward were used.typer.seml via pip install seml[ssh_forward].seml configure --ssh_forward
stash_all_py_files to the seml configuration block to stash all python files within the current directory (instead of only stashing the imported ones).seml.experiment.Experiment directly from seml.seml now provides template(s), if you want to initialize a new project simply run seml project init <project_name> and a new folder project_name will be created with sensible defaults. List available templates via seml project list-templates.
SETTINGS.EXPERIMENT.TERMINAL_WIDTH.seml.get_results
print-output.sacred.Experiment has been deprecated. Please use seml.experiment.Experiment going forward. See https://github.com/TUM-DAML/seml/blob/master/examples/example_experiment.py
seml.experiment.Experiment already includes the common MongoDB, logging and experiment statistics collection setupsettings.py or for a specific experiment with seml.experiment.Experiment(capture_output=True).seml <col> print-output: print experiment outputs directly via CLI. This command first attempts to gather the output via the slurm log file. If that fails it will search for the correct log within the MongoDB.seml --version
old
import sacred
import seml
ex = sacred.Experiment()
seml.setup_logger(ex)
@ex.post_run_hook
def collect_stats(_run):
seml.collect_exp_stats(_run)
@ex.config
def config():
overwrite = None
db_collection = None
if db_collection is not None:
ex.observers.append(
seml.create_mongodb_observer(db_collection, overwrite=overwrite)
)
new
from seml.experiment import Experiment
ex = Experiment()
seml drop <regex> to delete all collections matching the regex pattern. (be careful with this one)seml's CLI has been reimplemented for better usability, scalability, and optics. (#116)
rich for pretty printing.seml now supports sacred's named configurations (#119)
$config[_<optional_name>]: str | {'name': str, 'priority': int}. Use the optional priority to indicate the import order. The optional name allows for importing multiple configs.seml now supports descriptions! (#118)
seml.description in your yaml files to set descriptions.seml collection description set
status, list and print-fail-trace
seml <collection> add *.yaml
seml <collection> add config1.yaml config2.yaml
seml list <regex> to list the state of all collections matching the regex (if left empty, all collections will be shown). (#114)
seml <collection> print-fail-trace to print all fail traces in a collection. (#114)
status and print-fail-trace (#118)
seml collection status -p config.dataset -p config.model
seml --install-completion {shell}
seml will complete collection names, command names, and flags.seml configure has been adjusted to incorporate more configurations in the future. For now, the default behavior is identical.seml <collection> delete. (#100)seml is now faster
time seml: 1.2s - 2stime seml: 0.5 - 0.8smunch and sacred: ~0.3sgrid.TMP_DIRECTORY in settings.py (#94)directConnection (ca34fbf52b552cab4b7a6811d750f4f1c233175c)seed parameter (#93)sbatch / srun fails (#77).numpy.py and raise an error accordingly (#78).TensorFlow >= 2.5
pymongo >= 4 (#76)seml jupyter (#86)max_jobs_per_batch config option.reload-sources command for updating the source code of specified experiments (#57, #64)seml coll cancel reset reload-sources start (#61)-y option to confirm changes directly (#62)max_simultaneous_jobs config option that limits the number of simultaneously running jobs per job arrayrepr and shlex for proper shell-escaping and handling of parameter edge cases (#51, #53, #69)CONDA_DEFAULT_ENV is not defined