),
Arg("--csv", action="store_true", help="print as CSV"),
],
is_default=True,
),
Cmd("config", config, "display experiment config", [experiment_id_arg("experiment ID")]),
Cmd(
"describe",
describe,
"describe experiment",
[
Arg("experiment_ids", help="comma-separated list of experiment IDs to describe"),
Arg("--metrics", action="store_true", help="display full metrics"),
Group(
Arg("--csv", action="store_true", help="print as CSV"),
Arg("--json", action="store_true", help="print as JSON"),
Arg("--outdir", type=Path, help="directory to save output"),
),
],
),
Cmd(
"download-model-def",
download_model_def,
"download model definition",
[
experiment_id_arg("experiment ID"),
Arg("--output-dir", type=Path, help="output directory", default="."),
],
),
Cmd(
"list-trials lt",
# fmt: off
args_description = [
Cmd("tensorboard", None, "manage TensorBoard instances", [
Cmd("list ls",
partial(command.list_tasks),
"list TensorBoard instances", [
Arg("-q",
"--quiet",
action="store_true",
help="only display the IDs"),
Arg("--all",
"-a",
action="store_true",
help="show all TensorBoards (including other users')"),
Group(format_args["json"], format_args["csv"]),
],
is_default=True),
Cmd("start", start_tensorboard, "start new TensorBoard instance", [
Arg("experiment_ids",
type=int,
nargs="*",
help=
"experiment IDs to load into TensorBoard. At most 100 trials from "
"the specified experiment will be loaded into TensorBoard. If the "
"experiment has more trials, the 100 best-performing trials will "
"be used."),
Arg("-t",
"--trial-ids",
nargs=ONE_OR_MORE,
type=int,
return
api.pprint_task_logs(args.master, resp["id"], follow=True)
# fmt: off
args_description = [
Cmd("command cmd", None, "manage commands", [
Cmd("list ls", command.list_tasks, "list commands", [
Arg("-q", "--quiet", action="store_true",
help="only display the IDs"),
Arg("--all", "-a", action="store_true",
help="show all commands (including other users')"),
Group(
Arg("--csv", action="store_true", help="print as CSV"),
Arg("--json", action="store_true", help="print as JSON"),
),
], is_default=True),
Cmd("config", command.config,
"display command config", [
Arg("command_id", type=str, help="command ID"),
]),
Cmd("run", run_command, "create command", [
Arg("entrypoint", type=str, nargs=REMAINDER,
help="entrypoint command and arguments to execute"),
Arg("--config-file", default=None, type=FileType("r"),
help="command config file (.yaml)"),
Arg("-v", "--volume", action="append", default=[],
help=VOLUME_DESC),
Arg("-c", "--context", default=None, type=Path, help=CONTEXT_DESC),
Arg("--config", action="append", default=[], help=CONFIG_DESC),
None,
"local help",
[
Cmd(
"cluster-up",
handle_cluster_up,
"Create a Determined cluster",
[
Group(
Arg(
"--master-config-path",
type=Path,
default=None,
help="path to master configuration",
),
Arg(
"--storage-host-path",
type=Path,
default=DEFAULT_STORAGE_HOST_PATH,
help=
"Storage location for cluster data (e.g. checkpoints)",
),
),
Arg(
"--agents",
type=int,
default=1,
help="number of agents to start (on this machine)",
),
Arg("--master-port",
type=int,
args_description = [
Cmd(
"t|rial",
None,
"manage trials",
[
Cmd(
"describe",
describe_trial,
"describe trial",
[
Arg("trial_id", type=int, help="trial ID"),
Arg("--metrics", action="store_true", help="display full metrics"),
Group(
Arg("--csv", action="store_true", help="print as CSV"),
Arg("--json", action="store_true", help="print JSON"),
),
],
),
Cmd(
"download",
download,
"download checkpoint for trial",
[
Arg("trial_id", type=int, help="trial ID"),
Group(
Arg(
"--best",
action="store_true",
help="download the checkpoint with the best validation metric",
),
# The `tail` parameter only makes sense the first time we
# fetch logs.
response = api.get(
args.master, "logs", params={"greater_than_id": str(latest_log_id)}
)
latest_log_id = process_response(response, latest_log_id)
except KeyboardInterrupt:
break
# fmt: off
args_description = [
Cmd("master", None, "manage master", [
Cmd("config", config, "fetch master config", [
Group(format_args["json"], format_args["yaml"])
]),
Cmd("info", get_master, "fetch master info", [
Group(format_args["json"], format_args["yaml"])
]),
Cmd("logs", logs, "fetch master logs", [
Arg("-f", "--follow", action="store_true",
help="follow the logs of master, similar to tail -f"),
Arg("--tail", type=int,
help="number of lines to show, counting from the end "
"of the log (default is all)")
]),
])
] # type: List[Any]
# fmt: on
def agent_id_completer(_1: str, parsed_args: argparse.Namespace,
_2: Any) -> List[str]:
r = api.get(parsed_args.master, "agents")
return list(r.json().keys())
# fmt: off
args_description = [
Cmd("a|gent", None, "manage agents", [
Cmd("list",
list_agents,
"list agents", [
Group(
Arg("--csv", action="store_true", help="print as CSV"),
Arg("--json", action="store_true", help="print as JSON"),
),
],
is_default=True),
Cmd("enable", patch_agent(True), "enable agent", [
Group(
Arg("agent_id",
help="agent ID",
nargs="?",
completer=agent_id_completer),
Arg("--all", action="store_true", help="enable all agents"),
)
]),
Cmd("disable", patch_agent(False), "disable agent", [
Group(
Arg("agent_id",
common_log_options = [
Arg(
"-f",
"--follow",
action="store_true",
help="follow the logs of a running task, similar to tail -f",
),
Group(
Arg(
"--head",
type=int,
help=
"number of lines to show, counting from the beginning of the log",
),
Arg(
"--tail",
type=int,
help="number of lines to show, counting from the end of the log",
),
),
Arg(
"--allocation-id",
dest="allocation_ids",
action="append",
help="allocations to show logs from (repeat for multiple values)",
),
Arg(
"--agent-id",
dest="agent_ids",
[
Cmd(
"list ls",
ls,
"list jobs",
[
Arg(
"-p",
"--resource-pool",
type=str,
help="The target resource pool, if any.",
),
*pagination_args,
Group(
format_args["json"],
format_args["yaml"],
format_args["table"],
format_args["csv"],
),
],
is_default=True,
),
Cmd(
"u|pdate",
update,
"update job",
[
Arg("job_id", type=str, help="The target job ID"),
Group(
Arg(
"-p",
"--priority",