deepmd.dpmodel.train.entrypoint

Contents

deepmd.dpmodel.train.entrypoint#

Backend-independent training entrypoint pipeline.

Attributes#

Classes#

`TrainEntrypointOptions`	Common command options for backend train entrypoints.
`AbstractTrainEntrypoint`	Shared pipeline for backend train entrypoints.

Module Contents#

deepmd.dpmodel.train.entrypoint.log[source]#

class deepmd.dpmodel.train.entrypoint.TrainEntrypointOptions[source]#

Common command options for backend train entrypoints.

input_file: str[source]#

output: str = 'out.json'[source]#

init_model: str | None = None[source]#

restart: str | None = None[source]#

init_frz_model: str | None = None[source]#

finetune: str | None = None[source]#

model_branch: str = ''[source]#

use_pretrain_script: bool = False[source]#

skip_neighbor_stat: bool = False[source]#

class deepmd.dpmodel.train.entrypoint.AbstractTrainEntrypoint[source]#

Bases: abc.ABC

Shared pipeline for backend train entrypoints.

Backend subclasses keep ownership of backend-specific feature handling, neighbor-stat updates, distributed setup, data construction, and trainer construction. This pipeline only coordinates the common command flow.

run(options: TrainEntrypointOptions) → None[source]#: Run the training entrypoint.

prepare_options(options: TrainEntrypointOptions) → TrainEntrypointOptions[source]#: Normalize command options before reading or preprocessing config.

load_config(input_file: str) → dict[str, Any][source]#: Load the JSON/YAML training config.

validate_options(config: dict[str, Any], options: TrainEntrypointOptions) → None[source]#: Validate backend feature support before mutating the config.

preprocess_config(config: dict[str, Any], options: TrainEntrypointOptions) → dict[str, Any][source]#: Apply backend-specific config preprocessing before argcheck.

is_multi_task(config: dict[str, Any]) → bool[source]#: Return whether the config is in multi-task layout.

update_input(config: dict[str, Any]) → dict[str, Any][source]#: Apply DeePMD input-version compatibility conversion.

normalize_config(config: dict[str, Any], *, multi_task: bool) → dict[str, Any][source]#: Run DeePMD argcheck normalization.

update_neighbor_stat(config: dict[str, Any], options: TrainEntrypointOptions, *, multi_task: bool) → tuple[dict[str, Any], Any][source]#: Update descriptor selections from neighbor statistics.

dump_config(config: dict[str, Any], output: str) → None[source]#: Dump the normalized config used for training.

print_summary() → None[source]#: Print backend summary information.

setup_run(options: TrainEntrypointOptions, config: dict[str, Any]) → None[source]#: Set up backend runtime state before trainer execution.

teardown_run(options: TrainEntrypointOptions, config: dict[str, Any]) → None[source]#: Tear down backend runtime state after trainer execution.

abstractmethod run_training(config: dict[str, Any], options: TrainEntrypointOptions, neighbor_stat: Any) → None[source]#: Build backend data/trainer objects and run training.