nshtrainer 1.0.0b47__py3-none-any.whl → 1.0.0b48__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- nshtrainer/callbacks/checkpoint/best_checkpoint.py +3 -3
- nshtrainer/callbacks/early_stopping.py +1 -1
- nshtrainer/callbacks/metric_validation.py +3 -3
- nshtrainer/callbacks/rlp_sanity_checks.py +1 -1
- nshtrainer/data/datamodule.py +2 -2
- nshtrainer/loggers/__init__.py +0 -1
- nshtrainer/lr_scheduler/reduce_lr_on_plateau.py +7 -7
- nshtrainer/metrics/_config.py +2 -19
- nshtrainer/model/base.py +4 -4
- nshtrainer/model/mixins/debug.py +1 -1
- nshtrainer/model/mixins/logger.py +12 -6
- {nshtrainer-1.0.0b47.dist-info → nshtrainer-1.0.0b48.dist-info}/METADATA +1 -1
- {nshtrainer-1.0.0b47.dist-info → nshtrainer-1.0.0b48.dist-info}/RECORD +14 -14
- {nshtrainer-1.0.0b47.dist-info → nshtrainer-1.0.0b48.dist-info}/WHEEL +0 -0
@@ -51,7 +51,7 @@ class BestCheckpointCallbackConfig(BaseCheckpointCallbackConfig):
|
|
51
51
|
class BestCheckpointCallback(CheckpointBase[BestCheckpointCallbackConfig]):
|
52
52
|
@property
|
53
53
|
def _metric_name_normalized(self):
|
54
|
-
return self.metric.
|
54
|
+
return self.metric.monitor.replace("/", "_").replace(" ", "_").replace(".", "_")
|
55
55
|
|
56
56
|
@override
|
57
57
|
def __init__(
|
@@ -69,12 +69,12 @@ class BestCheckpointCallback(CheckpointBase[BestCheckpointCallbackConfig]):
|
|
69
69
|
|
70
70
|
@override
|
71
71
|
def default_filename(self):
|
72
|
-
return f"epoch{{epoch}}-step{{step}}-{self._metric_name_normalized}{{{self.metric.
|
72
|
+
return f"epoch{{epoch}}-step{{step}}-{self._metric_name_normalized}{{{self.metric.monitor}}}"
|
73
73
|
|
74
74
|
@override
|
75
75
|
def topk_sort_key(self, metadata: CheckpointMetadata):
|
76
76
|
return metadata.metrics.get(
|
77
|
-
self.metric.
|
77
|
+
self.metric.monitor,
|
78
78
|
float("-inf" if self.metric.mode == "max" else "inf"),
|
79
79
|
)
|
80
80
|
|
@@ -68,7 +68,7 @@ class EarlyStoppingCallback(_EarlyStopping):
|
|
68
68
|
del config, metric
|
69
69
|
|
70
70
|
super().__init__(
|
71
|
-
monitor=self.metric.
|
71
|
+
monitor=self.metric.monitor,
|
72
72
|
mode=self.metric.mode,
|
73
73
|
patience=self.config.patience,
|
74
74
|
min_delta=self.config.min_delta,
|
@@ -55,14 +55,14 @@ class MetricValidationCallback(Callback):
|
|
55
55
|
self.metrics = metrics
|
56
56
|
|
57
57
|
def _check_metrics(self, trainer: Trainer):
|
58
|
-
metric_names = ", ".join(metric.
|
58
|
+
metric_names = ", ".join(metric.monitor for metric in self.metrics)
|
59
59
|
log.info(f"Validating metrics: {metric_names}...")
|
60
60
|
logged_metrics = set(trainer.logged_metrics.keys())
|
61
61
|
|
62
62
|
invalid_metrics: list[str] = []
|
63
63
|
for metric in self.metrics:
|
64
|
-
if metric.
|
65
|
-
invalid_metrics.append(metric.
|
64
|
+
if metric.monitor not in logged_metrics:
|
65
|
+
invalid_metrics.append(metric.monitor)
|
66
66
|
|
67
67
|
if invalid_metrics:
|
68
68
|
msg = (
|
@@ -171,7 +171,7 @@ class CustomRLPImplementation(Protocol):
|
|
171
171
|
__reduce_lr_on_plateau__: bool
|
172
172
|
|
173
173
|
|
174
|
-
class
|
174
|
+
class RLPSanityCheckModuleMixin(LightningModule):
|
175
175
|
def reduce_lr_on_plateau_config(
|
176
176
|
self,
|
177
177
|
lr_scheduler: LRSchedulerTypeUnion | LRSchedulerConfigType,
|
nshtrainer/data/datamodule.py
CHANGED
@@ -11,13 +11,13 @@ from lightning.pytorch import LightningDataModule
|
|
11
11
|
from typing_extensions import Never, TypeVar, deprecated, override
|
12
12
|
|
13
13
|
from ..model.mixins.callback import CallbackRegistrarModuleMixin
|
14
|
-
from ..model.mixins.debug import
|
14
|
+
from ..model.mixins.debug import DebugModuleMixin
|
15
15
|
|
16
16
|
THparams = TypeVar("THparams", bound=C.Config, infer_variance=True)
|
17
17
|
|
18
18
|
|
19
19
|
class LightningDataModuleBase(
|
20
|
-
|
20
|
+
DebugModuleMixin,
|
21
21
|
CallbackRegistrarModuleMixin,
|
22
22
|
LightningDataModule,
|
23
23
|
ABC,
|
nshtrainer/loggers/__init__.py
CHANGED
@@ -49,13 +49,13 @@ class ReduceLROnPlateauConfig(LRSchedulerConfigBase):
|
|
49
49
|
if (metric := self.metric) is None:
|
50
50
|
from ..trainer import Trainer
|
51
51
|
|
52
|
-
assert isinstance(
|
53
|
-
trainer
|
54
|
-
)
|
52
|
+
assert isinstance(trainer := lightning_module.trainer, Trainer), (
|
53
|
+
"The trainer must be a `nshtrainer.Trainer` instance."
|
54
|
+
)
|
55
55
|
|
56
|
-
assert (
|
57
|
-
metric
|
58
|
-
)
|
56
|
+
assert (metric := trainer.hparams.primary_metric) is not None, (
|
57
|
+
"Primary metric must be provided if metric is not specified."
|
58
|
+
)
|
59
59
|
|
60
60
|
lr_scheduler = ReduceLROnPlateau(
|
61
61
|
optimizer,
|
@@ -70,7 +70,7 @@ class ReduceLROnPlateauConfig(LRSchedulerConfigBase):
|
|
70
70
|
)
|
71
71
|
return {
|
72
72
|
"scheduler": lr_scheduler,
|
73
|
-
"monitor": metric.
|
73
|
+
"monitor": metric.monitor,
|
74
74
|
}
|
75
75
|
|
76
76
|
@override
|
nshtrainer/metrics/_config.py
CHANGED
@@ -7,8 +7,8 @@ import nshconfig as C
|
|
7
7
|
|
8
8
|
|
9
9
|
class MetricConfig(C.Config):
|
10
|
-
|
11
|
-
"""The name of the
|
10
|
+
monitor: str
|
11
|
+
"""The name of the metric to monitor."""
|
12
12
|
|
13
13
|
mode: Literal["min", "max"]
|
14
14
|
"""
|
@@ -17,23 +17,6 @@ class MetricConfig(C.Config):
|
|
17
17
|
- "max" for metrics that should be maximized (e.g., accuracy)
|
18
18
|
"""
|
19
19
|
|
20
|
-
@property
|
21
|
-
def validation_monitor(self) -> str:
|
22
|
-
return f"val/{self.name}"
|
23
|
-
|
24
|
-
def __post_init__(self):
|
25
|
-
for split in ("train", "val", "test", "predict"):
|
26
|
-
if self.name.startswith(f"{split}/"):
|
27
|
-
raise ValueError(
|
28
|
-
f"Primary metric name should not start with '{split}/'. "
|
29
|
-
f"Just use '{self.name[len(split) + 1:]}' instead. "
|
30
|
-
"The split name is automatically added depending on the context."
|
31
|
-
)
|
32
|
-
|
33
|
-
@classmethod
|
34
|
-
def loss(cls, mode: Literal["min", "max"] = "min"):
|
35
|
-
return cls(name="loss", mode=mode)
|
36
|
-
|
37
20
|
@property
|
38
21
|
def best(self):
|
39
22
|
return builtins.min if self.mode == "min" else builtins.max
|
nshtrainer/model/base.py
CHANGED
@@ -15,9 +15,9 @@ from lightning.pytorch.utilities.model_helpers import is_overridden
|
|
15
15
|
from lightning.pytorch.utilities.rank_zero import rank_zero_warn
|
16
16
|
from typing_extensions import Never, TypeVar, deprecated, override
|
17
17
|
|
18
|
-
from ..callbacks.rlp_sanity_checks import
|
18
|
+
from ..callbacks.rlp_sanity_checks import RLPSanityCheckModuleMixin
|
19
19
|
from .mixins.callback import CallbackModuleMixin
|
20
|
-
from .mixins.debug import
|
20
|
+
from .mixins.debug import DebugModuleMixin
|
21
21
|
from .mixins.logger import LoggerLightningModuleMixin
|
22
22
|
|
23
23
|
log = logging.getLogger(__name__)
|
@@ -54,8 +54,8 @@ VALID_REDUCE_OPS = (
|
|
54
54
|
|
55
55
|
|
56
56
|
class LightningModuleBase(
|
57
|
-
|
58
|
-
|
57
|
+
DebugModuleMixin,
|
58
|
+
RLPSanityCheckModuleMixin,
|
59
59
|
LoggerLightningModuleMixin,
|
60
60
|
CallbackModuleMixin,
|
61
61
|
LightningModule,
|
nshtrainer/model/mixins/debug.py
CHANGED
@@ -54,6 +54,12 @@ class _LogContextKwargs:
|
|
54
54
|
d = dataclasses.asdict(self)
|
55
55
|
for field in self.__ignore_fields__:
|
56
56
|
d.pop(field, None)
|
57
|
+
|
58
|
+
# Pop all None values
|
59
|
+
for k in list(d.keys()):
|
60
|
+
if d[k] is None:
|
61
|
+
d.pop(k)
|
62
|
+
|
57
63
|
return d
|
58
64
|
|
59
65
|
|
@@ -134,18 +140,18 @@ class LoggerLightningModuleMixin(mixin_base_type(LightningModule)):
|
|
134
140
|
self,
|
135
141
|
name: str,
|
136
142
|
value: _METRIC,
|
137
|
-
prog_bar: bool =
|
143
|
+
prog_bar: bool | None = None,
|
138
144
|
logger: bool | None = None,
|
139
145
|
on_step: bool | None = None,
|
140
146
|
on_epoch: bool | None = None,
|
141
|
-
reduce_fx: str | Callable =
|
142
|
-
enable_graph: bool =
|
143
|
-
sync_dist: bool =
|
147
|
+
reduce_fx: str | Callable | None = None,
|
148
|
+
enable_graph: bool | None = None,
|
149
|
+
sync_dist: bool | None = None,
|
144
150
|
sync_dist_group: Any | None = None,
|
145
|
-
add_dataloader_idx: bool =
|
151
|
+
add_dataloader_idx: bool | None = None,
|
146
152
|
batch_size: int | None = None,
|
147
153
|
metric_attribute: str | None = None,
|
148
|
-
rank_zero_only: bool =
|
154
|
+
rank_zero_only: bool | None = None,
|
149
155
|
) -> None:
|
150
156
|
# If logging is disabled, then do nothing.
|
151
157
|
if not self.logging_enabled:
|
@@ -11,22 +11,22 @@ nshtrainer/callbacks/actsave.py,sha256=NSXIIu62MNYe5gz479SMW33bdoKYoYtWtd_iTWFpK
|
|
11
11
|
nshtrainer/callbacks/base.py,sha256=Alaou1IHAIlMEM7g58d_02ozY2xWlshBN7fsw5Ee21s,3683
|
12
12
|
nshtrainer/callbacks/checkpoint/__init__.py,sha256=l8tkHc83_mLiU0-wT09SWdRzwpm2ulbkLzcuCmuTwzE,620
|
13
13
|
nshtrainer/callbacks/checkpoint/_base.py,sha256=f7lpk8W4xqxk3PolBEU3AWt9VTIpoLW7wMUhC5DNm3c,6345
|
14
|
-
nshtrainer/callbacks/checkpoint/best_checkpoint.py,sha256=
|
14
|
+
nshtrainer/callbacks/checkpoint/best_checkpoint.py,sha256=aCs3E1eucfDlUeW2Iq_Ke7hb96BxHanmvn7PCCbqq0E,2648
|
15
15
|
nshtrainer/callbacks/checkpoint/last_checkpoint.py,sha256=vn-as3ex7kaTRcKsIurVtM6kUSHYNwHJeYG82j2dMcc,3554
|
16
16
|
nshtrainer/callbacks/checkpoint/on_exception_checkpoint.py,sha256=nljzETqkHwA-4g8mxaeFK5HxA8My0dlIPzIUscSMWyk,3525
|
17
17
|
nshtrainer/callbacks/debug_flag.py,sha256=96fuP0C7C6dSs1GiMeUYzzs0X3Q4Pjt9JVWg3b75fU4,1748
|
18
18
|
nshtrainer/callbacks/directory_setup.py,sha256=wPas_Ren8ANejogmIdKhqqgj4ulxz9AS_8xVIAfRXa0,2565
|
19
|
-
nshtrainer/callbacks/early_stopping.py,sha256=
|
19
|
+
nshtrainer/callbacks/early_stopping.py,sha256=rC_qYKCQWjRQJFo0ky46uG0aDJdYP8vsSlKunk0bUVI,4765
|
20
20
|
nshtrainer/callbacks/ema.py,sha256=dBFiUXG0xmyCw8-ayuSzJMKqSbepl6Ii5VIbhFlT5ug,12255
|
21
21
|
nshtrainer/callbacks/finite_checks.py,sha256=3lZ3kEIjmYQfqTF0DcrgZ9_98ZLQhQj8usH7SgWst3o,2185
|
22
22
|
nshtrainer/callbacks/gradient_skipping.py,sha256=8g7oC7PF0LTAEzwiNoaS5tWOnkjk_EB0QG3JdHkQ8ek,3523
|
23
23
|
nshtrainer/callbacks/interval.py,sha256=UCzUzt3XCFVyQyCWL9lOrStkkxesvduNOYk8yMrGTTk,8116
|
24
24
|
nshtrainer/callbacks/log_epoch.py,sha256=B5Dm8XVZwCzKUhUWfT_5PDdDac993191OsbcxxuSVJE,1457
|
25
25
|
nshtrainer/callbacks/lr_monitor.py,sha256=qy_C0R40J0hBAukzBwng5FI2jJUpWuXOi5N6FU6ym3I,1210
|
26
|
-
nshtrainer/callbacks/metric_validation.py,sha256=
|
26
|
+
nshtrainer/callbacks/metric_validation.py,sha256=4RDr1FuNKfro-6QEtmcFqT4iNf2twmJVNk9y-8nq9bg,2882
|
27
27
|
nshtrainer/callbacks/norm_logging.py,sha256=nVIDWe-ASl5zN830-ODR8QMCqI1ma-QPCIwoy0Wb-Nk,6390
|
28
28
|
nshtrainer/callbacks/print_table.py,sha256=VaS4JgI963do79laXK4lUkFQx8v6aRSy22W0zyal_LA,3035
|
29
|
-
nshtrainer/callbacks/rlp_sanity_checks.py,sha256=
|
29
|
+
nshtrainer/callbacks/rlp_sanity_checks.py,sha256=Df9Prq2QKXnaeMBIvMQBhDhJTDeru5UbiuXJOJR16Gk,10050
|
30
30
|
nshtrainer/callbacks/shared_parameters.py,sha256=s94jJTAIbDGukYJu6l247QonVOCudGClU4t5kLt8XrY,3076
|
31
31
|
nshtrainer/callbacks/timer.py,sha256=gDcw_K_ikf0bkVgxQ0cDhvvNvz6GLZVLcatuKfh0ORU,4731
|
32
32
|
nshtrainer/callbacks/wandb_upload_code.py,sha256=shV7UtnXgY2bUlXdVrXiaDs0PNLlIt7TzNJkJPkzvzI,2414
|
@@ -100,9 +100,9 @@ nshtrainer/configs/util/config/dtype/__init__.py,sha256=PmGF-O4r6SXqEaagVsQ5YxEq
|
|
100
100
|
nshtrainer/configs/util/config/duration/__init__.py,sha256=44lS2irOIPVfgshMTfnZM2jC6l0Pjst9w2M_lJoS_MU,353
|
101
101
|
nshtrainer/data/__init__.py,sha256=K4i3Tw4g9EOK2zlMMbidi99y0SyI4w8P7_XUf1n42Ts,260
|
102
102
|
nshtrainer/data/balanced_batch_sampler.py,sha256=r1cBKRXKHD8E1Ax6tj-FUbE-z1qpbO58mQ9VrK9uLnc,5481
|
103
|
-
nshtrainer/data/datamodule.py,sha256=
|
103
|
+
nshtrainer/data/datamodule.py,sha256=0M-HjGZQkLG77HXn4ZgLSypnbSjkjTq6GEJwGWe_gbM,4136
|
104
104
|
nshtrainer/data/transform.py,sha256=qd0lIocO59Fk_m90xyOHgFezbymd1mRwly8nbYIfHGc,2263
|
105
|
-
nshtrainer/loggers/__init__.py,sha256=
|
105
|
+
nshtrainer/loggers/__init__.py,sha256=fI0OHEltHP4tZI-KFB3npdzoxm_M2QsEYKxY3um05_s,592
|
106
106
|
nshtrainer/loggers/actsave.py,sha256=wgNrpBB6wQM7qff8iLDb_sQnbiAcYHRmH56pcEJPB3o,1409
|
107
107
|
nshtrainer/loggers/base.py,sha256=ON92XbwTSgadQOSyw5PiRRFzyH6uJ-xLtE0nB3cbgPc,1205
|
108
108
|
nshtrainer/loggers/csv.py,sha256=xJ8mSmw4vJwinIfqhF6t2HWmh_1dXEYyLfGuXwL7WHo,1160
|
@@ -111,14 +111,14 @@ nshtrainer/loggers/wandb.py,sha256=KZXAUWrrmdX_L8rqej77oUHaM0JxZRM8y9z6JP9PISw,6
|
|
111
111
|
nshtrainer/lr_scheduler/__init__.py,sha256=daMMK3erUcNXGGd_nZB8AWu3ZTYqfS1RSWeK4FV2udw,851
|
112
112
|
nshtrainer/lr_scheduler/base.py,sha256=LE53JRBTuAlA1fqbMgCZ7m39D1z0rGj2TizhJ62CPvE,3756
|
113
113
|
nshtrainer/lr_scheduler/linear_warmup_cosine.py,sha256=MsoXgCcWTKsrkNZiGnKS6yC-slRuleuwFxeM_lmG_pQ,5560
|
114
|
-
nshtrainer/lr_scheduler/reduce_lr_on_plateau.py,sha256=
|
114
|
+
nshtrainer/lr_scheduler/reduce_lr_on_plateau.py,sha256=zKO_4Cl28m3TopoNFmc5H6GSUuVUGYUoAlXpMh_EJIk,2931
|
115
115
|
nshtrainer/metrics/__init__.py,sha256=Nqkn_jsDf3n5WtfMcnaaEftYjIIT2b-S7rmsB1MOMkU,86
|
116
|
-
nshtrainer/metrics/_config.py,sha256=
|
116
|
+
nshtrainer/metrics/_config.py,sha256=ox_ScK6V0J9nzIMhEB0qpToNKpt83VVgOVSRFCV-wBc,595
|
117
117
|
nshtrainer/model/__init__.py,sha256=3G-bwPPSRStWdsdwG9-rn0bXcRpEiP1BiQpF_qavtls,97
|
118
|
-
nshtrainer/model/base.py,sha256=
|
118
|
+
nshtrainer/model/base.py,sha256=bZMNap0rkxRbAbu2BOHV_6YS2iZZnvy6wVSMOXGa_ZM,8680
|
119
119
|
nshtrainer/model/mixins/callback.py,sha256=0LPgve4VszHbLipid4mpI1qnnmdGS2spivs0dXLvqHw,3154
|
120
|
-
nshtrainer/model/mixins/debug.py,sha256=
|
121
|
-
nshtrainer/model/mixins/logger.py,sha256=
|
120
|
+
nshtrainer/model/mixins/debug.py,sha256=ydLuAAaa7M5bX0gougZ5gWuZnvn4Ra9assal3IZ9hq8,2086
|
121
|
+
nshtrainer/model/mixins/logger.py,sha256=IYfyyW_1VAD_HiTsfX28P-XNgz_SMb07t5lwb5rjlZ0,6221
|
122
122
|
nshtrainer/nn/__init__.py,sha256=5Gg3nieGSC5_dXaI9KUVUUbM13hHexH9831m4hcf6no,1475
|
123
123
|
nshtrainer/nn/mlp.py,sha256=nYUgAISzuhC8sav6PloAdyz0PdEoikwppiXIuToEVdE,7550
|
124
124
|
nshtrainer/nn/module_dict.py,sha256=9plb8aQUx5TUEPhX5jI9u8LrpTeKe7jZAHi8iIqcN8w,2365
|
@@ -154,6 +154,6 @@ nshtrainer/util/seed.py,sha256=diMV8iwBKN7Xxt5pELmui-gyqyT80_CZzomrWhNss0k,316
|
|
154
154
|
nshtrainer/util/slurm.py,sha256=HflkP5iI_r4UHMyPjw9R4dD5AHsJUpcfJw5PLvGYBRM,1603
|
155
155
|
nshtrainer/util/typed.py,sha256=Xt5fUU6zwLKSTLUdenovnKK0N8qUq89Kddz2_XeykVQ,164
|
156
156
|
nshtrainer/util/typing_utils.py,sha256=MjY-CUX9R5Tzat-BlFnQjwl1PQ_W2yZQoXhkYHlJ_VA,442
|
157
|
-
nshtrainer-1.0.
|
158
|
-
nshtrainer-1.0.
|
159
|
-
nshtrainer-1.0.
|
157
|
+
nshtrainer-1.0.0b48.dist-info/METADATA,sha256=b26a0GYVQcEszYiodjGF34N7gvEKONBVuB1bXTv35U4,988
|
158
|
+
nshtrainer-1.0.0b48.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
|
159
|
+
nshtrainer-1.0.0b48.dist-info/RECORD,,
|
File without changes
|