nshtrainer 0.29.1__py3-none-any.whl → 0.30.1__py3-none-any.whl

nshtrainer/__init__.py

@@ -1,5 +1,6 @@
 from . import _experimental as _experimental
 from . import callbacks as callbacks
+from . import config as config
 from . import data as data
 from . import lr_scheduler as lr_scheduler
 from . import metrics as metrics

nshtrainer/config/__init__.py

@@ -0,0 +1,3 @@
+from .duration import Duration as Duration
+from .duration import Epochs as Epochs
+from .duration import Steps as Steps

nshtrainer/config/duration.py

@@ -0,0 +1,31 @@
+import math
+from typing import Annotated, Literal
+
+import nshconfig as C
+
+
+class Steps(C.Config):
+    kind: Literal["steps"] = "steps"
+
+    value: Annotated[int, C.Field(ge=0)]
+    """Number of steps."""
+
+    def to_steps(self, steps_per_epoch: int):
+        return self
+
+
+class Epochs(C.Config):
+    kind: Literal["epochs"] = "epochs"
+
+    value: Annotated[int | float, C.Field(ge=0)]
+    """Number of epochs."""
+
+    def to_steps(self, steps_per_epoch: int):
+        value = self.value * steps_per_epoch
+        if not isinstance(value, int):
+            value = int(math.ceil(value))
+
+        return Steps(value=value)
+
+
+Duration = Annotated[Steps | Epochs, C.Field(discriminator="kind")]

nshtrainer/lr_scheduler/linear_warmup_cosine.py

@@ -2,11 +2,11 @@ import math
 import warnings
 from typing import Literal
 
-import nshconfig as C
 from torch.optim import Optimizer
 from torch.optim.lr_scheduler import LRScheduler
 from typing_extensions import override
 
+from ..config import Duration
 from ._base import LRSchedulerConfigBase, LRSchedulerMetadata
 
 
@@ -91,13 +91,13 @@ class LinearWarmupCosineAnnealingLR(LRScheduler):
 class LinearWarmupCosineDecayLRSchedulerConfig(LRSchedulerConfigBase):
     name: Literal["linear_warmup_cosine_decay"] = "linear_warmup_cosine_decay"
 
-    warmup_epochs: int = C.Field(ge=0)
-    r"""The number of epochs for the linear warmup phase.
-    The learning rate is linearly increased from `warmup_start_lr` to the initial learning rate over this number of epochs."""
+    warmup_duration: Duration
+    r"""The duration for the linear warmup phase.
+    The learning rate is linearly increased from `warmup_start_lr` to the initial learning rate over this duration."""
 
-    max_epochs: int = C.Field(gt=0)
-    r"""The total number of epochs.
-    The learning rate is decayed to `min_lr` over this number of epochs."""
+    max_duration: Duration
+    r"""The total duration.
+    The learning rate is decayed to `min_lr` over this duration."""
 
     warmup_start_lr_factor: float = 0.0
     r"""The initial learning rate for the linear warmup phase, as a factor of the initial learning rate.
@@ -121,11 +121,15 @@ class LinearWarmupCosineDecayLRSchedulerConfig(LRSchedulerConfigBase):
     @override
     def create_scheduler_impl(self, optimizer, lightning_module, lr):
         num_steps_per_epoch = self.compute_num_steps_per_epoch(lightning_module)
-        warmup_steps = self.warmup_epochs * num_steps_per_epoch
-        max_steps = self.max_epochs * num_steps_per_epoch
+        warmup_steps = self.warmup_duration.to_steps(num_steps_per_epoch).value
+        max_steps = self.max_duration.to_steps(num_steps_per_epoch).value
         warmup_start_lr = self.warmup_start_lr_factor * lr
         min_lr = self.min_lr_factor * lr
 
+        # Warmup and max steps should be at least 1.
+        warmup_steps = max(warmup_steps, 1)
+        max_steps = max(max_steps, 1)
+
         # Create the scheduler
         scheduler = LinearWarmupCosineAnnealingLR(
             optimizer=optimizer,

{nshtrainer-0.29.1.dist-info → nshtrainer-0.30.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nshtrainer
-Version: 0.29.1
+Version: 0.30.1
 Summary: 
 Author: Nima Shoghi
 Author-email: nimashoghi@gmail.com

{nshtrainer-0.29.1.dist-info → nshtrainer-0.30.1.dist-info}/RECORD

@@ -1,4 +1,4 @@
-nshtrainer/__init__.py,sha256=39loiLLXbaGiozEsAn8mPHopxaPsek8JsgR9DD2gxtY,583
+nshtrainer/__init__.py,sha256=sUb2yNdkHHhrKWCeWA5QKIA1Xx3jkO1QGD5Pa-HvgbA,614
 nshtrainer/_callback.py,sha256=A1zLsTy4b_wOYnInLLXGSRdHzT2yNa6mPEql-ozm0u0,1013
 nshtrainer/_checkpoint/loader.py,sha256=5vjg-OFChXJjgiOVv8vnV8nwTscfdDtEdxQRz6uPfDE,14158
 nshtrainer/_checkpoint/metadata.py,sha256=5D4PgKodzhLsmQvuF3xxkH49epKaegxi4wh_ImDTtns,4737
@@ -25,6 +25,8 @@ nshtrainer/callbacks/print_table.py,sha256=_FdAHhqylWGk4Z0c2FrLFeiMA4jhfA_beZRK_
 nshtrainer/callbacks/throughput_monitor.py,sha256=H_ocXErZxUO3dxFk8Tx_VQdpI9E_Ztvqof5WtFevLyQ,1838
 nshtrainer/callbacks/timer.py,sha256=quS79oYClDUvQxJkNWmDMe0hwRUkkREgTgqzVrnom50,4607
 nshtrainer/callbacks/wandb_watch.py,sha256=Y6SEXfIx3kDDQbI5zpP53BVq0FBLJbLd3RJsiHZk1-Y,2921
+nshtrainer/config/__init__.py,sha256=v9RtlM1Pqj_4fCDfskgxEtiGtbWH3Tj7lqNsKCDQ4gk,119
+nshtrainer/config/duration.py,sha256=f_obz0eorkktI3HzAuIawABDkvuL4lDqCxcPb3UW7Q4,692
 nshtrainer/data/__init__.py,sha256=7mk1tr7SWUZ7ySbsf0y0ZPszk7u4QznPhQ-7wnpH9ec,149
 nshtrainer/data/balanced_batch_sampler.py,sha256=dGBTDDtlBU6c-ZlVQOCnTW7SjTB5hczWsOWEdUWjvkA,4385
 nshtrainer/data/transform.py,sha256=6SNs3_TpNpfhcwTwvPKyEJ3opM1OT7LmMEYQNHKgRl8,2227
@@ -52,7 +54,7 @@ nshtrainer/loggers/tensorboard.py,sha256=wL2amRSdP68zbslZvBeM0ZQBnjF3hIKsz-_lBbd
 nshtrainer/loggers/wandb.py,sha256=FPwbf618AYmuPzHdhd1ZFhJ8qDjwTUiSe7cm7g3KCyM,5112
 nshtrainer/lr_scheduler/__init__.py,sha256=uEvgaFAs-4s_bAEMaildy0GT6OvgpgOEKTuzqutESHE,736
 nshtrainer/lr_scheduler/_base.py,sha256=7xOIuxQ86YHbFWG5a3gX46emQj1WN_LaY4-i0Q1TDBg,3659
-nshtrainer/lr_scheduler/linear_warmup_cosine.py,sha256=mn6cyizyI_stkXtg6zxIEGF9btIxMRWigUHUTlUYCSw,5221
+nshtrainer/lr_scheduler/linear_warmup_cosine.py,sha256=Fyontbfu4k2932xZenE63QL4CrVGWANXdTeq63dUko0,5347
 nshtrainer/lr_scheduler/reduce_lr_on_plateau.py,sha256=h76oTHYpMxauV_l6lviya5DW-WKArwxxf7ZQizhmbCw,2782
 nshtrainer/metrics/__init__.py,sha256=ObLIELGguIEcUpRsUkqh1ltrvZii6vglTpJGrPvoy00,50
 nshtrainer/metrics/_config.py,sha256=jgRBfDAQLFTW7AiUY7CRtdfts6CR6keeuqm0FFMWCzQ,1288
@@ -87,6 +89,6 @@ nshtrainer/util/seed.py,sha256=Or2wMPsnQxfnZ2xfBiyMcHFIUt3tGTNeMMyOEanCkqs,280
 nshtrainer/util/slurm.py,sha256=rofIU26z3SdL79SF45tNez6juou1cyDLz07oXEZb9Hg,1566
 nshtrainer/util/typed.py,sha256=NGuDkDzFlc1fAoaXjOFZVbmj0mRFjsQi1E_hPa7Bn5U,128
 nshtrainer/util/typing_utils.py,sha256=8ptjSSLZxlmy4FY6lzzkoGoF5fGNClo8-B_c0XHQaNU,385
-nshtrainer-0.29.1.dist-info/METADATA,sha256=Qck1QY1pNnjQH9zLMyAMKVVvYMovEeIyP5zV7VlZios,916
-nshtrainer-0.29.1.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
-nshtrainer-0.29.1.dist-info/RECORD,,
+nshtrainer-0.30.1.dist-info/METADATA,sha256=LV0wQlmotpfC3qO76dFVCbS26bEl-9YMiTetEeqVQsU,916
+nshtrainer-0.30.1.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
+nshtrainer-0.30.1.dist-info/RECORD,,