kostyl-toolkit 0.1.10__py3-none-any.whl → 0.1.11__py3-none-any.whl

kostyl/{ml_core → ml}/clearml/config_mixin.py

@@ -5,7 +5,7 @@ import clearml
 from caseconverter import pascalcase
 from caseconverter import snakecase
 
-from kostyl.ml_core.configs.base_model import KostylBaseModel
+from kostyl.ml.configs.base_model import KostylBaseModel
 from kostyl.utils.dict_manipulations import convert_to_flat_dict
 from kostyl.utils.dict_manipulations import flattened_dict_to_nested
 from kostyl.utils.fs import load_config

kostyl/{ml_core → ml}/clearml/pulling_utils.py

@@ -3,12 +3,13 @@ from typing import cast
 
 from clearml import InputModel
 from clearml import Task
+from transformers import Any
 from transformers import AutoModel
 from transformers import AutoTokenizer
 from transformers import PreTrainedModel
 from transformers import PreTrainedTokenizerBase
 
-from kostyl.ml_core.lightning.extenstions.pretrained_model import (
+from kostyl.ml.lightning.extenstions.pretrained_model import (
     LightningCheckpointLoaderMixin,
 )
 
@@ -48,6 +49,7 @@ def get_model_from_clearml[
     model: type[TModel],
     task: Task | None = None,
     ignore_remote_overrides: bool = True,
+    **kwargs: Any,
 ) -> TModel:
     """
     Retrieve a pretrained model from ClearML and instantiate it using the appropriate loader.
@@ -59,6 +61,7 @@ def get_model_from_clearml[
             will be connected to this task, with remote overrides optionally ignored.
         ignore_remote_overrides: When connecting the input model to the provided task, determines whether
             remote configuration overrides should be ignored.
+        **kwargs: Additional keyword arguments to pass to the model loading method.
 
     Returns:
         An instantiated model loaded either from a ClearML package directory or a Lightning checkpoint.
@@ -72,13 +75,13 @@ def get_model_from_clearml[
     local_path = Path(input_model.get_local_copy(raise_on_error=True))
 
     if local_path.is_dir() and input_model._is_package():
-        model_instance = model.from_pretrained(local_path)
+        model_instance = model.from_pretrained(local_path, **kwargs)
     elif local_path.suffix == ".ckpt":
         if not issubclass(model, LightningCheckpointLoaderMixin):
             raise ValueError(
                 f"Model class {model.__name__} is not compatible with Lightning checkpoints."
             )
-        model_instance = model.from_lighting_checkpoint(local_path)
+        model_instance = model.from_lighting_checkpoint(local_path, **kwargs)
     else:
         raise ValueError(
             f"Unsupported model format for path: {local_path}. "

kostyl/{ml_core → ml}/configs/hyperparams.py

@@ -2,7 +2,7 @@ from pydantic import BaseModel
 from pydantic import Field
 from pydantic import model_validator
 
-from kostyl.ml_core.clearml.config_mixin import ClearMLConfigMixin
+from kostyl.ml.clearml.config_mixin import ClearMLConfigMixin
 from kostyl.utils.logging import setup_logger
 
 

kostyl/{ml_core → ml}/configs/training_settings.py

@@ -3,7 +3,7 @@ from typing import Literal
 from pydantic import BaseModel
 from pydantic import Field
 
-from kostyl.ml_core.clearml.config_mixin import ClearMLConfigMixin
+from kostyl.ml.clearml.config_mixin import ClearMLConfigMixin
 from kostyl.utils.logging import setup_logger
 
 from .base_model import KostylBaseModel

kostyl/{ml_core → ml}/lightning/callbacks/checkpoint.py

@@ -3,8 +3,8 @@ from shutil import rmtree
 
 from lightning.pytorch.callbacks import ModelCheckpoint
 
-from kostyl.ml_core.configs import CheckpointConfig
-from kostyl.ml_core.dist_utils import is_main_process
+from kostyl.ml.configs import CheckpointConfig
+from kostyl.ml.dist_utils import is_main_process
 from kostyl.utils import setup_logger
 
 

kostyl/{ml_core → ml}/lightning/callbacks/early_stopping.py

@@ -1,6 +1,6 @@
 from lightning.pytorch.callbacks import EarlyStopping
 
-from kostyl.ml_core.configs import EarlyStoppingConfig
+from kostyl.ml.configs import EarlyStoppingConfig
 
 
 def setup_early_stopping_callback(

kostyl/{ml_core → ml}/lightning/callbacks/registry_uploading.py

@@ -7,9 +7,9 @@ from lightning import Trainer
 from lightning.pytorch.callbacks import Callback
 from lightning.pytorch.callbacks import ModelCheckpoint
 
-from kostyl.ml_core.clearml.logging_utils import find_version_in_tags
-from kostyl.ml_core.clearml.logging_utils import increment_version
-from kostyl.ml_core.lightning import KostylLightningModule
+from kostyl.ml.clearml.logging_utils import find_version_in_tags
+from kostyl.ml.clearml.logging_utils import increment_version
+from kostyl.ml.lightning import KostylLightningModule
 from kostyl.utils.logging import setup_logger
 
 

kostyl/{ml_core → ml}/lightning/extenstions/custom_module.py

@@ -15,9 +15,8 @@ from torchmetrics import MetricCollection
 from transformers import PretrainedConfig
 from transformers import PreTrainedModel
 
-from kostyl.ml_core.configs import HyperparamsConfig
-from kostyl.ml_core.metrics_formatting import apply_suffix
-from kostyl.ml_core.schedulers.base import BaseScheduler
+from kostyl.ml.metrics_formatting import apply_suffix
+from kostyl.ml.schedulers.base import BaseScheduler
 from kostyl.utils import setup_logger
 
 
@@ -28,7 +27,6 @@ class KostylLightningModule(L.LightningModule):
     """Custom PyTorch Lightning Module with logging, checkpointing, and distributed training utilities."""
 
     model: PreTrainedModel | nn.Module | None
-    hyperparams: HyperparamsConfig
 
     def get_process_group(self) -> ProcessGroup | None:
         """
@@ -70,6 +68,11 @@ class KostylLightningModule(L.LightningModule):
             return model.config  # type: ignore
         return None
 
+    @property
+    def grad_clip_val(self) -> float | None:
+        """Returns the gradient clipping value from hyperparameters if set."""
+        raise NotImplementedError
+
     @override
     def on_save_checkpoint(self, checkpoint: dict[str, Any]) -> None:
         model = self.get_model()
@@ -93,20 +96,16 @@ class KostylLightningModule(L.LightningModule):
     def on_before_optimizer_step(self, optimizer) -> None:
         if self.model is None:
             raise ValueError("Model must be configured before optimizer step.")
-        if not hasattr(self, "hyperparams"):
-            logger.warning_once("cannot clip gradients, hyperparams attr missing")
-            return
-        if self.hyperparams.grad_clip_val is None:
+
+        grad_clip_val = self.grad_clip_val
+        if grad_clip_val is None:
             return
 
         if not isinstance(self.trainer.strategy, FSDPStrategy):
-            norm = torch.nn.utils.clip_grad_norm_(
-                self.parameters(), self.hyperparams.grad_clip_val
-            )
+            norm = torch.nn.utils.clip_grad_norm_(self.parameters(), grad_clip_val)
         else:
             module: FSDP = self.trainer.strategy.model  # type: ignore
-            norm = module.clip_grad_norm_(self.hyperparams.grad_clip_val)
-
+            norm = module.clip_grad_norm_(grad_clip_val)
         self.log(
             "grad_norm",
             norm,

kostyl/{ml_core → ml}/lightning/extenstions/pretrained_model.py

@@ -1,4 +1,5 @@
 from pathlib import Path
+from typing import Any
 from typing import cast
 
 import torch
@@ -28,6 +29,7 @@ class LightningCheckpointLoaderMixin(PreTrainedModel):
         config_key: str = "config",
         weights_prefix: str = "model.",
         should_log_incompatible_keys: bool = True,
+        **kwargs: Any,
     ) -> TModelInstance:
         """
         Load a model from a Lightning checkpoint file.
@@ -48,6 +50,7 @@ class LightningCheckpointLoaderMixin(PreTrainedModel):
             weights_prefix (str, optional): Prefix to strip from state dict keys. Defaults to "model.".
                 If not empty and doesn't end with ".", a "." is appended.
             should_log_incompatible_keys (bool, optional): Whether to log incompatible keys. Defaults to True.
+            **kwargs: Additional keyword arguments to pass to the model loading method.
 
         Returns:
             TModelInstance: The loaded model instance.
@@ -75,10 +78,17 @@ class LightningCheckpointLoaderMixin(PreTrainedModel):
         )
 
         config_cls = cast(PretrainedConfig, type(cls.config_class))
-        config = config_cls.from_dict(checkpoint_dict[config_key])
+        config_dict = checkpoint_dict[config_key]
+        config_dict.update(kwargs)
+        config = config_cls.from_dict(config_dict)
+
+        kwargs_for_model = {}
+        for key in kwargs:
+            if not hasattr(config, key):
+                kwargs_for_model[key] = kwargs[key]
 
         with torch.device("meta"):
-            model = cls(config)
+            model = cls(config, **kwargs_for_model)
 
             if "peft_config" in checkpoint_dict:
                 if PeftConfig is None:

kostyl/ml/lightning/loggers/__init__.py

@@ -0,0 +1,4 @@
+from .tb_logger import setup_tb_logger
+
+
+__all__ = ["setup_tb_logger"]

kostyl/{ml_core → ml}/lightning/loggers/tb_logger.py

@@ -3,7 +3,7 @@ from shutil import rmtree
 
 from lightning.pytorch.loggers import TensorBoardLogger
 
-from kostyl.ml_core.dist_utils import is_main_process
+from kostyl.ml.dist_utils import is_main_process
 from kostyl.utils.logging import setup_logger
 
 

{kostyl_toolkit-0.1.10.dist-info → kostyl_toolkit-0.1.11.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: kostyl-toolkit
-Version: 0.1.10
+Version: 0.1.11
 Summary: Kickass Orchestration System for Training, Yielding & Logging 
 Requires-Dist: case-converter>=1.2.0
 Requires-Dist: loguru>=0.7.3

kostyl_toolkit-0.1.11.dist-info/RECORD

@@ -0,0 +1,36 @@
+kostyl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+kostyl/ml/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+kostyl/ml/clearml/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+kostyl/ml/clearml/config_mixin.py,sha256=Wkz92BLL9y9aDk8qPtla6BrtZegQe7aw40G-oNkBIGs,3363
+kostyl/ml/clearml/dataset_utils.py,sha256=eij_sr2KDhm8GxEbVbK8aBjPsuVvLl9-PIGGaKVgXLA,1729
+kostyl/ml/clearml/logging_utils.py,sha256=GBjIIZbH_itd5sj7XpvxjkyZwxxGOpEcQ3BiWaJTyq8,1210
+kostyl/ml/clearml/pulling_utils.py,sha256=FOk9rtwu9PZ4tcIexiK0LvF6MfuuCwpZxtg2oDUWlTw,3526
+kostyl/ml/configs/__init__.py,sha256=pBuEhcumGBNncprqz-t_IGll2XtMmogTn9S6ZPHXCTg,997
+kostyl/ml/configs/base_model.py,sha256=nOuiBIQn5pYDRRq-F4HbfWi5cn8pfaT2kM02VGoo4TE,1622
+kostyl/ml/configs/hyperparams.py,sha256=UqMk2y1LISK7QMPbh3LyICjJ1DnUQhsYh1NDMR6YnR8,3006
+kostyl/ml/configs/training_settings.py,sha256=JznYSXpTRraFEH7dvk4gt2Ym-XpPGlN-ALzklD7kzCw,2848
+kostyl/ml/dist_utils.py,sha256=G8atjzkRbXZZiZh9rdEYBmeXqX26rJdDDovft2n6xiU,3201
+kostyl/ml/lightning/__init__.py,sha256=-F3JAyq8KU1d-nACWryGu8d1CbvWbQ1rXFdeRwfE2X8,175
+kostyl/ml/lightning/callbacks/__init__.py,sha256=Vd-rozY4T9Prr3IMqbliXxj6sC6y9XsovHQqRwzc2HI,297
+kostyl/ml/lightning/callbacks/checkpoint.py,sha256=FooGeeUz6TtoXQglpcK16NWAmSX3fbu6wntRtK3a_Io,1936
+kostyl/ml/lightning/callbacks/early_stopping.py,sha256=D5nyjktCJ9XYAf28-kgXG8jORvXLl1N3nbDQnvValPM,615
+kostyl/ml/lightning/callbacks/registry_uploading.py,sha256=um-fHQUp-TSD_76rVZoNV2IMi05RFjKajGvMV7SAb6s,4691
+kostyl/ml/lightning/extenstions/__init__.py,sha256=OY6QGv1agYgqqKf1xJBrxgp_i8FunVfPzYezfaRrGXU,182
+kostyl/ml/lightning/extenstions/custom_module.py,sha256=N4kPObp4jqrCJscPKzJib3AkvcQQ3AHal7A1Cm3ZobA,6201
+kostyl/ml/lightning/extenstions/pretrained_model.py,sha256=ZOKtrVl095cwvI43wAz-Xdzu4l0v0lHH2mfh4WXwxKQ,5059
+kostyl/ml/lightning/loggers/__init__.py,sha256=e51dszaoJbuzwBkbdugmuDsPldoSO4yaRgmZUg1Bdy0,71
+kostyl/ml/lightning/loggers/tb_logger.py,sha256=I0WQV3SVVIds9kHhAYnN3SXodX64jfTk7_5gH-j3uzA,932
+kostyl/ml/lightning/steps_estimation.py,sha256=fTZ0IrUEZV3H6VYlx4GYn56oco56mMiB7FO9F0Z7qc4,1511
+kostyl/ml/metrics_formatting.py,sha256=w0rTz61z0Um_d2pomYLvcQFcZX_C-KolZcIPRsa1efE,1421
+kostyl/ml/params_groups.py,sha256=nUyw5d06Pvy9QPiYtZzLYR87xwXqJLxbHthgQH8oSCM,3583
+kostyl/ml/schedulers/__init__.py,sha256=bxXbsU_WYnVbhvNNnuI7cOAh2Axz7D25TaleBTZhYfc,197
+kostyl/ml/schedulers/base.py,sha256=9M2iOoOVSRojR_liPX1qo3Nn4iMXSM5ZJuAFWZTulUk,1327
+kostyl/ml/schedulers/composite.py,sha256=ee4xlMDMMtjKPkbTF2ue9GTr9DuGCGjZWf11mHbi6aE,2387
+kostyl/ml/schedulers/cosine.py,sha256=jufULVHn_L_ZZEc3ZTG3QCY_pc0jlAMH5Aw496T31jo,8203
+kostyl/utils/__init__.py,sha256=hkpmB6c5pr4Ti5BshOROebb7cvjDZfNCw83qZ_FFKMM,240
+kostyl/utils/dict_manipulations.py,sha256=e3vBicID74nYP8lHkVTQc4-IQwoJimrbFELy5uSF6Gk,1073
+kostyl/utils/fs.py,sha256=gAQNIU4R_2DhwjgzOS8BOMe0gZymtY1eZwmdgOdDgqo,510
+kostyl/utils/logging.py,sha256=3MvfDPArZhwakHu5nMlp_LpOsWg0E0SP26y41clsBtA,5232
+kostyl_toolkit-0.1.11.dist-info/WHEEL,sha256=3id4o64OvRm9dUknh3mMJNcfoTRK08ua5cU6DFyVy-4,79
+kostyl_toolkit-0.1.11.dist-info/METADATA,sha256=ImcjKFPWoQznLjTGCKjcvmf_Nn2F6NXNWzfQNPXPFDU,4269
+kostyl_toolkit-0.1.11.dist-info/RECORD,,

kostyl_toolkit-0.1.10.dist-info/RECORD

@@ -1,36 +0,0 @@
-kostyl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-kostyl/ml_core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-kostyl/ml_core/clearml/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-kostyl/ml_core/clearml/config_mixin.py,sha256=sN24RPs8kWW1iXpIr_6w1bfEZZSXIV_i_RXypqzjwXg,3368
-kostyl/ml_core/clearml/dataset_utils.py,sha256=eij_sr2KDhm8GxEbVbK8aBjPsuVvLl9-PIGGaKVgXLA,1729
-kostyl/ml_core/clearml/logging_utils.py,sha256=GBjIIZbH_itd5sj7XpvxjkyZwxxGOpEcQ3BiWaJTyq8,1210
-kostyl/ml_core/clearml/pulling_utils.py,sha256=mXL17sQxRaSLJKmLIjm7hCsZ7OjzLIKxrtebowBOtvQ,3379
-kostyl/ml_core/configs/__init__.py,sha256=pBuEhcumGBNncprqz-t_IGll2XtMmogTn9S6ZPHXCTg,997
-kostyl/ml_core/configs/base_model.py,sha256=nOuiBIQn5pYDRRq-F4HbfWi5cn8pfaT2kM02VGoo4TE,1622
-kostyl/ml_core/configs/hyperparams.py,sha256=u-7FIM-cD3nz9Sycuvg7r0Vdiu4pduafiCoAxq8JK0s,3011
-kostyl/ml_core/configs/training_settings.py,sha256=35EgIfqfoLAXtxwJ6XiWOalOcHDxrlsoBgjdtLbUI5k,2853
-kostyl/ml_core/dist_utils.py,sha256=G8atjzkRbXZZiZh9rdEYBmeXqX26rJdDDovft2n6xiU,3201
-kostyl/ml_core/lightning/__init__.py,sha256=-F3JAyq8KU1d-nACWryGu8d1CbvWbQ1rXFdeRwfE2X8,175
-kostyl/ml_core/lightning/callbacks/__init__.py,sha256=Vd-rozY4T9Prr3IMqbliXxj6sC6y9XsovHQqRwzc2HI,297
-kostyl/ml_core/lightning/callbacks/checkpoint.py,sha256=RgkNNmsbAz9fdMYGlEgn9Qs_DF8LiuY7Bp1Hu4ZW98s,1946
-kostyl/ml_core/lightning/callbacks/early_stopping.py,sha256=nEj3OkMNJkpQzR6pt0Z0kvHDND6OWzWHw_aCZRmGE1s,620
-kostyl/ml_core/lightning/callbacks/registry_uploading.py,sha256=1aqT38FVOMQo4JphXcyjyK3ZY6A6HF1JBOsKqYNXar8,4706
-kostyl/ml_core/lightning/extenstions/__init__.py,sha256=OY6QGv1agYgqqKf1xJBrxgp_i8FunVfPzYezfaRrGXU,182
-kostyl/ml_core/lightning/extenstions/custom_module.py,sha256=M8kAvKSlrHT2-MPmzRUfbLhNOOMQzhwnN8FazZzkYg0,6311
-kostyl/ml_core/lightning/extenstions/pretrained_model.py,sha256=4Ng4xjt3KFWKT0UThC_9vM5Y95AhBdzrU4y4We1gemc,4685
-kostyl/ml_core/lightning/loggers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-kostyl/ml_core/lightning/loggers/tb_logger.py,sha256=Zh9n-lLu-bXMld-FIUO3lJfCyDf0IQFhS3JVShDJmvg,937
-kostyl/ml_core/lightning/steps_estimation.py,sha256=fTZ0IrUEZV3H6VYlx4GYn56oco56mMiB7FO9F0Z7qc4,1511
-kostyl/ml_core/metrics_formatting.py,sha256=w0rTz61z0Um_d2pomYLvcQFcZX_C-KolZcIPRsa1efE,1421
-kostyl/ml_core/params_groups.py,sha256=nUyw5d06Pvy9QPiYtZzLYR87xwXqJLxbHthgQH8oSCM,3583
-kostyl/ml_core/schedulers/__init__.py,sha256=bxXbsU_WYnVbhvNNnuI7cOAh2Axz7D25TaleBTZhYfc,197
-kostyl/ml_core/schedulers/base.py,sha256=9M2iOoOVSRojR_liPX1qo3Nn4iMXSM5ZJuAFWZTulUk,1327
-kostyl/ml_core/schedulers/composite.py,sha256=ee4xlMDMMtjKPkbTF2ue9GTr9DuGCGjZWf11mHbi6aE,2387
-kostyl/ml_core/schedulers/cosine.py,sha256=jufULVHn_L_ZZEc3ZTG3QCY_pc0jlAMH5Aw496T31jo,8203
-kostyl/utils/__init__.py,sha256=hkpmB6c5pr4Ti5BshOROebb7cvjDZfNCw83qZ_FFKMM,240
-kostyl/utils/dict_manipulations.py,sha256=e3vBicID74nYP8lHkVTQc4-IQwoJimrbFELy5uSF6Gk,1073
-kostyl/utils/fs.py,sha256=gAQNIU4R_2DhwjgzOS8BOMe0gZymtY1eZwmdgOdDgqo,510
-kostyl/utils/logging.py,sha256=3MvfDPArZhwakHu5nMlp_LpOsWg0E0SP26y41clsBtA,5232
-kostyl_toolkit-0.1.10.dist-info/WHEEL,sha256=3id4o64OvRm9dUknh3mMJNcfoTRK08ua5cU6DFyVy-4,79
-kostyl_toolkit-0.1.10.dist-info/METADATA,sha256=WRGWDJWvhk-VsChASkzG5vlzPLj89nv37gOVh9_FgPo,4269
-kostyl_toolkit-0.1.10.dist-info/RECORD,,