mostlyai-engine 1.1.2__tar.gz → 1.1.4__tar.gz

{mostlyai_engine-1.1.2 → mostlyai_engine-1.1.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mostlyai-engine
-Version: 1.1.2
+Version: 1.1.4
 Summary: Synthetic Data Engine
 Project-URL: homepage, https://github.com/mostly-ai/mostlyai-engine
 Project-URL: repository, https://github.com/mostly-ai/mostlyai-engine
@@ -24,15 +24,16 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Software Development :: Libraries
 Classifier: Typing :: Typed
 Requires-Python: >=3.10
-Requires-Dist: accelerate<0.32,>=0.31.0
-Requires-Dist: datasets<3,>=2.20.0
+Requires-Dist: accelerate>=0.31.0
+Requires-Dist: datasets<4,>=3.0.0
 Requires-Dist: formatron<0.5,>=0.4.11
+Requires-Dist: huggingface-hub<0.29,>=0.28.0
 Requires-Dist: joblib<2,>=1.3.0
-Requires-Dist: json-repair<0.31,>=0.30.0
+Requires-Dist: json-repair>=0.30.0
 Requires-Dist: numpy<2,>=1.26.3
-Requires-Dist: opacus==1.5.2
+Requires-Dist: opacus<2,>=1.5.2
 Requires-Dist: pandas~=2.2.0
-Requires-Dist: peft<0.12,>=0.11.1
+Requires-Dist: peft>=0.12.0
 Requires-Dist: psutil<6,>=5.9.5
 Requires-Dist: pyarrow>=16.0.0
 Requires-Dist: rich<14,>=13.9.4
@@ -192,4 +193,5 @@ engine.generate(                      # use model to generate synthetic samples
     workspace_dir=ws,
     sample_size=10,
 )
+pd.read_parquet(ws / "SyntheticData") # load synthetic data
 ```

{mostlyai_engine-1.1.2 → mostlyai_engine-1.1.4}/README.md

@@ -141,4 +141,5 @@ engine.generate(                      # use model to generate synthetic samples
     workspace_dir=ws,
     sample_size=10,
 )
+pd.read_parquet(ws / "SyntheticData") # load synthetic data
 ```

{mostlyai_engine-1.1.2 → mostlyai_engine-1.1.4}/mostlyai/engine/__init__.py

@@ -22,7 +22,7 @@ from mostlyai.engine.generation import generate
 
 
 __all__ = ["split", "analyze", "encode", "train", "generate", "init_logging"]
-__version__ = "1.1.2"
+__version__ = "1.1.4"
 
 # suppress specific warning related to os.fork() in multi-threaded processes
 warnings.filterwarnings("ignore", category=DeprecationWarning, message=".*multi-threaded.*fork.*")

{mostlyai_engine-1.1.2 → mostlyai_engine-1.1.4}/mostlyai/engine/_language/training.py

@@ -679,7 +679,8 @@ def train(
                 do_stop = True
 
             # check for max_training_time
-            if (total_time_init + time.time() - start_trn_time) > max_training_time:
+            total_training_time = total_time_init + time.time() - start_trn_time
+            if total_training_time > max_training_time:
                 do_stop = True
 
         # no checkpoint is saved yet because the training stopped before the first epoch ended
@@ -691,8 +692,13 @@ def train(
                 lr_scheduler=lr_scheduler,
                 dp_accountant=privacy_engine.accountant if with_dp else None,
             )
-            with forward_ctx_mgr:
-                val_loss = _calculate_val_loss(model=model, val_dataloader=val_dataloader)
+            if total_training_time > max_training_time:
+                _LOG.info("skip validation loss calculation due to time-capped early stopping")
+                val_loss = None
+            else:
+                _LOG.info("calculate validation loss")
+                with forward_ctx_mgr:
+                    val_loss = _calculate_val_loss(model=model, val_dataloader=val_dataloader)
             dp_epsilon = privacy_engine.get_epsilon(dp_delta) if with_dp else None
             # send a final message to inform how far we've progressed
             progress_message = ProgressMessage(
@@ -702,7 +708,7 @@ def train(
                 samples=samples,
                 trn_loss=None,
                 val_loss=val_loss,
-                total_time=total_time_init + time.time() - start_trn_time,
+                total_time=total_training_time,
                 learn_rate=current_lr,
                 dp_eps=dp_epsilon,
                 dp_delta=dp_delta,

{mostlyai_engine-1.1.2 → mostlyai_engine-1.1.4}/mostlyai/engine/_tabular/training.py

@@ -822,7 +822,8 @@ def train(
                 do_stop = True
 
             # check for max_training_time
-            if (total_time_init + time.time() - start_trn_time) > max_training_time:
+            total_training_time = total_time_init + time.time() - start_trn_time
+            if total_training_time > max_training_time:
                 do_stop = True
 
         # no checkpoint is saved yet because the training stopped before the first epoch ended
@@ -834,7 +835,12 @@ def train(
                 lr_scheduler=lr_scheduler,
                 dp_accountant=privacy_engine.accountant if with_dp else None,
             )
-            val_loss = _calculate_val_loss(model=argn, val_dataloader=val_dataloader)
+            if total_training_time > max_training_time:
+                _LOG.info("skip validation loss calculation due to time-capped early stopping")
+                val_loss = None
+            else:
+                _LOG.info("calculate validation loss")
+                val_loss = _calculate_val_loss(model=argn, val_dataloader=val_dataloader)
             dp_epsilon = privacy_engine.get_epsilon(dp_delta) if with_dp else None
             # send a final message to inform how far we've progressed
             trn_loss = _calculate_average_trn_loss(trn_sample_losses)
@@ -845,7 +851,7 @@ def train(
                 samples=samples,
                 trn_loss=trn_loss,
                 val_loss=val_loss,
-                total_time=total_time_init + time.time() - start_trn_time,
+                total_time=total_training_time,
                 learn_rate=current_lr,
                 dp_eps=dp_epsilon,
                 dp_delta=dp_delta,

{mostlyai_engine-1.1.2 → mostlyai_engine-1.1.4}/mostlyai/engine/logging.py

@@ -20,15 +20,13 @@ _LOG = logging.getLogger(__name__.rsplit(".", 1)[0])  # get the logger with the
 
 def init_logging() -> None:
     """
-    Initialize the logging configuration. Either log to stdout or to a file.
+    Initialize the logging configuration to stdout.
     """
 
-    # log to stdout
-    handler = logging.StreamHandler(stream=sys.stdout)
-    handler.setFormatter(logging.Formatter("[%(asctime)s] %(levelname)-7s: %(message)s"))
-    handler.setLevel(logging.INFO)
-
+    _LOG.propagate = False
     if not _LOG.hasHandlers():
+        handler = logging.StreamHandler(stream=sys.stdout)
+        handler.setFormatter(logging.Formatter("[%(asctime)s] %(levelname)-7s: %(message)s"))
+        handler.setLevel(logging.INFO)
         _LOG.addHandler(handler)
         _LOG.setLevel(logging.INFO)
-        _LOG.propagate = False

{mostlyai_engine-1.1.2 → mostlyai_engine-1.1.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "mostlyai-engine"
-version = "1.1.2"
+version = "1.1.4"
 description = "Synthetic Data Engine"
 authors = [{ name = "MOSTLY AI", email = "dev@mostly.ai" }]
 requires-python = ">=3.10"
@@ -33,12 +33,13 @@ dependencies = [
     "psutil>=5.9.5,<6",  # upgrade when colab psutil is updated
     "tokenizers>=0.21.0,<0.22",
     "transformers>=4.45.2,<5",
-    "datasets>=2.20.0,<3",
-    "accelerate>=0.31.0,<0.32",
-    "peft>=0.11.1,<0.12",
-    "opacus==1.5.2",  # pin as 1.5.3 breaks tests
+    "datasets>=3.0.0,<4",
+    "accelerate>=0.31.0",
+    "peft>=0.12.0",
+    "huggingface-hub>=0.28.0,<0.29",
+    "opacus>=1.5.2,<2",
     "formatron>=0.4.11,<0.5",
-    "json-repair>=0.30.0,<0.31",
+    "json-repair>=0.30.0",
     "rich>=13.9.4,<14",
     "torch>=2.5.1; sys_platform == 'darwin'",
 ]