truefoundry 0.11.3rc2__tar.gz → 0.11.5rc1__tar.gz

{truefoundry-0.11.3rc2 → truefoundry-0.11.5rc1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truefoundry
-Version: 0.11.3rc2
+Version: 0.11.5rc1
 Summary: TrueFoundry CLI
 Author-email: TrueFoundry Team <abhishek@truefoundry.com>
 Requires-Python: <3.14,>=3.8.1
@@ -30,7 +30,7 @@ Requires-Dist: requirements-parser<0.12.0,>=0.11.0
 Requires-Dist: rich-click<2.0.0,>=1.2.1
 Requires-Dist: rich<14.0.0,>=13.7.1
 Requires-Dist: tqdm<5.0.0,>=4.0.0
-Requires-Dist: truefoundry-sdk<0.2.0,>=0.1.9
+Requires-Dist: truefoundry-sdk<0.2.0,>=0.1.10
 Requires-Dist: typing-extensions>=4.0
 Requires-Dist: urllib3<3,>=1.26.18
 Requires-Dist: yq<4.0.0,>=3.1.0

{truefoundry-0.11.3rc2 → truefoundry-0.11.5rc1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "truefoundry"
-version = "0.11.3rc2"  # This is a placeholder version, it will be replaced when releasing
+version = "0.11.5rc1"  # This is a placeholder version, it will be replaced when releasing
 requires-python = ">=3.8.1,<3.14"
 description = "TrueFoundry CLI"
 authors = [{ name = "TrueFoundry Team", email = "abhishek@truefoundry.com" }]
@@ -31,7 +31,7 @@ dependencies = [
   "rich-click>=1.2.1,<2.0.0",
   "rich>=13.7.1,<14.0.0",
   "tqdm>=4.0.0,<5.0.0",
-  "truefoundry-sdk>=0.1.9,<0.2.0",
+  "truefoundry-sdk>=0.1.10,<0.2.0",
   "typing-extensions>=4.0",
   "urllib3>=1.26.18,<3",
   "yq>=3.1.0,<4.0.0",
@@ -58,22 +58,25 @@ truefoundry = "truefoundry.cli.__main__:main"
 
 [dependency-groups]
 dev = [
-  "Pillow>=9.1.1,<11.0.0",
+  "accelerate>=1.0.0,<2.0.0",
+  "datasets>=3.0.0,<5.0.0",
   "ipykernel>=6.16.0",
-  "ipywidgets>=8.0.0",
   "ipython>=7.10.0",
+  "ipywidgets>=8.0.0",
   "matplotlib>=3.0.0,<4.0.0",
   "mypy>=1.14.1",
   "omegaconf==2.3",
-  "plotly>=5.8.2,<6.0.0",
   "pandas>=1.0.0,<3.0.0; python_version < '3.10'",
   "pandas>=1.4.0,<3.0.0; python_version >= '3.10'",
+  "Pillow>=9.1.1,<11.0.0",
+  "plotly>=5.8.2,<6.0.0",
   "pre-commit>=3.5.0,<4.0.0",
-  "pytest>=8.2.2,<9.0.0",
   "pytest-env>=1.1.5,<2.0.0",
+  "pytest>=8.2.2,<9.0.0",
   "ruff>=0.9.4",
-  "seaborn>=0.13.0,<0.14.0",
   "scikit-learn>=1.3.0,<2.0.0",
+  "seaborn>=0.13.0,<0.14.0",
+  "transformers>=4.0.0,<5.0.0",
   "xgboost>=2.1.3,<3.0.0",
 
   # typing stubs

{truefoundry-0.11.3rc2 → truefoundry-0.11.5rc1}/truefoundry/deploy/_autogen/models.py

@@ -1,6 +1,6 @@
 # generated by datamodel-codegen:
 #   filename:  application.json
-#   timestamp: 2025-07-28T14:11:36+00:00
+#   timestamp: 2025-08-06T14:39:55+00:00
 
 from __future__ import annotations
 
@@ -1088,11 +1088,11 @@ class TrueFoundryInteractiveLogin(BaseModel):
 
 
 class VolumeBrowser(BaseModel):
-    username: constr(regex=r"^[a-z][a-z0-9]{1,8}[a-z0-9]$") = Field(
-        ..., description="Username for logging in the volume browser."
+    username: Optional[constr(regex=r"^[a-z][a-z0-9]{1,8}[a-z0-9]$")] = Field(
+        None, description="Username for logging in the volume browser."
     )
-    password_secret_fqn: constr(regex=r"^tfy-secret:\/\/.+:.+:.+$") = Field(
-        ...,
+    password_secret_fqn: Optional[constr(regex=r"^tfy-secret:\/\/.+:.+:.+$")] = Field(
+        None,
         description="TFY Secret containing the password for logging in the volume browser.",
     )
     endpoint: Endpoint
@@ -1282,11 +1282,11 @@ class HealthProbe(BaseModel):
         1,
         description="Time to wait for a response from the endpoint before considering it down",
     )
-    success_threshold: conint(ge=1, le=100) = Field(
+    success_threshold: conint(ge=1, le=5000) = Field(
         1,
         description="Number of successful responses from the endpoint before container is considered healthy",
     )
-    failure_threshold: conint(ge=1, le=100) = Field(
+    failure_threshold: conint(ge=1, le=5000) = Field(
         3,
         description="Number of consecutive failures before the container is considered down",
     )
@@ -1593,6 +1593,7 @@ class BaseService(BaseModel):
     kustomize: Optional[Kustomize] = None
     liveness_probe: Optional[HealthProbe] = None
     readiness_probe: Optional[HealthProbe] = None
+    startup_probe: Optional[HealthProbe] = None
     workspace_fqn: Optional[str] = Field(
         None, description="Fully qualified name of the workspace"
     )

{truefoundry-0.11.3rc2 → truefoundry-0.11.5rc1}/truefoundry/ml/artifact/truefoundry_artifact_repo.py

@@ -654,7 +654,7 @@ class MlFoundryArtifactsRepository:
                 artifact_identifier=self.artifact_identifier, paths=[remote_file_path]
             )[0]
 
-        if progress_bar is None or not progress_bar.disable:
+        if progress_bar is None or progress_bar.disable:
             logger.info("Downloading %s to %s", remote_file_path, local_path)
 
         if progress_bar is not None:

truefoundry-0.11.5rc1/truefoundry/ml/integrations/huggingface/trainer_callback.py

@@ -0,0 +1,198 @@
+import logging
+import math
+import os
+from typing import TYPE_CHECKING, Any, Dict, Optional
+
+import numpy as np
+
+from truefoundry import ml
+
+try:
+    from transformers.integrations.integration_utils import rewrite_logs
+    from transformers.trainer_callback import TrainerCallback
+except ImportError as e:
+    raise ImportError(
+        "Importing this module requires `transformers` to be installed"
+    ) from e
+
+if TYPE_CHECKING:
+    from transformers.trainer_callback import TrainerControl, TrainerState
+    from transformers.training_args import TrainingArguments
+
+    from truefoundry.ml import MlFoundryRun
+
+logger = logging.getLogger(__name__)
+
+
+class TrueFoundryMLCallback(TrainerCallback):
+    def __init__(
+        self,
+        run: "MlFoundryRun",
+        log_checkpoints: bool = True,
+        checkpoint_artifact_name: Optional[str] = None,
+        auto_end_run_on_train_end: bool = False,
+    ):
+        """
+        Args:
+            run: The run entity to log metrics to.
+            log_checkpoints: Whether to log checkpoints or not, defaults to True.
+            checkpoint_artifact_name: The name of the artifact to log checkpoints to, required if log_checkpoints is True.
+            auto_end_run_on_train_end: Whether to end the run automatically when training ends, defaults to False.
+
+            Usage:
+                from transformers import Trainer
+                from truefoundry.ml.integrations.huggingface.trainer_callback import TrueFoundryMLCallback
+                from truefoundry.ml import get_client
+
+                client = get_client()
+                run = client.create_run(ml_repo="my-ml-repo", run_name="my-run", auto_end=False)
+
+                callback = TrueFoundryMLCallback(
+                    run=run,
+                    log_checkpoints=True,
+                    checkpoint_artifact_name="my-checkpoint",
+                    auto_end_run_on_train_end=True,
+                )
+
+                trainer = Trainer(
+                    ...,
+                    callbacks=[callback]
+                )
+        """
+        self._run = run
+        self._log_checkpoints = log_checkpoints
+        if self._log_checkpoints and not checkpoint_artifact_name:
+            raise ValueError(
+                "`checkpoint_artifact_name` is required when `log_checkpoints` is True"
+            )
+        self._checkpoint_artifact_name = checkpoint_artifact_name
+        self._auto_end_run_on_train_end = auto_end_run_on_train_end
+
+    @classmethod
+    def with_managed_run(
+        cls,
+        ml_repo: str,
+        run_name: Optional[str] = None,
+        log_checkpoints: bool = True,
+        checkpoint_artifact_name: Optional[str] = None,
+        auto_end_run_on_train_end: bool = True,
+    ) -> "TrueFoundryMLCallback":
+        """
+        Args:
+            ml_repo: The name of the ML Repository to log metrics and data to.
+            run_name: The name of the run, if not provided, a random name will be generated.
+            log_checkpoints: Whether to log checkpoints or not, defaults to True.
+            checkpoint_artifact_name: The name of the artifact to log checkpoints to, required if log_checkpoints is True.
+            auto_end_run_on_train_end: Whether to end the run automatically when training ends, defaults to True.
+
+        Usage:
+            from transformers import Trainer
+            from truefoundry.ml.integrations.huggingface.trainer_callback import TrueFoundryMLCallback
+
+            callback = TrueFoundryMLCallback.with_managed_run(
+                ml_repo="my-ml-repo",
+                run_name="my-run",
+                log_checkpoints=True,
+                checkpoint_artifact_name="my-checkpoint",
+                auto_end_run_on_train_end=True,
+            )
+            trainer = Trainer(
+                ...,
+                callbacks=[callback]
+            )
+        """
+        run = ml.get_client().create_run(
+            ml_repo=ml_repo, run_name=run_name, auto_end=False
+        )
+        return cls(
+            run=run,
+            log_checkpoints=log_checkpoints,
+            checkpoint_artifact_name=checkpoint_artifact_name,
+            auto_end_run_on_train_end=auto_end_run_on_train_end,
+        )
+
+    def _drop_non_finite_values(self, dct: Dict[str, Any]) -> Dict[str, Any]:
+        sanitized = {}
+        for k, v in dct.items():
+            if isinstance(v, (int, float, np.integer, np.floating)) and math.isfinite(
+                v
+            ):
+                sanitized[k] = v
+            else:
+                logger.warning(
+                    f'Trainer is attempting to log a value of "{v}" of'
+                    f' type {type(v)} for key "{k}" as a metric.'
+                    " Mlfoundry's log_metric() only accepts finite float and"
+                    " int types so we dropped this attribute."
+                )
+        return sanitized
+
+    @property
+    def run(self) -> "MlFoundryRun":
+        return self._run
+
+    # noinspection PyMethodOverriding
+    def on_log(
+        self,
+        args: "TrainingArguments",
+        state: "TrainerState",
+        control: "TrainerControl",
+        logs: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ):
+        logs = logs or {}
+        if not state.is_world_process_zero:
+            return
+
+        metrics = self._drop_non_finite_values(logs)
+        self._run.log_metrics(rewrite_logs(metrics), step=state.global_step)
+
+    def on_save(
+        self,
+        args: "TrainingArguments",
+        state: "TrainerState",
+        control: "TrainerControl",
+        **kwargs,
+    ):
+        if not state.is_world_process_zero:
+            return
+
+        if not self._log_checkpoints:
+            return
+
+        if not self._checkpoint_artifact_name:
+            return
+
+        ckpt_dir = f"checkpoint-{state.global_step}"
+        artifact_path = os.path.join(args.output_dir, ckpt_dir)
+        description = None
+        _job_name = os.getenv("TFY_INTERNAL_COMPONENT_NAME")
+        _job_run_name = os.getenv("TFY_INTERNAL_JOB_RUN_NAME")
+        if _job_name:
+            description = f"Checkpoint from job={_job_name} run={_job_run_name}"
+        logger.info(f"Uploading checkpoint {ckpt_dir} ...")
+        metadata = {}
+        for log in state.log_history:
+            if isinstance(log, dict) and log.get("step") == state.global_step:
+                metadata = log.copy()
+        metadata = self._drop_non_finite_values(metadata)
+        self._run.log_artifact(
+            name=self._checkpoint_artifact_name,
+            artifact_paths=[(artifact_path, None)],
+            metadata=metadata,
+            step=state.global_step,
+            description=description,
+        )
+
+    def on_train_end(
+        self,
+        args: "TrainingArguments",
+        state: "TrainerState",
+        control: "TrainerControl",
+        **kwargs,
+    ):
+        """
+        Event called at the end of training.
+        """
+        if self._auto_end_run_on_train_end:
+            self._run.end()