truefoundry 0.5.0rc7__py3-none-any.whl → 0.5.1__py3-none-any.whl

truefoundry/ml/autogen/entities/artifacts.py

@@ -1,6 +1,6 @@
 # generated by datamodel-codegen:
 #   filename:  artifacts.json
-#   timestamp: 2024-11-28T08:27:16+00:00
+#   timestamp: 2024-12-09T09:04:12+00:00
 
 from __future__ import annotations
 
@@ -52,11 +52,11 @@ class AgentWithFQN(Agent):
 class BaseArtifactVersion(BaseModel):
     description: Optional[constr(max_length=512)] = Field(
         None,
-        description="+label=Description\n+docs=Description of the artifact version",
+        description="+label=Description\n+usage=Description of the artifact or model version\n+docs=Description of the artifact or model version",
     )
     metadata: Dict[str, Any] = Field(
         ...,
-        description="+label=Metadata\n+docs=Metadata for the model version\n+usage=Metadata for the model version\n+uiType=JsonInput",
+        description="+label=Metadata\n+docs=Metadata for the artifact or model version\n+usage=Metadata for the artifact or model version\n+uiType=JsonInput",
     )
 
 
@@ -228,7 +228,7 @@ class ModelConfiguration(BaseModel):
 
 class ModelVersionEnvironment(BaseModel):
     """
-    +label=ModelVersionEnvironment
+    +label=Environment
     """
 
     python_version: Optional[constr(regex=r"^\d+(\.\d+){1,2}([\-\.a-z0-9]+)?$")] = (
@@ -276,32 +276,44 @@ class PyTorchFramework(BaseModel):
     )
 
 
-class SerializationFormat(str, Enum):
+class InferMethodName(str, Enum):
     """
-    +label=Serialization format
-    +usage=Serialization format used for the model
+    +label=Inference Method Name
+    +usage=Name of the method used for inference
     """
 
-    cloudpickle = "cloudpickle"
-    joblib = "joblib"
-    pickle = "pickle"
+    predict = "predict"
+    predict_proba = "predict_proba"
 
 
-class SklearnFramework(BaseModel):
+class SklearnModelSchema(BaseModel):
     """
-    +docs=Scikit-learn framework for the model version
-    +label=Sklearn
+    +label=Sklearn Model Schema
     """
 
-    type: Literal["sklearn"] = Field(
-        ..., description="+label=Type\n+usage=Type of the framework\n+value=sklearn"
+    infer_method_name: InferMethodName = Field(
+        ...,
+        description="+label=Inference Method Name\n+usage=Name of the method used for inference",
     )
-    serialization_format: Optional[SerializationFormat] = Field(
-        None,
-        description="+label=Serialization format\n+usage=Serialization format used for the model",
+    inputs: List[Dict[str, Any]] = Field(
+        ..., description="+label= Input Schema\n+usage=Schema of the input"
+    )
+    outputs: List[Dict[str, Any]] = Field(
+        ..., description="+label= Output Schema\n+usage=Schema of the output"
     )
 
 
+class SklearnSerializationFormat(str, Enum):
+    """
+    +label=Serialization format
+    +usage=Serialization format used for sklearn models
+    """
+
+    cloudpickle = "cloudpickle"
+    joblib = "joblib"
+    pickle = "pickle"
+
+
 class SpaCyFramework(BaseModel):
     """
     +docs=spaCy framework for the model version
@@ -389,11 +401,11 @@ class TransformersFramework(BaseModel):
     )
     pipeline_tag: Optional[str] = Field(
         None,
-        description="+label=Pipeline Tag\n+usage=Pipeline tag\n+docs=Pipeline tag for the framework",
+        description="+label=Pipeline Tag\n+usage=The `pipeline()` task this model can be used with e.g. `text-generation`. See [huggingface docs](https://huggingface.co/docs/transformers/main/en/main_classes/pipelines#transformers.pipeline.task) for all possible values\n+docs=Pipeline tag for the framework",
     )
     base_model: Optional[str] = Field(
         None,
-        description="+label=Base Model\n+usage=Base model\n+docs=Base model Id. If this is a finetuned model, this points to the base model used for finetuning",
+        description="+label=Base Model\n+usage=Base model Id. If this is a finetuned model, this points to the base model used for finetuning\n+docs=Base model Id. If this is a finetuned model, this points to the base model used for finetuning",
     )
 
 
@@ -433,19 +445,33 @@ class UserMessage(BaseModel):
     )
 
 
-class XGBoostFramework(BaseModel):
+class XGBoostModelSchema(BaseModel):
     """
-    +docs=XGBoost framework for the model version
-    +label=XGBoost
+    +label=XGBoost Model Schema
     """
 
-    type: Literal["xgboost"] = Field(
-        ..., description="+label=Type\n+usage=Type of the framework\n+value=xgboost"
+    infer_method_name: Literal["predict"] = Field(
+        ...,
+        description="+label=Inference Method Name\n+usage=Name of the method used for inference",
     )
-    serialization_format: Optional[SerializationFormat] = Field(
-        None,
-        description="+label=Serialization format\n+usage=Serialization format used for the model",
+    inputs: List[Dict[str, Any]] = Field(
+        ..., description="+label= Input Schema\n+usage=Schema of the input"
     )
+    outputs: List[Dict[str, Any]] = Field(
+        ..., description="+label= Output Schema\n+usage=Schema of the output"
+    )
+
+
+class XGBoostSerializationFormat(str, Enum):
+    """
+    +label=Serialization format
+    +usage=Serialization format used for XGBoost models
+    """
+
+    cloudpickle = "cloudpickle"
+    joblib = "joblib"
+    pickle = "pickle"
+    json = "json"
 
 
 class AgentOpenAPITool(BaseModel):
@@ -541,6 +567,47 @@ class ChatPrompt(BasePrompt):
     )
 
 
+class SklearnFramework(BaseModel):
+    """
+    +docs=Scikit-learn framework for the model version
+    +label=Sklearn
+    """
+
+    type: Literal["sklearn"] = Field(
+        ..., description="+label=Type\n+usage=Type of the framework\n+value=sklearn"
+    )
+    model_filepath: Optional[str] = Field(
+        None,
+        description="+label=Model file path\n+usage=Relative path to the model file",
+    )
+    serialization_format: Optional[SklearnSerializationFormat] = None
+    model_schema: Optional[SklearnModelSchema] = None
+
+
+class XGBoostFramework(BaseModel):
+    """
+    +docs=XGBoost framework for the model version
+    +label=XGBoost
+    """
+
+    type: Literal["xgboost"] = Field(
+        ..., description="+label=Type\n+usage=Type of the framework\n+value=xgboost"
+    )
+    serialization_format: Optional[XGBoostSerializationFormat] = None
+    model_filepath: Optional[str] = Field(
+        None,
+        description="+label=Model file path\n+usage=Relative path to the model file",
+    )
+    model_schema: Optional[XGBoostModelSchema] = None
+
+
+class AgentApp(BaseModel):
+    type: Literal["agent-app"] = Field(..., description="+value=agent-app")
+    tools: List[AgentOpenAPIToolWithFQN]
+    agents: List[AgentWithFQN]
+    root_agent: constr(min_length=1)
+
+
 class ModelVersion(BaseArtifactVersion):
     type: Literal["model-version"] = Field(
         ..., description='+label=Type\n+usage=Model Version\n+value="model-version"'
@@ -571,17 +638,6 @@ class ModelVersion(BaseArtifactVersion):
     )
     environment: Optional[ModelVersionEnvironment] = None
     step: conint(ge=0) = Field(0, description="+label=Step")
-    model_schema: Optional[Dict[str, Any]] = Field(
-        None,
-        description="+label=Model Schema\n+usage=Schema of the model\n+uiType=Hidden",
-    )
-
-
-class AgentApp(BaseModel):
-    type: Literal["agent-app"] = Field(..., description="+value=agent-app")
-    tools: List[AgentOpenAPIToolWithFQN]
-    agents: List[AgentWithFQN]
-    root_agent: constr(min_length=1)
 
 
 class VersionedArtifactType(BaseModel):

truefoundry/ml/autogen/models/signature.py

@@ -1,13 +1,16 @@
 from dataclasses import dataclass, is_dataclass
-from typing import Any, Dict, Optional, Union
+from typing import TYPE_CHECKING, Any, Dict, Optional, Union
 
 import numpy as np
-import pandas as pd
+
+if TYPE_CHECKING:
+    import pandas as pd
+
 
 from .schema import ParamSchema, Schema, convert_dataclass_to_schema
 from .utils import infer_param_schema, infer_schema
 
-MlflowInferableDataset = Union[pd.DataFrame, np.ndarray, Dict[str, np.ndarray]]
+MlflowInferableDataset = Union["pd.DataFrame", "np.ndarray", Dict[str, "np.ndarray"]]
 
 
 class ModelSignature:

truefoundry/ml/autogen/models/utils.py

@@ -4,7 +4,12 @@ from collections import defaultdict
 from typing import Any, Dict, List, Optional, Union
 
 import numpy as np
-import pandas as pd
+
+try:
+    import pandas as pd
+except ImportError:
+    pd = None
+
 
 from .exceptions import MlflowException
 from .schema import (
@@ -330,7 +335,7 @@ def infer_schema(data: Any) -> Schema:  # noqa: C901
                 ]
             )
     # pandas.Series
-    elif isinstance(data, pd.Series):
+    elif pd and isinstance(data, pd.Series):
         name = getattr(data, "name", None)
         schema = Schema(
             [
@@ -342,7 +347,7 @@ def infer_schema(data: Any) -> Schema:  # noqa: C901
             ]
         )
     # pandas.DataFrame
-    elif isinstance(data, pd.DataFrame):
+    elif pd and isinstance(data, pd.DataFrame):
         schema = Schema(
             [
                 ColSpec(
@@ -473,13 +478,13 @@ def _is_none_or_nan(x):
 
 
 def _infer_required(col) -> bool:
-    if isinstance(col, (list, pd.Series)):
+    if pd and isinstance(col, (list, pd.Series)):
         return not any(_is_none_or_nan(x) for x in col)
     return not _is_none_or_nan(col)
 
 
-def _infer_pandas_column(col: pd.Series) -> DataType:
-    if not isinstance(col, pd.Series):
+def _infer_pandas_column(col: "pd.Series") -> DataType:
+    if pd and not isinstance(col, pd.Series):
         raise TypeError(f"Expected pandas.Series, got '{type(col)}'.")
     if len(col.values.shape) > 1:
         raise MlflowException(f"Expected 1d array, got array with shape {col.shape}")
@@ -496,7 +501,7 @@ def _infer_pandas_column(col: pd.Series) -> DataType:
             # For backwards compatibility, we fall back to string
             # if the provided array is of string type
             # This is for diviner test where df field is ('key2', 'key1', 'key0')
-            if pd.api.types.is_string_dtype(col):
+            if pd and pd.api.types.is_string_dtype(col):
                 return DataType.string
             raise MlflowException(
                 f"Failed to infer schema for pandas.Series {col}. Error: {e}"

truefoundry/ml/cli/commands/model_init.py

@@ -0,0 +1,97 @@
+import os
+from typing import Optional
+
+import rich_click as click
+
+from truefoundry.deploy.cli.console import console
+from truefoundry.deploy.cli.const import COMMAND_CLS
+from truefoundry.deploy.cli.util import handle_exception_wrapper
+from truefoundry.ml.autogen.client.models import ModelServer
+from truefoundry.ml.cli.utils import (
+    AppName,
+    NonEmptyString,
+)
+from truefoundry.ml.mlfoundry_api import get_client
+
+
+@click.command(
+    name="model",
+    cls=COMMAND_CLS,
+    help="Generating application code for the specified model version.",
+)
+@click.option(
+    "--name",
+    required=True,
+    type=AppName(),
+    help="Name for the model server deployment",
+    show_default=True,
+)
+@click.option(
+    "--model-version-fqn",
+    "--model_version_fqn",
+    type=NonEmptyString(),
+    required=True,
+    show_default=True,
+    help="Fully Qualified Name (FQN) of the model version to deploy, e.g., 'model:tenant_name/my-model/linear-regression:2'",
+)
+@click.option(
+    "-w",
+    "--workspace-fqn",
+    "--workspace_fqn",
+    type=NonEmptyString(),
+    required=True,
+    show_default=True,
+    help="Fully Qualified Name (FQN) of the workspace to deploy",
+)
+@click.option(
+    "--model-server",
+    "--model_server",
+    type=click.Choice(ModelServer, case_sensitive=False),
+    default=ModelServer.FASTAPI.value,
+    show_default=True,
+    help="Specify the model server (Case Insensitive).",
+)
+@click.option(
+    "--output-dir",
+    "--output_dir",
+    type=click.Path(exists=True, file_okay=False, writable=True),
+    help="Output directory for the model server code",
+    required=False,
+    show_default=True,
+    default=os.getcwd(),
+)
+@handle_exception_wrapper
+def model_init_command(
+    name: str,
+    model_version_fqn: str,
+    workspace_fqn: str,
+    model_server: ModelServer,
+    output_dir: Optional[str],
+):
+    """
+    Generates application code for the specified model version.
+    """
+    ml_client = get_client()
+    console.print(f"Generating application code for {model_version_fqn!r}")
+    output_dir = ml_client._initialize_model_server(
+        name=name,
+        model_version_fqn=model_version_fqn,
+        workspace_fqn=workspace_fqn,
+        model_server=ModelServer[model_server.upper()],
+        output_dir=output_dir,
+    )
+    message = f"""
+[bold green]Model Server code initialized successfully![/bold green]
+
+[bold]Code Location:[/bold] {output_dir}
+
+[bold]Next Steps:[/bold]
+- Navigate to the model server directory:
+[green]cd {output_dir}[/green]
+- Refer to the README file in the directory for further instructions.
+"""
+    console.print(message)
+
+
+def get_model_init_command():
+    return model_init_command

truefoundry/ml/cli/utils.py

@@ -0,0 +1,34 @@
+import rich_click as click
+
+from truefoundry.ml import MlFoundryException
+from truefoundry.ml.validation_utils import (
+    _APP_NAME_REGEX,
+)
+
+
+class AppName(click.ParamType):
+    """
+    Custom ParamType to validate application names.
+    """
+
+    name = "application-name"
+
+    def convert(self, value, param, ctx):
+        try:
+            if not value or not _APP_NAME_REGEX.match(value):
+                raise MlFoundryException(
+                    f"{value!r} must be lowercase and cannot contain spaces. It can only contain alphanumeric characters and hyphens. "
+                    f"Length must be between 1 and 30 characters."
+                )
+        except MlFoundryException as e:
+            self.fail(str(e), param, ctx)
+        return value
+
+
+class NonEmptyString(click.ParamType):
+    name = "non-empty-string"
+
+    def convert(self, value, param, ctx):
+        if isinstance(value, str) and not value.strip():
+            self.fail("Value cannot be empty or contain only spaces.", param, ctx)
+        return value

truefoundry/ml/log_types/artifacts/model.py

@@ -8,7 +8,7 @@ import typing
 import uuid
 import warnings
 from pathlib import Path
-from typing import TYPE_CHECKING, Any, Dict, Optional, Sequence, Tuple, Union
+from typing import TYPE_CHECKING, Any, Dict, Optional, Union
 
 from truefoundry.ml.artifact.truefoundry_artifact_repo import (
     ArtifactIdentifier,
@@ -31,6 +31,7 @@ from truefoundry.ml.autogen.client import (  # type: ignore[attr-defined]
     TrueFoundryArtifactSource,
     UpdateModelVersionRequestDto,
 )
+from truefoundry.ml.autogen.models import infer_signature as _infer_signature
 from truefoundry.ml.enums import ModelFramework
 from truefoundry.ml.exceptions import MlFoundryException
 from truefoundry.ml.log_types.artifacts.artifact import BlobStorageDirectory
@@ -44,11 +45,19 @@ from truefoundry.ml.log_types.artifacts.utils import (
     _validate_description,
     calculate_total_size,
 )
-from truefoundry.ml.model_framework import ModelFrameworkType, _ModelFramework
+from truefoundry.ml.model_framework import (
+    ModelFrameworkType,
+    _ModelFramework,
+    auto_update_environment_details,
+    auto_update_model_framework_details,
+)
 from truefoundry.ml.session import _get_api_client
 from truefoundry.pydantic_v1 import BaseModel, Extra
 
 if TYPE_CHECKING:
+    import numpy as np
+    import pandas as pd
+
     from truefoundry.ml.mlfoundry_run import MlFoundryRun
 
 
@@ -97,7 +106,7 @@ class ModelVersion:
         self._deleted = False
         self._description: str = ""
         self._metadata: Dict[str, Any] = {}
-        self._model_schema: Optional[Dict[str, Any]] = None
+        self._environment: Optional[ModelVersionEnvironment] = None
         self._framework: Optional[ModelFrameworkType] = None
         self._set_mutable_attrs()
 
@@ -140,22 +149,19 @@ class ModelVersion:
         if self._model_version.manifest:
             self._description = self._model_version.manifest.description or ""
             self._metadata = copy.deepcopy(self._model_version.manifest.metadata)
-            self._model_schema = copy.deepcopy(
-                self._model_version.manifest.model_schema
+            self._environment = copy.deepcopy(self._model_version.manifest.environment)
+            self._framework = (
+                copy.deepcopy(self._model_version.manifest.framework.actual_instance)
+                if self._model_version.manifest.framework
+                else None
             )
-            if self._model_version.manifest.framework:
-                self._framework = copy.deepcopy(
-                    self._model_version.manifest.framework.actual_instance
-                )
-            else:
-                self._framework = None
         else:
             self._description = self._model_version.description or ""
             self._metadata = copy.deepcopy(self._model_version.artifact_metadata)
+            self._environment = None
             self._framework = _ModelFramework.to_model_framework_type(
                 self._model_version.model_framework
             )
-            self._model_schema = None
 
     def _refetch_model_version(self, reset_mutable_attrs: bool = True):
         _model_version = self._mlfoundry_artifacts_api.get_model_version_get(
@@ -226,21 +232,21 @@ class ModelVersion:
         self._metadata = copy.deepcopy(value)
 
     @property
-    def model_schema(self) -> Optional[Dict[str, Any]]:
-        """Get model_schema for the current model"""
-        return self._model_schema
+    def environment(self) -> Optional[ModelVersionEnvironment]:
+        """Get the environment details for the model"""
+        return self._environment
 
-    @model_schema.setter
-    def model_schema(self, value: Optional[Dict[str, Any]]):
-        """set the model_schema for current model"""
+    @environment.setter
+    def environment(self, value: Optional[Dict[str, Any]]):
+        """set the environment details for the model"""
         if not self._model_version.manifest:
             warnings.warn(
-                message="This model version was created using an older serialization format. model_schema will not be updated",
+                message="This model version was created using an older serialization format. Environment will not be updated",
                 category=DeprecationWarning,
                 stacklevel=2,
             )
             return
-        self._model_schema = copy.deepcopy(value)
+        self._environment = copy.deepcopy(value)
 
     @property
     def framework(self) -> Optional["ModelFrameworkType"]:
@@ -439,7 +445,7 @@ class ModelVersion:
         if self._model_version.manifest:
             self._model_version.manifest.description = self.description
             self._model_version.manifest.metadata = self.metadata
-            self._model_version.manifest.model_schema = self.model_schema
+            self._model_version.manifest.environment = self.environment
             self._model_version.manifest.framework = (
                 Framework.from_dict(self.framework.dict()) if self.framework else None
             )
@@ -467,14 +473,12 @@ def _log_model_version(  # noqa: C901
     model_file_or_folder: Union[str, BlobStorageDirectory],
     mlfoundry_artifacts_api: Optional[MlfoundryArtifactsApi] = None,
     ml_repo_id: Optional[str] = None,
-    additional_files: Sequence[Tuple[Union[str, Path], Optional[str]]] = (),
     description: Optional[str] = None,
     metadata: Optional[Dict[str, Any]] = None,
     step: Optional[int] = 0,
     progress: Optional[bool] = None,
     framework: Optional[Union[str, ModelFramework, "ModelFrameworkType"]] = None,
     environment: Optional[ModelVersionEnvironment] = None,
-    model_schema: Optional[Dict[str, Any]] = None,
 ) -> ModelVersion:
     if (run and mlfoundry_artifacts_api) or (not run and not mlfoundry_artifacts_api):
         raise MlFoundryException(
@@ -498,7 +502,6 @@ def _log_model_version(  # noqa: C901
     step = step or 0
     total_size = None
     metadata = metadata or {}
-    additional_files = additional_files or {}
 
     _validate_description(description)
     _validate_artifact_metadata(metadata)
@@ -516,17 +519,6 @@ def _log_model_version(  # noqa: C901
                 ignore_model_dir_dest_conflict=True,
             )
 
-            # verify additional files and paths, copy additional files
-            if additional_files:
-                logger.info("Adding `additional_files` to model version contents")
-                temp_dest_to_src_map = _copy_additional_files(
-                    root_dir=temp_dir.name,
-                    files_dir="",
-                    model_dir=None,
-                    additional_files=additional_files,
-                    ignore_model_dir_dest_conflict=False,
-                    existing_dest_to_src_map=temp_dest_to_src_map,
-                )
         except Exception as e:
             temp_dir.cleanup()
             raise MlFoundryException("Failed to log model") from e
@@ -580,16 +572,24 @@ def _log_model_version(  # noqa: C901
     else:
         raise MlFoundryException("Invalid model_file_or_folder provided")
 
-    _framework = _ModelFramework.to_model_framework_type(framework)
     _source_cls = typing.get_type_hints(ModelVersionManifest)["source"]
+
+    # Auto fetch the framework & environment details if not provided
+    framework = _ModelFramework.to_model_framework_type(framework)
+    if framework and isinstance(model_file_or_folder, str):
+        auto_update_model_framework_details(
+            framework=framework, model_file_or_folder=model_file_or_folder
+        )
+    environment = environment or ModelVersionEnvironment()
+    auto_update_environment_details(environment=environment, framework=framework)
+
     model_manifest = ModelVersionManifest(
         description=description,
         metadata=metadata,
         source=_source_cls.from_dict(source.dict()),
-        framework=Framework.from_dict(_framework.dict()) if _framework else None,
+        framework=Framework.from_dict(framework.dict()) if framework else None,
         environment=environment,
         step=step,
-        model_schema=model_schema,
     )
     artifact_version_response = mlfoundry_artifacts_api.finalize_artifact_version_post(
         finalize_artifact_version_request_dto=FinalizeArtifactVersionRequestDto(
@@ -603,3 +603,15 @@ def _log_model_version(  # noqa: C901
         )
     )
     return ModelVersion.from_fqn(fqn=artifact_version_response.artifact_version.fqn)
+
+
+def infer_signature(
+    model_input: Any = None,
+    model_output: Optional[
+        Union["pd.DataFrame", "np.ndarray", Dict[str, "np.ndarray"]]
+    ] = None,
+    params: Optional[Dict[str, Any]] = None,
+):
+    return _infer_signature(
+        model_input=model_input, model_output=model_output, params=params
+    )

truefoundry/ml/log_types/artifacts/utils.py

@@ -2,7 +2,7 @@ import json
 import logging
 import os
 import posixpath
-from pathlib import Path
+from pathlib import Path, PureWindowsPath
 from typing import Any, Dict, Optional, Sequence, Tuple, Union
 
 from truefoundry.ml.exceptions import MlFoundryException
@@ -11,6 +11,13 @@ from truefoundry.ml.log_types.artifacts.constants import DESCRIPTION_MAX_LENGTH
 logger = logging.getLogger(__name__)
 
 
+def to_unix_path(path):
+    path = os.path.normpath(path)
+    if os.path.sep == "\\":
+        path = PureWindowsPath(path).as_posix()
+    return path
+
+
 def _copy_tree(
     root_dir: str, src_path: str, dest_path: str, dest_to_src: Dict[str, str]
 ):
@@ -42,6 +49,35 @@ def is_destination_path_dirlike(dest_path) -> bool:
     return False
 
 
+def get_single_file_path_if_only_one_in_directory(path: str) -> Optional[str]:
+    """
+    Get the filename from a path, or return a filename from a directory if a single file exists.
+    Args:
+        path: The file or folder path.
+    Returns:
+        Optional[str]: The filename or None if no files are found or multiple files are found.
+    """
+    # If it's already a file, return it as-is
+    if os.path.isfile(path):
+        return path
+
+    # If it's a directory, check if it contains a single file
+    if is_destination_path_dirlike(path):
+        all_files = []
+        for root, _, files in os.walk(path):
+            # Collect all files found in any subdirectory
+            all_files.extend(os.path.join(root, f) for f in files)
+            # If more than one file is found, stop early
+            if len(all_files) > 1:
+                return None
+
+        # If only one file is found, return it
+        if len(all_files) == 1:
+            return all_files[0]
+
+    return None  # No file found or Multiple files found
+
+
 def _copy_additional_files(
     root_dir: str,
     files_dir: str,  # relative to root dir e.g. "files/"
@@ -136,7 +172,7 @@ def _get_src_dest_pairs(
     dest_to_src_map: Dict[str, str],
 ) -> Sequence[Tuple[str, str]]:
     src_dest_pairs = [
-        (src_path, os.path.relpath(dest_abs_path, root_dir))
+        (src_path, to_unix_path(os.path.relpath(dest_abs_path, root_dir)))
         for dest_abs_path, src_path in dest_to_src_map.items()
     ]
     return src_dest_pairs