fxn 0.0.52__tar.gz → 0.0.54__tar.gz

{fxn-0.0.52 → fxn-0.0.54}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fxn
-Version: 0.0.52
+Version: 0.0.54
 Summary: Run prediction functions locally in Python. Register at https://fxn.ai.
 Author-email: "NatML Inc." <hi@fxn.ai>
 License:                                  Apache License

fxn-0.0.54/fxn/beta/__init__.py

@@ -0,0 +1,13 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from .metadata import (
+    CoreMLInferenceMetadata, LiteRTInferenceMetadata, LlamaCppInferenceMetadata,
+    OnnxInferenceMetadata, OnnxRuntimeInferenceSessionMetadata, OpenVINOInferenceMetadata,
+    QnnInferenceMetadata, QnnInferenceBackend, QnnInferenceQuantization,
+    # Deprecated
+    ONNXInferenceMetadata, ONNXRuntimeInferenceSessionMetadata
+)
+from .services import RemoteAcceleration

fxn-0.0.54/fxn/beta/cli/__init__.py

@@ -0,0 +1,7 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from .llm import app as llm_app
+from .mcp import app as mcp_app

fxn-0.0.54/fxn/beta/cli/llm.py

@@ -0,0 +1,22 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from pathlib import Path
+from typer import Argument, Option, Typer
+from typing_extensions import Annotated
+
+app = Typer(no_args_is_help=True)
+
+@app.command(name="chat", help="Start a chat session.")
+def chat (
+    model: Annotated[str, Argument(help="Model to chat with.")]
+):
+    pass
+
+@app.command(name="serve", help="Start an LLM server.")
+def serve (
+    port: Annotated[int, Option(help="Port to start the server on.")] = 11435
+):
+    pass

fxn-0.0.54/fxn/beta/cli/mcp.py

@@ -0,0 +1,16 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from pathlib import Path
+from typer import Argument, Option, Typer
+from typing_extensions import Annotated
+
+app = Typer(no_args_is_help=True)
+
+@app.command(name="serve", help="Start an MCP server.")
+def serve (
+    port: Annotated[int, Option(help="Port to start the server on.")] = 11436
+):
+    pass

{fxn-0.0.52 → fxn-0.0.54}/fxn/beta/client.py

@@ -10,8 +10,7 @@ from typing import get_origin, Callable, Generator, Iterator, TypeVar
 from ..client import FunctionClient
 from ..services import PredictionService as EdgePredictionService
 from ..types import Acceleration
-from .prediction import PredictionService
-from .remote import RemoteAcceleration
+from .services import PredictionService, RemoteAcceleration
 
 F = TypeVar("F", bound=Callable[..., object])
 

fxn-0.0.54/fxn/beta/llm/__init__.py

@@ -0,0 +1,5 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+

fxn-0.0.54/fxn/beta/llm/server.py

@@ -0,0 +1,5 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+

fxn-0.0.54/fxn/beta/metadata.py

@@ -0,0 +1,189 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from pathlib import Path
+from pydantic import BaseModel, BeforeValidator, ConfigDict, Field
+from typing import Annotated, Literal
+
+def _validate_torch_module (module: "torch.nn.Module") -> "torch.nn.Module": # type: ignore
+    try:
+        from torch.nn import Module
+        if not isinstance(module, Module):
+            raise ValueError(f"Expected `torch.nn.Module` model but got `{type(module).__qualname__}`")
+        return module
+    except ImportError:
+        raise ImportError("PyTorch is required to create this metadata but it is not installed.")
+
+def _validate_ort_inference_session (session: "onnxruntime.InferenceSession") -> "onnxruntime.InferenceSession": # type: ignore
+    try:
+        from onnxruntime import InferenceSession
+        if not isinstance(session, InferenceSession):
+            raise ValueError(f"Expected `onnxruntime.InferenceSession` model but got `{type(session).__qualname__}`")
+        return session
+    except ImportError:
+        raise ImportError("ONNXRuntime is required to create this metadata but it is not installed.")
+
+def _validate_torch_tensor_args (args: list) -> list:
+    try:
+        from torch import Tensor
+        for idx, arg in enumerate(args):
+            if not isinstance(arg, Tensor):
+                raise ValueError(f"Expected `torch.Tensor` instance at `model_args[{idx}]` but got `{type(arg).__qualname__}`")
+        return args
+    except ImportError:
+        raise ImportError("PyTorch is required to create this metadata but it is not installed.")
+    
+def _validate_llama_cpp_model (model: "llama_cpp.llama.Llama") -> "llama_cpp.llama.Llama": # type: ignore
+    try:
+        from llama_cpp import Llama
+        if not isinstance(model, Llama):
+            raise ValueError(f"Expected `llama_cpp.llama.Llama` model but got `{type(model).__qualname__}`")
+        return model
+    except ImportError:
+        raise ImportError("Llama-cpp-python is required to create this metadata but it is not installed.")
+
+class CoreMLInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower a PyTorch model for inference on iOS, macOS, and visionOS with CoreML.
+
+    Members:
+        model (torch.nn.Module): PyTorch module to apply metadata to.
+        model_args (tuple[Tensor,...]): Positional inputs to the model.
+    """
+    kind: Literal["meta.inference.coreml"] = "meta.inference.coreml"
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(
+        description="PyTorch module to apply metadata to.",
+        exclude=True
+    )
+    model_args: Annotated[list[object], BeforeValidator(_validate_torch_tensor_args)] = Field(
+        description="Positional inputs to the model.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+class OnnxInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower a PyTorch model for inference.
+
+    Members:
+        model (torch.nn.Module): PyTorch module to apply metadata to.
+        model_args (tuple[Tensor,...]): Positional inputs to the model.
+    """
+    kind: Literal["meta.inference.onnx"] = "meta.inference.onnx"
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(
+        description="PyTorch module to apply metadata to.",
+        exclude=True
+    )
+    model_args: Annotated[list[object], BeforeValidator(_validate_torch_tensor_args)] = Field(
+        description="Positional inputs to the model.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+class OnnxRuntimeInferenceSessionMetadata (BaseModel):
+    """
+    Metadata required to lower an ONNXRuntime `InferenceSession` for inference.
+
+    Members:
+        session (onnxruntime.InferenceSession): ONNXRuntime inference session to apply metadata to.
+        model_path (str | Path): ONNX model path. The model must exist at this path in the compiler sandbox.
+    """
+    kind: Literal["meta.inference.onnxruntime"] = "meta.inference.onnxruntime"
+    session: Annotated[object, BeforeValidator(_validate_ort_inference_session)] = Field(
+        description="ONNXRuntime inference session to apply metadata to.",
+        exclude=True
+    )
+    model_path: str | Path = Field(
+        description="ONNX model path. The model must exist at this path in the compiler sandbox.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+class LiteRTInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower PyTorch model for inference with LiteRT (fka TensorFlow Lite).
+
+    Members:
+        model (torch.nn.Module): PyTorch module to apply metadata to.
+        model_args (tuple[Tensor,...]): Positional inputs to the model.
+    """
+    kind: Literal["meta.inference.litert"] = "meta.inference.litert"
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(
+        description="PyTorch module to apply metadata to.",
+        exclude=True
+    )
+    model_args: Annotated[list[object], BeforeValidator(_validate_torch_tensor_args)] = Field(
+        description="Positional inputs to the model.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+class OpenVINOInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower PyTorch model for interence with Intel OpenVINO.
+
+    Members:
+        model (torch.nn.Module): PyTorch module to apply metadata to.
+        model_args (tuple[Tensor,...]): Positional inputs to the model.
+    """
+    kind: Literal["meta.inference.openvino"] = "meta.inference.openvino"
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(
+        description="PyTorch module to apply metadata to.",
+        exclude=True
+    )
+    model_args: Annotated[list[object], BeforeValidator(_validate_torch_tensor_args)] = Field(
+        description="Positional inputs to the model.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+QnnInferenceBackend = Literal["cpu", "gpu", "htp"]
+QnnInferenceQuantization = Literal["w8a8", "w8a16", "w4a8", "w4a16"]
+
+class QnnInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower a PyTorch model for inference on Qualcomm accelerators with QNN SDK.
+
+    Members:
+        model (torch.nn.Module): PyTorch module to apply metadata to.
+        model_args (tuple[Tensor,...]): Positional inputs to the model.
+        backend (QnnInferenceBackend): QNN inference backend. Defaults to `cpu`.
+        quantization (QnnInferenceQuantization): QNN model quantization mode. This MUST only be specified when backend is `htp`.
+    """
+    kind: Literal["meta.inference.qnn"] = "meta.inference.qnn"
+    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(
+        description="PyTorch module to apply metadata to.",
+        exclude=True
+    )
+    model_args: Annotated[list[object], BeforeValidator(_validate_torch_tensor_args)] = Field(
+        description="Positional inputs to the model.",
+        exclude=True
+    )
+    backend: QnnInferenceBackend = Field(
+        default="cpu",
+        description="QNN backend to execute the model.",
+        exclude=True
+    )
+    quantization: QnnInferenceQuantization | None = Field(
+        default=None,
+        description="QNN model quantization mode. This MUST only be specified when backend is `htp`.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+class LlamaCppInferenceMetadata (BaseModel):
+    """
+    Metadata required to lower a Llama.cpp model for LLM inference.
+    """
+    kind: Literal["meta.inference.llama_cpp"] = "meta.inference.llama_cpp"
+    model: Annotated[object, BeforeValidator(_validate_llama_cpp_model)] = Field(
+        description="Llama model that metadata applies to.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+# DEPRECATED
+ONNXInferenceMetadata = OnnxInferenceMetadata
+ONNXRuntimeInferenceSessionMetadata = OnnxRuntimeInferenceSessionMetadata

fxn-0.0.54/fxn/beta/services/__init__.py

@@ -0,0 +1,7 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from .prediction import PredictionService
+from .remote import RemoteAcceleration

{fxn-0.0.52/fxn/beta → fxn-0.0.54/fxn/beta/services}/prediction.py

@@ -3,7 +3,7 @@
 #   Copyright © 2025 NatML Inc. All Rights Reserved.
 #
 
-from ..client import FunctionClient
+from ...client import FunctionClient
 from .remote import RemotePredictionService
 
 class PredictionService:

{fxn-0.0.52/fxn/beta → fxn-0.0.54/fxn/beta/services}/remote.py

@@ -15,10 +15,10 @@ from requests import get, put
 from typing import Literal
 from urllib.request import urlopen
 
-from ..c import Configuration
-from ..client import FunctionClient
-from ..services import Value
-from ..types import Dtype, Prediction
+from ...c import Configuration
+from ...client import FunctionClient
+from ...services.prediction import Value
+from ...types import Dtype, Prediction
 
 RemoteAcceleration = Literal["auto", "cpu", "a40", "a100"]
 

fxn-0.0.54/fxn/cli/__init__.py

@@ -0,0 +1,91 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+import typer
+
+from ..logging import TracebackMarkupConsole
+from ..version import __version__
+
+from .auth import app as auth_app
+from .compile import compile_predictor
+from .misc import cli_options
+from .predictions import create_prediction
+from .predictors import archive_predictor, delete_predictor, retrieve_predictor
+from .sources import retrieve_source
+from ..beta.cli import llm_app, mcp_app
+
+# Define CLI
+typer.main.console_stderr = TracebackMarkupConsole()
+app = typer.Typer(
+    name=f"Function CLI {__version__}",
+    no_args_is_help=True,
+    pretty_exceptions_show_locals=False,
+    pretty_exceptions_short=True,
+    add_completion=False
+)
+
+# Add top level options
+app.callback()(cli_options)
+
+# Predictions
+app.command(
+    name="predict",
+    help="Make a prediction.",
+    context_settings={ "allow_extra_args": True, "ignore_unknown_options": True },
+    rich_help_panel="Predictions"
+)(create_prediction)
+app.command(
+    name="source",
+    help="Retrieve the generated C++ code for a given prediction.",
+    rich_help_panel="Predictions"
+)(retrieve_source)
+
+# Predictors
+app.command(
+    name="compile",
+    help="Create a predictor by compiling a Python function.",
+    rich_help_panel="Predictors"
+)(compile_predictor)
+app.command(
+    name="retrieve",
+    help="Retrieve a predictor.",
+    rich_help_panel="Predictors"
+)(retrieve_predictor)
+app.command(
+    name="archive",
+    help="Archive a predictor." ,
+    rich_help_panel="Predictors"
+)(archive_predictor)
+app.command(
+    name="delete",
+    help="Delete a predictor.",
+    rich_help_panel="Predictors"
+)(delete_predictor)
+
+# Subcommands
+app.add_typer(
+    auth_app,
+    name="auth",
+    help="Login, logout, and check your authentication status.",
+    rich_help_panel="Auth"
+)
+app.add_typer(
+    llm_app,
+    name="llm",
+    hidden=True,
+    help="Work with large language models (LLMs).",
+    rich_help_panel="Beta"
+)
+app.add_typer(
+    mcp_app,
+    name="mcp",
+    hidden=True,
+    help="Provide prediction functions as tools for use by AI assistants.",
+    rich_help_panel="Beta"
+)
+
+# Run
+if __name__ == "__main__":
+    app()

{fxn-0.0.52 → fxn-0.0.54}/fxn/cli/compile.py

@@ -86,6 +86,8 @@ def _load_predictor_func (path: str) -> Callable[...,object]:
     if "" not in sys.path:
         sys.path.insert(0, "")
     path: Path = Path(path).resolve()
+    if not path.exists():
+        raise ValueError(f"Cannot compile predictor because no Python module exists at the given path.")
     sys.path.insert(0, str(path.parent))
     name = getmodulename(path)
     spec = spec_from_file_location(name, path)

{fxn-0.0.52 → fxn-0.0.54}/fxn/cli/sources.py

@@ -15,8 +15,8 @@ from ..logging import CustomProgress, CustomProgressTask
 from .auth import get_access_key
 
 def retrieve_source (
-    predictor: Annotated[str, Option(help="Predictor tag.")] = None,
     prediction: Annotated[str, Option(help="Prediction identifier. If specified, this MUST be from a prediction returned by the Function API.")] = None,
+    predictor: Annotated[str, Option(help="Predictor tag. If specified, a prediction will be made with this predictor before retrieving the source.")] = None,
     output: Annotated[Path, Option(help="Path to output source file.")] = Path("predictor.cpp")
 ):
     if not ((predictor is not None) ^ (prediction is not None)):

{fxn-0.0.52 → fxn-0.0.54}/fxn/compile.py

@@ -13,11 +13,11 @@ from typing import Any, Callable, Literal, ParamSpec, TypeVar, cast
 
 from .beta import (
     CoreMLInferenceMetadata, LiteRTInferenceMetadata, LlamaCppInferenceMetadata,
-    ONNXInferenceMetadata, ONNXRuntimeInferenceSessionMetadata, OpenVINOInferenceMetadata,
+    OnnxInferenceMetadata, OnnxRuntimeInferenceSessionMetadata, OpenVINOInferenceMetadata,
     QnnInferenceMetadata
 )
 from .sandbox import Sandbox
-from .types import AccessMode
+from .types import PredictorAccess
 
 CompileTarget = Literal[
     "android",
@@ -33,8 +33,8 @@ CompileMetadata = (
     CoreMLInferenceMetadata             |
     LiteRTInferenceMetadata             |
     LlamaCppInferenceMetadata           |
-    ONNXInferenceMetadata               |
-    ONNXRuntimeInferenceSessionMetadata |
+    OnnxInferenceMetadata               |
+    OnnxRuntimeInferenceSessionMetadata |
     OpenVINOInferenceMetadata           |
     QnnInferenceMetadata
 )
@@ -51,7 +51,7 @@ class PredictorSpec (BaseModel):
     sandbox: Sandbox = Field(description="Sandbox to compile the function.")
     targets: list[str] | None = Field(description="Targets to compile this predictor for. Pass `None` to compile for our default targets.")
     metadata: list[object] = Field(default=[], description="Metadata to use while compiling the function.")
-    access: AccessMode = Field(description="Predictor access.")
+    access: PredictorAccess = Field(description="Predictor access.")
     card: str | None = Field(default=None, description="Predictor card (markdown).")
     media: str | None = Field(default=None, description="Predictor media URL.")
     license: str | None = Field(default=None, description="Predictor license URL. This is required for public predictors.")
@@ -65,7 +65,7 @@ def compile (
     trace_modules: list[ModuleType]=[],
     targets: list[CompileTarget]=None,
     metadata: list[CompileMetadata]=[],
-    access: AccessMode=AccessMode.Private,
+    access: PredictorAccess="private",
     card: str | Path=None,
     media: Path=None,
     license: str=None,
@@ -81,7 +81,7 @@ def compile (
         trace_modules (list): Modules to trace and compile.
         targets (list): Targets to compile this predictor for. Pass `None` to compile for our default targets.
         metadata (list): Metadata to use while compiling the function.
-        access (AccessMode): Predictor access.
+        access (PredictorAccess): Predictor access.
         card (str | Path): Predictor card markdown string or path to card.
         media (Path): Predictor thumbnail image (jpeg or png) path.
         license (str): Predictor license URL. This is required for public predictors.

{fxn-0.0.52 → fxn-0.0.54}/fxn/sandbox.py

@@ -57,6 +57,8 @@ class EntrypointCommand (UploadableCommand):
 class PipInstallCommand (BaseModel):
     kind: Literal["pip_install"] = "pip_install"
     packages: list[str]
+    index_url: str | None
+    flags: str
 
 class AptInstallCommand (BaseModel):
     kind: Literal["apt_install"] = "apt_install"
@@ -133,14 +135,25 @@ class Sandbox (BaseModel):
         )
         return Sandbox(commands=self.commands + [command])
 
-    def pip_install (self, *packages: str) -> Sandbox:
+    def pip_install (
+        self,
+        *packages: str,
+        index_url: str=None,
+        flags: str=""
+    ) -> Sandbox:
         """
         Install Python packages in the sandbox.
 
         Parameters:
             packages (list): Packages to install.
+            index_url (str | None): Index URL to search for package.
+            flags (str): Additional flags to pass to `pip`.
         """
-        command = PipInstallCommand(packages=packages)
+        command = PipInstallCommand(
+            packages=packages,
+            index_url=index_url,
+            flags=flags
+        )
         return Sandbox(commands=self.commands + [command])
 
     def apt_install (self, *packages: str) -> Sandbox:

{fxn-0.0.52 → fxn-0.0.54}/fxn/services/prediction.py

@@ -43,18 +43,6 @@ class PredictionService:
         self.__cache_dir = self.__class__.__get_home_dir() / ".fxn" / "cache"
         self.__cache_dir.mkdir(parents=True, exist_ok=True)
 
-    def ready (self, tag: str, **kwargs) -> bool:
-        """
-        Check whether a predictor has been preloaded and is ready to make predictions.
-
-        Parameters:
-            tag (str): Predictor tag.
-
-        Returns:
-            bool: Whether the predictor is ready to make predictions.
-        """
-        return tag in self.__cache
-
     def create (
         self,
         tag: str,

{fxn-0.0.52 → fxn-0.0.54}/fxn/types/__init__.py

@@ -5,5 +5,5 @@
 
 from .dtype import Dtype
 from .prediction import Acceleration, Prediction, PredictionResource
-from .predictor import AccessMode, EnumerationMember, Parameter, Predictor, PredictorStatus, Signature
+from .predictor import EnumerationMember, Parameter, Predictor, PredictorAccess, PredictorStatus, Signature
 from .user import User

{fxn-0.0.52 → fxn-0.0.54}/fxn/types/predictor.py

@@ -3,28 +3,15 @@
 #   Copyright © 2025 NatML Inc. All Rights Reserved.
 #
 
-from enum import Enum
 from pydantic import AliasChoices, BaseModel, ConfigDict, Field
-from typing import Any
+from typing import Any, Literal
 
 from .dtype import Dtype
 from .user import User
 
-class AccessMode (str, Enum):
-    """
-    Predictor access mode.
-    """
-    Public = "PUBLIC"
-    Private = "PRIVATE"
+PredictorAccess = Literal["public", "private", "unlisted"]
 
-class PredictorStatus (str, Enum):
-    """
-    Predictor status.
-    """
-    Compiling = "COMPILING"
-    Active = "ACTIVE"
-    Invalid = "INVALID"
-    Archived = "ARCHIVED"
+PredictorStatus = Literal["compiling", "active", "archived"]
 
 class EnumerationMember (BaseModel):
     """
@@ -79,7 +66,7 @@ class Predictor (BaseModel):
         owner (User): Predictor owner.
         name (str): Predictor name.
         status (PredictorStatus): Predictor status.
-        access (AccessMode): Predictor access.
+        access (PredictorAccess): Predictor access.
         signature (Signature): Predictor signature.
         created (str): Date created.
         description (str): Predictor description.
@@ -91,7 +78,7 @@ class Predictor (BaseModel):
     owner: User = Field(description="Predictor owner.")
     name: str = Field(description="Predictor name.")
     status: PredictorStatus = Field(description="Predictor status.")
-    access: AccessMode = Field(description="Predictor access.")
+    access: PredictorAccess = Field(description="Predictor access.")
     signature: Signature = Field(description="Predictor signature.")
     created: str = Field(description="Date created.")
     description: str | None = Field(default=None, description="Predictor description.")

{fxn-0.0.52 → fxn-0.0.54}/fxn/version.py

@@ -3,4 +3,4 @@
 #   Copyright © 2025 NatML Inc. All Rights Reserved.
 #
 
-__version__ = "0.0.52"
+__version__ = "0.0.54"

{fxn-0.0.52 → fxn-0.0.54}/fxn.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fxn
-Version: 0.0.52
+Version: 0.0.54
 Summary: Run prediction functions locally in Python. Register at https://fxn.ai.
 Author-email: "NatML Inc." <hi@fxn.ai>
 License:                                  Apache License

{fxn-0.0.52 → fxn-0.0.54}/fxn.egg-info/SOURCES.txt

@@ -17,8 +17,14 @@ fxn.egg-info/top_level.txt
 fxn/beta/__init__.py
 fxn/beta/client.py
 fxn/beta/metadata.py
-fxn/beta/prediction.py
-fxn/beta/remote.py
+fxn/beta/cli/__init__.py
+fxn/beta/cli/llm.py
+fxn/beta/cli/mcp.py
+fxn/beta/llm/__init__.py
+fxn/beta/llm/server.py
+fxn/beta/services/__init__.py
+fxn/beta/services/prediction.py
+fxn/beta/services/remote.py
 fxn/c/__init__.py
 fxn/c/configuration.py
 fxn/c/fxnc.py

fxn-0.0.52/fxn/beta/__init__.py

@@ -1,11 +0,0 @@
-# 
-#   Function
-#   Copyright © 2025 NatML Inc. All Rights Reserved.
-#
-
-from .metadata import (
-    CoreMLInferenceMetadata, LiteRTInferenceMetadata, LlamaCppInferenceMetadata,
-    ONNXInferenceMetadata, ONNXRuntimeInferenceSessionMetadata, OpenVINOInferenceMetadata,
-    QnnInferenceMetadata
-)
-from .remote import RemoteAcceleration

fxn-0.0.52/fxn/beta/metadata.py

@@ -1,89 +0,0 @@
-# 
-#   Function
-#   Copyright © 2025 NatML Inc. All Rights Reserved.
-#
-
-from pathlib import Path
-from pydantic import BaseModel, BeforeValidator, ConfigDict, Field
-from typing import Annotated, Literal
-
-def _validate_torch_module (module: "torch.nn.Module") -> "torch.nn.Module": # type: ignore
-    try:
-        from torch.nn import Module # type: ignore
-        if not isinstance(module, Module):
-            raise ValueError(f"Expected torch.nn.Module, got {type(module)}")
-        return module
-    except ImportError:
-        raise ImportError("PyTorch is required to create this metadata but is not installed.")
-
-def _validate_ort_inference_session (session: "onnxruntime.InferenceSession") -> "onnxruntime.InferenceSession": # type: ignore
-    try:
-        from onnxruntime import InferenceSession # type: ignore
-        if not isinstance(session, InferenceSession):
-            raise ValueError(f"Expected onnxruntime.InferenceSession, got {type(session)}")
-        return session
-    except ImportError:
-        raise ImportError("ONNXRuntime is required to create this metadata but is not installed.")
-
-class CoreMLInferenceMetadata (BaseModel):
-    """
-    Metadata required to lower a PyTorch model for inference on iOS, macOS, and visionOS with CoreML.
-    """
-    kind: Literal["meta.inference.coreml"] = "meta.inference.coreml"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
-    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
-    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
-
-class ONNXInferenceMetadata (BaseModel):
-    """
-    Metadata required to lower a PyTorch model for inference.
-    """
-    kind: Literal["meta.inference.onnx"] = "meta.inference.onnx"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
-    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
-    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
-
-class ONNXRuntimeInferenceSessionMetadata (BaseModel):
-    """
-    Metadata required to lower an ONNXRuntime `InferenceSession` for inference.
-    """
-    kind: Literal["meta.inference.onnxruntime"] = "meta.inference.onnxruntime"
-    session: Annotated[object, BeforeValidator(_validate_ort_inference_session)] = Field(description="ONNXRuntime inference session to apply metadata to.", exclude=True)
-    model_path: Path = Field(description="ONNX model path. The model must exist at this path in the compiler sandbox.", exclude=True)
-    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
-
-class LiteRTInferenceMetadata (BaseModel):
-    """
-    Metadata required to lower PyTorch model for inference with LiteRT (fka TensorFlow Lite).
-    """
-    kind: Literal["meta.inference.litert"] = "meta.inference.litert"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
-    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
-    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
-
-class OpenVINOInferenceMetadata (BaseModel):
-    """
-    Metadata required to lower PyTorch model for interence with Intel OpenVINO.
-    """
-    kind: Literal["meta.inference.openvino"] = "meta.inference.openvino"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
-    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
-    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
-
-class QnnInferenceMetadata (BaseModel):
-    """
-    Metadata required to lower a PyTorch model for inference on Qualcomm accelerators with QNN SDK.
-    """
-    kind: Literal["meta.inference.qnn"] = "meta.inference.qnn"
-    model: Annotated[object, BeforeValidator(_validate_torch_module)] = Field(description="PyTorch module to apply metadata to.", exclude=True)
-    model_args: list[object] = Field(description="Positional inputs to the model.", exclude=True)
-    backend: Literal["cpu", "gpu"] = Field(default="cpu", description="QNN backend to execute the model.", exclude=True) # CHECK # Add `htp`
-    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
-
-class LlamaCppInferenceMetadata (BaseModel): # INCOMPLETE
-    """
-    Metadata required to lower a GGUF model for LLM inference.
-    """
-    kind: Literal["meta.inference.gguf"] = "meta.inference.gguf"
-    model_path: Path = Field(description="GGUF model path. The model must exist at this path in the compiler sandbox.", exclude=True)
-    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)

fxn-0.0.52/fxn/cli/__init__.py

@@ -1,51 +0,0 @@
-# 
-#   Function
-#   Copyright © 2025 NatML Inc. All Rights Reserved.
-#
-
-import typer
-
-from ..logging import TracebackMarkupConsole
-from ..version import __version__
-
-from .auth import app as auth_app
-from .compile import compile_predictor
-from .misc import cli_options
-from .predictions import create_prediction
-from .predictors import archive_predictor, delete_predictor, retrieve_predictor
-from .sources import retrieve_source
-
-# Define CLI
-typer.main.console_stderr = TracebackMarkupConsole()
-app = typer.Typer(
-    name=f"Function CLI {__version__}",
-    no_args_is_help=True,
-    pretty_exceptions_show_locals=False,
-    pretty_exceptions_short=True,
-    add_completion=False
-)
-
-# Add top level options
-app.callback()(cli_options)
-
-# Add subcommands
-app.add_typer(auth_app, name="auth", help="Login, logout, and check your authentication status.")
-
-# Add top-level commands
-app.command(
-    name="predict",
-    help="Make a prediction.",
-    context_settings={ "allow_extra_args": True, "ignore_unknown_options": True }
-)(create_prediction)
-app.command(
-    name="compile",
-    help="Create a predictor by compiling a Python function."
-)(compile_predictor)
-app.command(name="retrieve", help="Retrieve a predictor.")(retrieve_predictor)
-app.command(name="archive", help="Archive a predictor.")(archive_predictor)
-app.command(name="delete", help="Delete a predictor.")(delete_predictor)
-app.command(name="source", help="Retrieve the native source code for a given prediction.")(retrieve_source)
-
-# Run
-if __name__ == "__main__":
-    app()