fxn 0.0.52__tar.gz → 0.0.53__tar.gz

{fxn-0.0.52 → fxn-0.0.53}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fxn
-Version: 0.0.52
+Version: 0.0.53
 Summary: Run prediction functions locally in Python. Register at https://fxn.ai.
 Author-email: "NatML Inc." <hi@fxn.ai>
 License:                                  Apache License

fxn-0.0.53/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-0.0.53/fxn/beta/cli/__init__.py

@@ -0,0 +1,6 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from .llm import app as llm_app

fxn-0.0.53/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-0.0.52 → fxn-0.0.53}/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-0.0.53/fxn/beta/llm/__init__.py

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

fxn-0.0.53/fxn/beta/llm/server.py

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

fxn-0.0.53/fxn/beta/metadata.py

@@ -0,0 +1,171 @@
+# 
+#   Function
+#   Copyright © 2025 NatML Inc. All Rights Reserved.
+#
+
+from os import PathLike
+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.
+
+    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: 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.
+
+    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: 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.
+
+    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: 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.
+
+    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: list[object] = Field(
+        description="Positional inputs to the model.",
+        exclude=True
+    )
+    model_config = ConfigDict(arbitrary_types_allowed=True, frozen=True)
+
+QnnInferenceBackend = Literal["cpu", "gpu"] # `htp` coming soon
+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: list[object] = 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): # 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)
+
+# DEPRECATED
+ONNXInferenceMetadata = OnnxInferenceMetadata
+ONNXRuntimeInferenceSessionMetadata = OnnxRuntimeInferenceSessionMetadata

fxn-0.0.53/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-0.0.53/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-0.0.53/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-0.0.52 → fxn-0.0.53}/fxn/cli/__init__.py

@@ -14,6 +14,7 @@ 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
 
 # Define CLI
 typer.main.console_stderr = TracebackMarkupConsole()
@@ -30,6 +31,7 @@ app.callback()(cli_options)
 
 # Add subcommands
 app.add_typer(auth_app, name="auth", help="Login, logout, and check your authentication status.")
+app.add_typer(llm_app, name="llm", hidden=True, help="Work with large language models (LLMs).")
 
 # Add top-level commands
 app.command(
@@ -44,7 +46,7 @@ app.command(
 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)
+app.command(name="source", help="Retrieve the generated native code for a given predictor.")(retrieve_source)
 
 # Run
 if __name__ == "__main__":

{fxn-0.0.52 → fxn-0.0.53}/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-0.0.53}/fxn/compile.py

@@ -13,7 +13,7 @@ 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
@@ -33,8 +33,8 @@ CompileMetadata = (
     CoreMLInferenceMetadata             |
     LiteRTInferenceMetadata             |
     LlamaCppInferenceMetadata           |
-    ONNXInferenceMetadata               |
-    ONNXRuntimeInferenceSessionMetadata |
+    OnnxInferenceMetadata               |
+    OnnxRuntimeInferenceSessionMetadata |
     OpenVINOInferenceMetadata           |
     QnnInferenceMetadata
 )

{fxn-0.0.52 → fxn-0.0.53}/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-0.0.53}/fxn/version.py

@@ -3,4 +3,4 @@
 #   Copyright © 2025 NatML Inc. All Rights Reserved.
 #
 
-__version__ = "0.0.52"
+__version__ = "0.0.53"

{fxn-0.0.52 → fxn-0.0.53}/fxn.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: fxn
-Version: 0.0.52
+Version: 0.0.53
 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-0.0.53}/fxn.egg-info/SOURCES.txt

@@ -17,8 +17,13 @@ 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/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)