depthtensor 2.4.0__tar.gz → 2.4.2__tar.gz

{depthtensor-2.4.0 → depthtensor-2.4.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: depthtensor
-Version: 2.4.0
+Version: 2.4.2
 Summary: A Hardware-Accelerated Tensor Computation and Autograd Engine
 Author-email: Le Hong Ha <lehahong1310@gmail.com>
 License: MIT License
@@ -40,7 +40,7 @@ Dynamic: license-file
 
 # DepthTensor: A Hardware-Accelerated Tensor Computation and Autograd Engine
 
-**DepthTensor** is a light-weight, high-performance library for reverse-mode automatic differentiation (AD). It is useful in building the mathematical foundation of deep learning frameworks, with the use of a `Tensor` object which dynamically builds computational graphs and computes gradients using **Vector-Jacobian Products (VJP)**, generalized for tensors of abritrary rank.
+**DepthTensor** is a light-weight, high-performance library for reverse-mode automatic differentiation (AD). It is useful in building the mathematical foundation of deep learning frameworks, with the use of a `Tensor` object which dynamically builds computational graphs and computes gradients using **Vector-Jacobian Products (VJP)**, generalized for tensors of arbitrary rank.
 
 > **Note**: This is the core autograd and tensor computation engine. For the full deep learning framework, refer to [DepthML](https://github.com/l-h-ha/DepthML).
 
@@ -82,7 +82,7 @@ All gradients are aggregated from all gradient downstreams.
 
 ## 2. Architecture and System Design
 
-### 2.1. The differentiable Primitive (`Tensor`)
+### 2.1. The Differentiable Primitive (`Tensor`)
 
 The `Tensor` class acts like a node in the Directed Acyclic Graph (DAG).
 
@@ -109,13 +109,13 @@ a, b = dt.Tensor([1], device="gpu"), dt.Tensor(
     [100], device="gpu"
 )
 
-# 2. Optimization Loop
+# Optimization Loop
 lr = 0.001
 for i in range(500):
     # Rosenbrock: f(x,y) = (a-x)^2 + b(y-x^2)^2
     loss = (a - x) ** 2 + b * (y - x**2) ** 2
 
-    # Backpward pass
+    # Backward pass
     dt.differentiate(loss)
 
     # Gradient Descent

{depthtensor-2.4.0 → depthtensor-2.4.2}/README.md

@@ -1,6 +1,6 @@
 # DepthTensor: A Hardware-Accelerated Tensor Computation and Autograd Engine
 
-**DepthTensor** is a light-weight, high-performance library for reverse-mode automatic differentiation (AD). It is useful in building the mathematical foundation of deep learning frameworks, with the use of a `Tensor` object which dynamically builds computational graphs and computes gradients using **Vector-Jacobian Products (VJP)**, generalized for tensors of abritrary rank.
+**DepthTensor** is a light-weight, high-performance library for reverse-mode automatic differentiation (AD). It is useful in building the mathematical foundation of deep learning frameworks, with the use of a `Tensor` object which dynamically builds computational graphs and computes gradients using **Vector-Jacobian Products (VJP)**, generalized for tensors of arbitrary rank.
 
 > **Note**: This is the core autograd and tensor computation engine. For the full deep learning framework, refer to [DepthML](https://github.com/l-h-ha/DepthML).
 
@@ -42,7 +42,7 @@ All gradients are aggregated from all gradient downstreams.
 
 ## 2. Architecture and System Design
 
-### 2.1. The differentiable Primitive (`Tensor`)
+### 2.1. The Differentiable Primitive (`Tensor`)
 
 The `Tensor` class acts like a node in the Directed Acyclic Graph (DAG).
 
@@ -69,13 +69,13 @@ a, b = dt.Tensor([1], device="gpu"), dt.Tensor(
     [100], device="gpu"
 )
 
-# 2. Optimization Loop
+# Optimization Loop
 lr = 0.001
 for i in range(500):
     # Rosenbrock: f(x,y) = (a-x)^2 + b(y-x^2)^2
     loss = (a - x) ** 2 + b * (y - x**2) ** 2
 
-    # Backpward pass
+    # Backward pass
     dt.differentiate(loss)
 
     # Gradient Descent

{depthtensor-2.4.0 → depthtensor-2.4.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "depthtensor"
-version = "2.4.0"
+version = "2.4.2"
 description = "A Hardware-Accelerated Tensor Computation and Autograd Engine"
 readme = "README.md"
 authors = [
@@ -13,8 +13,8 @@ authors = [
 license = { file = "LICENSE" }
 classifiers = [
     "Programming Language :: Python :: 3",
-    "Topic :: Scientific/Engineering :: Artificial Intelligence",  # Fixed
-    "License :: OSI Approved :: MIT License",                      # Fixed
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "License :: OSI Approved :: MIT License",                   
     "Operating System :: OS Independent"
 ]
 requires-python = ">=3.8"

{depthtensor-2.4.0 → depthtensor-2.4.2}/src/DepthTensor/__init__.py

@@ -6,4 +6,4 @@ from ._core.ops import *
 from ._core.exceptions import *
 from ._core import random
 
-__version__ = "2.4.0"
+__version__ = "2.4.2"

{depthtensor-2.4.0 → depthtensor-2.4.2}/src/DepthTensor/_core/ops/comparison.py

@@ -1,12 +1,12 @@
-from typing import Union, Optional, Tuple, overload
+from typing import overload, Union, Tuple
 
 from ...typing import (
-    TensorLike,
-    DeviceLike,
-    NDArrayLikeBool,
+    TensorType,
+    Device,
+    TensorDataBool,
     Casting,
     Order,
-    OperandLike,
+    TensorLike,
 )
 
 from ..exceptions import (
@@ -16,7 +16,7 @@ from ..exceptions import (
     DEVICE_MISMATCH_MSG,
 )
 
-from ..utils import to_xp_array, get_device, get_two_operand_op_device
+from ..utils import to_tensordata, get_device, get_two_operand_op_device
 
 import numpy as np
 
@@ -32,35 +32,35 @@ except (ImportError, ModuleNotFoundError):
 
 @overload
 def where(
-    condition: OperandLike,
+    condition: TensorLike,
     /,
     *,
-    device: DeviceLike = "cpu",
+    device: Device = "cpu",
     requires_grad: bool = False,
-) -> Tuple[TensorLike, ...]: ...
+) -> Tuple[TensorType, ...]: ...
 
 
 @overload
 def where(
-    condition: OperandLike,
-    x: Optional[OperandLike],
-    y: Optional[OperandLike],
+    condition: TensorLike,
+    x: TensorLike | None,
+    y: TensorLike | None,
     /,
     *,
-    device: DeviceLike = "cpu",
+    device: Device = "cpu",
     requires_grad: bool = False,
-) -> TensorLike: ...
+) -> TensorType: ...
 
 
 def where(
-    condition: OperandLike,
-    x: Optional[OperandLike] = None,
-    y: Optional[OperandLike] = None,
+    condition: TensorLike,
+    x: TensorLike | None = None,
+    y: TensorLike | None = None,
     /,
     *,
-    device: Optional[DeviceLike] = None,
+    device: Device | None = None,
     requires_grad: bool = False,
-) -> Union[Tuple[TensorLike, ...], TensorLike]:
+) -> Union[tuple[TensorType, ...], TensorType]:
     from ...tensor import Tensor
 
     if device is None:
@@ -68,9 +68,9 @@ def where(
 
     # * One parameter overload
     if (x is None) and (y is None):
-        data = to_xp_array(condition, device=device)
+        data = to_tensordata(condition, device=device)
         if device == "cpu":
-            result = np.where(data)
+            result = np.where(data)  # type: ignore
         else:
             if cp is None:
                 raise CuPyNotFound(CUPY_NOT_FOUND_MSG)
@@ -85,9 +85,9 @@ def where(
         ):
             raise DeviceMismatch(DEVICE_MISMATCH_MSG)
 
-        data = to_xp_array(condition, device=device)
-        x_data = to_xp_array(x, device=device)
-        y_data = to_xp_array(y, device=device)
+        data = to_tensordata(condition, device=device)
+        x_data = to_tensordata(x, device=device)
+        y_data = to_tensordata(y, device=device)
         if device == "cpu":
             result = np.where(data, x_data, y_data)
         else:
@@ -105,24 +105,24 @@ def where(
 
 
 def wrapper_2in_1out(
-    x1: OperandLike,
-    x2: OperandLike,
+    x1: TensorLike,
+    x2: TensorLike,
     /,
-    out: Optional[NDArrayLikeBool] = None,
+    out: TensorDataBool | None = None,
     *,
     func_name: str,
-    device: Optional[DeviceLike] = None,
-    where: Union[bool, NDArrayLikeBool] = True,
+    device: Device | None = None,
+    where: TensorDataBool | bool = True,
     casting: Casting = "same_kind",
     order: Order = "K",
     dtype: None = None,
     subok: bool = True,
-) -> TensorLike:
+) -> TensorType:
     from ...tensor import Tensor
 
     op_device = get_two_operand_op_device(x1, x2, device)
 
-    x1, x2 = to_xp_array(x1, device=op_device), to_xp_array(x2, device=op_device)
+    x1, x2 = to_tensordata(x1, device=op_device), to_tensordata(x2, device=op_device)
     if op_device == "cpu":
         y = getattr(np, func_name)(
             x1,
@@ -142,18 +142,18 @@ def wrapper_2in_1out(
 
 
 def equal(
-    x1: OperandLike,
-    x2: OperandLike,
+    x1: TensorLike,
+    x2: TensorLike,
     /,
-    out: Optional[NDArrayLikeBool] = None,
+    out: TensorDataBool | None = None,
     *,
-    device: Optional[DeviceLike] = None,
-    where: Union[bool, NDArrayLikeBool] = True,
+    device: Device | None = None,
+    where: TensorDataBool | bool = True,
     casting: Casting = "same_kind",
     order: Order = "K",
     dtype: None = None,
     subok: bool = True,
-) -> TensorLike:
+) -> TensorType:
     return wrapper_2in_1out(
         x1,
         x2,
@@ -169,18 +169,18 @@ def equal(
 
 
 def not_equal(
-    x1: OperandLike,
-    x2: OperandLike,
+    x1: TensorLike,
+    x2: TensorLike,
     /,
-    out: Optional[NDArrayLikeBool] = None,
+    out: TensorDataBool | None = None,
     *,
-    device: Optional[DeviceLike] = None,
-    where: Union[bool, NDArrayLikeBool] = True,
+    device: Device | None = None,
+    where: TensorDataBool | bool = True,
     casting: Casting = "same_kind",
     order: Order = "K",
     dtype: None = None,
     subok: bool = True,
-) -> TensorLike:
+) -> TensorType:
     return wrapper_2in_1out(
         x1,
         x2,
@@ -196,18 +196,18 @@ def not_equal(
 
 
 def greater(
-    x1: OperandLike,
-    x2: OperandLike,
+    x1: TensorLike,
+    x2: TensorLike,
     /,
-    out: Optional[NDArrayLikeBool] = None,
+    out: TensorDataBool | None = None,
     *,
-    device: Optional[DeviceLike] = None,
-    where: Union[bool, NDArrayLikeBool] = True,
+    device: Device | None = None,
+    where: TensorDataBool | bool = True,
     casting: Casting = "same_kind",
     order: Order = "K",
     dtype: None = None,
     subok: bool = True,
-) -> TensorLike:
+) -> TensorType:
     return wrapper_2in_1out(
         x1,
         x2,
@@ -223,18 +223,18 @@ def greater(
 
 
 def greater_equal(
-    x1: OperandLike,
-    x2: OperandLike,
+    x1: TensorLike,
+    x2: TensorLike,
     /,
-    out: Optional[NDArrayLikeBool] = None,
+    out: TensorDataBool | None = None,
     *,
-    device: Optional[DeviceLike] = None,
-    where: Union[bool, NDArrayLikeBool] = True,
+    device: Device | None = None,
+    where: TensorDataBool | bool = True,
     casting: Casting = "same_kind",
     order: Order = "K",
     dtype: None = None,
     subok: bool = True,
-) -> TensorLike:
+) -> TensorType:
     return wrapper_2in_1out(
         x1,
         x2,
@@ -250,18 +250,18 @@ def greater_equal(
 
 
 def less(
-    x1: OperandLike,
-    x2: OperandLike,
+    x1: TensorLike,
+    x2: TensorLike,
     /,
-    out: Optional[NDArrayLikeBool] = None,
+    out: TensorDataBool | None = None,
     *,
-    device: Optional[DeviceLike] = None,
-    where: Union[bool, NDArrayLikeBool] = True,
+    device: Device | None = None,
+    where: TensorDataBool | bool = True,
     casting: Casting = "same_kind",
     order: Order = "K",
     dtype: None = None,
     subok: bool = True,
-) -> TensorLike:
+) -> TensorType:
     return wrapper_2in_1out(
         x1,
         x2,
@@ -277,18 +277,18 @@ def less(
 
 
 def less_equal(
-    x1: OperandLike,
-    x2: OperandLike,
+    x1: TensorLike,
+    x2: TensorLike,
     /,
-    out: Optional[NDArrayLikeBool] = None,
+    out: TensorDataBool | None = None,
     *,
-    device: Optional[DeviceLike] = None,
-    where: Union[bool, NDArrayLikeBool] = True,
+    device: Device | None = None,
+    where: TensorDataBool | bool = True,
     casting: Casting = "same_kind",
     order: Order = "K",
     dtype: None = None,
     subok: bool = True,
-) -> TensorLike:
+) -> TensorType:
     return wrapper_2in_1out(
         x1,
         x2,

{depthtensor-2.4.0 → depthtensor-2.4.2}/src/DepthTensor/_core/ops/creation.py

@@ -1,19 +1,17 @@
-from typing import Optional, Literal, Union
+from typing import Literal
 
 from ...typing import (
+    TensorType,
     TensorLike,
     DTypeLike,
     Order,
-    AxisLike,
-    OperandLike,
-    DeviceLike,
-    ShapeLike,
-    NDArrayLike,
+    Axis,
+    Device,
+    Shape,
 )
 
 from ..exceptions import CuPyNotFound, CUPY_NOT_FOUND_MSG
-
-from ..utils import to_xp_array, get_device
+from ..utils import to_tensordata, get_device
 
 import numpy as np
 
@@ -28,23 +26,23 @@ except (ImportError, ModuleNotFoundError):
 
 
 def zeros_like(
-    a: OperandLike,
+    a: TensorLike,
     /,
     *,
-    device: Optional[DeviceLike] = None,
+    device: Device | None = None,
     requires_grad: bool = False,
-    dtype: Optional[DTypeLike] = None,
+    dtype: DTypeLike | None = None,
     order: Order = "K",
     subok: bool = True,
-    shape: Optional[AxisLike] = None,
-) -> TensorLike:
+    shape: Axis | None = None,
+) -> TensorType:
     from ...tensor import Tensor
 
     if device is None:
         device_op = get_device(a)
     else:
         device_op = device
-    a = to_xp_array(a)
+    a = to_tensordata(a)
     if device_op == "cpu":
         y = np.zeros_like(a, dtype=dtype, order=order, subok=subok, shape=shape)
     else:
@@ -55,23 +53,23 @@ def zeros_like(
 
 
 def ones_like(
-    a: OperandLike,
+    a: TensorLike,
     /,
     *,
-    device: Optional[DeviceLike] = None,
+    device: Device | None = None,
     requires_grad: bool = False,
-    dtype: Optional[DTypeLike] = None,
+    dtype: DTypeLike | None = None,
     order: Order = "K",
     subok: bool = True,
-    shape: Optional[AxisLike] = None,
-) -> TensorLike:
+    shape: Axis | None = None,
+) -> TensorType:
     from ...tensor import Tensor
 
     if device is None:
         device_op = get_device(a)
     else:
         device_op = device
-    a = to_xp_array(a)
+    a = to_tensordata(a)
     if device_op == "cpu":
         y = np.ones_like(a, dtype=dtype, order=order, subok=subok, shape=shape)
     else:
@@ -82,13 +80,13 @@ def ones_like(
 
 
 def zeros(
-    shape: ShapeLike,
+    shape: Shape,
     dtype: DTypeLike = float,
     order: Literal["C", "F"] = "C",
     *,
-    device: DeviceLike = "cpu",
+    device: Device = "cpu",
     requires_grad: bool = False,
-) -> TensorLike:
+) -> TensorType:
     from ...tensor import Tensor
 
     if device == "cpu":
@@ -101,13 +99,13 @@ def zeros(
 
 
 def ones(
-    shape: ShapeLike,
+    shape: Shape,
     dtype: DTypeLike = float,
     order: Literal["C", "F"] = "C",
     *,
-    device: DeviceLike = "cpu",
+    device: Device = "cpu",
     requires_grad: bool = False,
-) -> TensorLike:
+) -> TensorType:
     from ...tensor import Tensor
 
     if device == "cpu":