deep-atomic 0.2.0__tar.gz → 0.2.1__tar.gz

deep_atomic-0.2.1/.pre-commit-config.yaml

@@ -0,0 +1,7 @@
+repos:
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.15.21
+    hooks:
+      - id: ruff-check
+        args: [--fix, --exit-non-zero-on-fix]
+      - id: ruff-format

{deep_atomic-0.2.0 → deep_atomic-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deep_atomic
-Version: 0.2.0
+Version: 0.2.1
 Summary: A simple deep learning framework built upon numpy only
 Author-email: Elliot Zhang <elliot_zh@proton.me>
 License: MIT
@@ -8,11 +8,12 @@ License-File: LICENSE
 Classifier: License :: OSI Approved :: MIT License
 Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3
-Requires-Python: >=3.8
+Requires-Python: >=3.10
 Requires-Dist: numpy>=1.24.4
 Requires-Dist: pytest>=8.3.5
 Provides-Extra: dev
 Requires-Dist: black>=23.0; extra == 'dev'
+Requires-Dist: pre-commit>=4.6.0; extra == 'dev'
 Requires-Dist: pytest-cov>=4.0; extra == 'dev'
 Requires-Dist: pytest>=7.0; extra == 'dev'
 Requires-Dist: ruff>=0.1.0; extra == 'dev'
@@ -80,13 +81,13 @@ c = da.maximum(a, b)
 c = da.fmin(a, b)
 c = da.minimum(a, b)
 
-c = d & e                     # element-wise and
-c = d | e                     # element-wise or
-c = d ^ e                     # element-wise xor
-c = da.logical_not(d)         # element-wise not
-c = d.all(axis=-1)            # reduction of and operation
-c = d.any(axis=-1)            # reduction of or operation
-c = da.where(d, a, b)         # return elements chosen from a or b depending on condition
+c = da.logical_and(d, e)                     # element-wise and
+c = da.logical_or(d, e)                      # element-wise or
+c = da.logical_xor(d, e)                     # element-wise xor
+c = da.logical_not(d)                        # element-wise not
+c = d.all(axis=-1, keepdims=False)           # reduction of and operation. axis=None, keepdims=False by default
+c = d.any(axis=-1, keepdims=False)           # reduction of or operation. axis=None, keepdims=False by default
+c = da.where(d, a, b)                        # return elements chosen from a or b depending on condition
 
 c = da.topk(a, 2, axis=-1, largest=True)    # same as pytorch. axis=-1, largest=True by default
 
@@ -94,6 +95,7 @@ c = da.sum(a)                               # shape: (1,)
 c = da.sum(a, axis=1)                       # shape: (3,)
 c = da.sum(a, axis=1, keepdims=True)        # shape: (3, 1)
 # min, max, argmin, argmax follow the same signature
+# all reductions set axis=None, keepdims=False by default
 
 c = da.softmax(a, axis=-1, temperature=0.6) # support temperature. temperature=1 by default
 c = da.log_softmax(a, axis=-1, temperature=0.6)
@@ -132,22 +134,29 @@ print(res.grad) # gradient get!
     - [x] gather or take_along_axis
     - [ ] scatter
     - [ ] convolution and 2d convolution
+    - [ ] pooling
+    - [ ] softmax attention
     - [ ] direct masking and indexing via `[]` syntax
     - [ ] einsum
+    - [ ] normalization
 - [ ] support backward with Vector-Jacobian Product like pytorch
-- [ ] basic neural network classes
+- [ ] basic neural network modules
 - [ ] optimizers and loss functions
+- [ ] dataset pipelines
+- [ ] save and load state dict file
 - [ ] full training test
 - [ ] benchmark with pytorch on CPU
-- [ ] attention layers and full LLM training
+- [ ] full LLM training
 - [ ] finer type annotations, comments and documentation
 
 ## Development
 
 Recommend manage dependencies using [uv](https://github.com/astral-sh/uv).
+We use [pre-commit](https://github.com/pre-commit/pre-commit) to manage hooks which help to lint and format our code.
 
 ```bash
 uv sync
+pre-commit install # install pre-commit git hooks for lint and format
 ```
 
 Run tests:

{deep_atomic-0.2.0 → deep_atomic-0.2.1}/README.md

@@ -60,13 +60,13 @@ c = da.maximum(a, b)
 c = da.fmin(a, b)
 c = da.minimum(a, b)
 
-c = d & e                     # element-wise and
-c = d | e                     # element-wise or
-c = d ^ e                     # element-wise xor
-c = da.logical_not(d)         # element-wise not
-c = d.all(axis=-1)            # reduction of and operation
-c = d.any(axis=-1)            # reduction of or operation
-c = da.where(d, a, b)         # return elements chosen from a or b depending on condition
+c = da.logical_and(d, e)                     # element-wise and
+c = da.logical_or(d, e)                      # element-wise or
+c = da.logical_xor(d, e)                     # element-wise xor
+c = da.logical_not(d)                        # element-wise not
+c = d.all(axis=-1, keepdims=False)           # reduction of and operation. axis=None, keepdims=False by default
+c = d.any(axis=-1, keepdims=False)           # reduction of or operation. axis=None, keepdims=False by default
+c = da.where(d, a, b)                        # return elements chosen from a or b depending on condition
 
 c = da.topk(a, 2, axis=-1, largest=True)    # same as pytorch. axis=-1, largest=True by default
 
@@ -74,6 +74,7 @@ c = da.sum(a)                               # shape: (1,)
 c = da.sum(a, axis=1)                       # shape: (3,)
 c = da.sum(a, axis=1, keepdims=True)        # shape: (3, 1)
 # min, max, argmin, argmax follow the same signature
+# all reductions set axis=None, keepdims=False by default
 
 c = da.softmax(a, axis=-1, temperature=0.6) # support temperature. temperature=1 by default
 c = da.log_softmax(a, axis=-1, temperature=0.6)
@@ -112,22 +113,29 @@ print(res.grad) # gradient get!
     - [x] gather or take_along_axis
     - [ ] scatter
     - [ ] convolution and 2d convolution
+    - [ ] pooling
+    - [ ] softmax attention
     - [ ] direct masking and indexing via `[]` syntax
     - [ ] einsum
+    - [ ] normalization
 - [ ] support backward with Vector-Jacobian Product like pytorch
-- [ ] basic neural network classes
+- [ ] basic neural network modules
 - [ ] optimizers and loss functions
+- [ ] dataset pipelines
+- [ ] save and load state dict file
 - [ ] full training test
 - [ ] benchmark with pytorch on CPU
-- [ ] attention layers and full LLM training
+- [ ] full LLM training
 - [ ] finer type annotations, comments and documentation
 
 ## Development
 
 Recommend manage dependencies using [uv](https://github.com/astral-sh/uv).
+We use [pre-commit](https://github.com/pre-commit/pre-commit) to manage hooks which help to lint and format our code.
 
 ```bash
 uv sync
+pre-commit install # install pre-commit git hooks for lint and format
 ```
 
 Run tests:

{deep_atomic-0.2.0 → deep_atomic-0.2.1}/pyproject.toml

@@ -4,10 +4,10 @@ build-backend = "hatchling.build"
 
 [project]
 name = "deep_atomic"
-version = "0.2.0"
+version = "0.2.1"
 description = "A simple deep learning framework built upon numpy only"
 readme = "README.md"
-requires-python = ">=3.8"
+requires-python = ">=3.10"
 license = { text = "MIT" }
 authors = [
     { name = "Elliot Zhang", email = "elliot_zh@proton.me" }
@@ -28,6 +28,7 @@ dev = [
     "pytest-cov>=4.0",
     "black>=23.0",
     "ruff>=0.1.0",
+    "pre-commit>=4.6.0"
 ]
 
 [tool.black]
@@ -36,9 +37,14 @@ target-version = ['py38']
 
 [tool.ruff]
 line-length = 88
-lint.select = ["E", "F", "W", "I"]
+lint.select = ["I"] # currently only use isort
 lint.fixable = ["I"]
 
+[tool.ruff.lint.isort]
+known-third-party = ["numpy"]
+known-first-party = ["deep-atomic"]
+force-sort-within-sections = true
+
 [tool.pytest.ini_options]
 testpaths = ["tests"]
 python_files = "*_test.py"

{deep_atomic-0.2.0 → deep_atomic-0.2.1}/src/deep_atomic/__init__.py

@@ -1,7 +1,7 @@
 """deep-atomic - A simple deep learning framework built upon numpy only"""
 
-__version__ = "0.2.0"
+__version__ = "0.2.1"
 __author__ = "elliot_zh@proton.me"
 
-from .tensor import *
 from .op import *
+from .tensor import *