Update develop for potential v8.2

.github/workflows/autoblack.yml

@@ -0,0 +1,44 @@
+# GitHub Action that uses Black to reformat all Python code and submits a PR
+# in regular intervals. Inspired by: https://github.com/cclauss/autoblack
+
+name: autoblack
+on:
+  workflow_dispatch:  # allow manual trigger
+  schedule:
+    - cron: '0 8 * * 5'  # every Friday at 8am UTC
+
+jobs:
+  autoblack:
+    if: github.repository_owner == 'explosion'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+        with:
+            ref: ${{ github.head_ref }}
+      - uses: actions/setup-python@v2
+      - run: pip install black
+      - name: Auto-format code if needed
+        run: black thinc
+      # We can't run black --check here because that returns a non-zero excit
+      # code and makes GitHub think the action failed
+      - name: Check for modified files
+        id: git-check
+        run: echo ::set-output name=modified::$(if git diff-index --quiet HEAD --; then echo "false"; else echo "true"; fi)
+      - name: Create Pull Request
+        if: steps.git-check.outputs.modified == 'true'
+        uses: peter-evans/create-pull-request@v3
+        with:
+            title: Auto-format code with black
+            labels: meta
+            commit-message: Auto-format code with black
+            committer: GitHub <noreply@github.com>
+            author: explosion-bot <explosion-bot@users.noreply.github.com>
+            body: _This PR is auto-generated._
+            branch: autoblack
+            delete-branch: true
+            draft: false
+      - name: Check outputs
+        if: steps.git-check.outputs.modified == 'true'
+        run: |
+          echo "Pull Request Number - ${{ steps.cpr.outputs.pull-request-number }}"
+          echo "Pull Request URL - ${{ steps.cpr.outputs.pull-request-url }}"

.github/workflows/explosionbot.yml

@@ -23,5 +23,5 @@ jobs:
         env:
           INPUT_TOKEN: ${{ secrets.EXPLOSIONBOT_TOKEN }}
           INPUT_BK_TOKEN: ${{ secrets.BUILDKITE_SECRET }}
-          ENABLED_COMMANDS: "test_gpu"
+          ENABLED_COMMANDS: "test_gpu,test_slow,test_slow_gpu"
           ALLOWED_TEAMS: "spacy-maintainers"

README.md

@@ -2,7 +2,7 @@
 
 # Thinc: A refreshing functional take on deep learning, compatible with your favorite libraries
 
-### From the makers of [spaCy](https://spacy.io), [Prodigy](https://prodi.gy) and [FastAPI](https://fastapi.tiangolo.com)
+### From the makers of [spaCy](https://spacy.io) and [Prodigy](https://prodi.gy)
 
 [Thinc](https://thinc.ai) is a **lightweight deep learning library** that offers an elegant,
 type-checked, functional-programming API for **composing models**, with support

azure-pipelines.yml

@@ -23,7 +23,7 @@ jobs:
         imageName: 'windows-2019'
         python.version: '3.6'
       Python37Mac:
-        imageName: 'macos-10.15'
+        imageName: 'macos-latest'
         python.version: '3.7'
       Python38Linux:
         imageName: 'ubuntu-latest'
@@ -63,6 +63,7 @@ jobs:
   - script: |
       python -m mypy thinc
     displayName: 'Run mypy'
+    condition: ne(variables['python.version'], '3.6')
 
   - task: DeleteFiles@1
     inputs:
@@ -82,25 +83,37 @@ jobs:
 
   - script: |
       pip install -r requirements.txt
-      pip install "tensorflow~=2.5.0"
-      pip install "mxnet; sys_platform != 'win32'"
-      pip install "torch==1.9.0+cpu" -f https://download.pytorch.org/whl/torch_stable.html
       pip install ipykernel pydot graphviz
       python -m ipykernel install --name thinc-notebook-tests --user
-    displayName: 'Install test dependencies'
+      python -m pytest --pyargs thinc --cov=thinc --cov-report=term
+    displayName: 'Run tests without extras'
+
+  - script: |
+      pip install "protobuf~=3.20.0" "tensorflow~=2.5.0"
+      pip install "mxnet; sys_platform != 'win32'"
+      pip install torch --extra-index-url https://download.pytorch.org/whl/cpu
+      # torch does not have a direct numpy requirement but is compiled against
+      # a newer version than the oldest supported numpy for windows and
+      # python 3.10; this version of numpy would not work with
+      # tensorflow~=2.5.0 as specified above, but there is no release for
+      # python 3.10 anyway
+      pip install "numpy~=1.23.0; python_version=='3.10' and sys_platform=='win32'"
+      pip install -r requirements.txt
+      pip uninstall -y mypy
+    displayName: 'Install extras for testing'
 
   - script: |
       python -m pytest --pyargs thinc --cov=thinc --cov-report=term
-    displayName: 'Run tests'
+    displayName: 'Run tests with extras'
 
   - script: |
       pip uninstall -y tensorflow
       pip install thinc-apple-ops
       python -m pytest --pyargs thinc_apple_ops
     displayName: 'Run tests for thinc-apple-ops'
-    condition: and(startsWith(variables['imageName'], 'macos'), eq(variables['python.version'], '3.9'))
+    condition: and(startsWith(variables['imageName'], 'macos'), eq(variables['python.version'], '3.10'))
 
   - script: |
       python -m pytest --pyargs thinc
     displayName: 'Run tests with thinc-apple-ops'
-    condition: and(startsWith(variables['imageName'], 'macos'), eq(variables['python.version'], '3.9'))
+    condition: and(startsWith(variables['imageName'], 'macos'), eq(variables['python.version'], '3.10'))

build-constraints.txt

@@ -1,6 +1,8 @@
 # build version constraints for use with wheelwright + multibuild
-numpy==1.15.0; python_version<='3.7'
-numpy==1.17.3; python_version=='3.8'
+numpy==1.15.0; python_version<='3.7' and platform_machine!='aarch64'
+numpy==1.19.2; python_version<='3.7' and platform_machine=='aarch64'
+numpy==1.17.3; python_version=='3.8' and platform_machine!='aarch64'
+numpy==1.19.2; python_version=='3.8' and platform_machine=='aarch64'
 numpy==1.19.3; python_version=='3.9'
 numpy==1.21.3; python_version=='3.10'
 numpy; python_version>='3.11'

examples/transformers_tagger.py

@@ -132,7 +132,9 @@ def forward(
         return TokensPlus(**token_data), lambda d_tokens: []
 
     return Model(
-        "tokenizer", forward, attrs={"tokenizer": AutoTokenizer.from_pretrained(name)},
+        "tokenizer",
+        forward,
+        attrs={"tokenizer": AutoTokenizer.from_pretrained(name)},
     )
 
 
@@ -166,11 +168,14 @@ def convert_transformer_outputs(model, inputs_outputs, is_train):
 
     def backprop(d_tokvecs: List[Floats2d]) -> ArgsKwargs:
         # Restore entries for bos and eos markers.
+        shim = model.shims[0]
         row = model.ops.alloc2f(1, d_tokvecs[0].shape[1])
         d_tokvecs = [model.ops.xp.vstack((row, arr, row)) for arr in d_tokvecs]
         return ArgsKwargs(
             args=(torch_tokvecs,),
-            kwargs={"grad_tensors": xp2torch(model.ops.pad(d_tokvecs))},
+            kwargs={
+                "grad_tensors": xp2torch(model.ops.pad(d_tokvecs, device=shim.device))
+            },
         )
 
     return tokvecs, backprop

pyproject.toml

@@ -5,7 +5,7 @@ requires = [
     "murmurhash>=1.0.2,<1.1.0",
     "cymem>=2.0.2,<2.1.0",
     "preshed>=3.0.2,<3.1.0",
-    "blis>=0.4.0,<0.8.0",
+    "blis>=0.7.8,<0.8.0",
     "numpy>=1.15.0",
 ]
 build-backend = "setuptools.build_meta"

requirements.txt

@@ -2,17 +2,18 @@
 murmurhash>=1.0.2,<1.1.0
 cymem>=2.0.2,<2.1.0
 preshed>=3.0.2,<3.1.0
-blis>=0.4.0,<0.8.0
+blis>=0.7.8,<0.8.0
 srsly>=2.4.0,<3.0.0
 wasabi>=0.8.1,<1.1.0
 catalogue>=2.0.4,<2.1.0
+confection>=0.0.1,<1.0.0
 ml_datasets>=0.2.0,<0.3.0
 # Third-party dependencies
-pydantic>=1.7.4,!=1.8,!=1.8.1,<1.10.0
+pydantic>=1.7.4,!=1.8,!=1.8.1,<1.11.0
 numpy>=1.15.0
 # Backports of modern Python features
 dataclasses>=0.6,<1.0; python_version < "3.7"
-typing_extensions>=3.7.4.1,<4.0.0.0; python_version < "3.8"
+typing_extensions>=3.7.4.1,<4.2.0; python_version < "3.8"
 contextvars>=2.4,<3; python_version < "3.7"
 # Development dependencies
 cython>=0.25.0,<3.0
@@ -22,7 +23,7 @@ pytest-cov>=2.7.0,<2.8.0
 coverage>=5.0.0,<6.0.0
 mock>=2.0.0,<3.0.0
 flake8>=3.5.0,<3.6.0
-mypy>=0.901,<0.960
+mypy>=0.980,<0.990; platform_machine != "aarch64" and python_version >= "3.7"
 types-mock>=0.1.1
 types-contextvars>=0.1.2; python_version < "3.7"
 types-dataclasses>=0.1.3; python_version < "3.7"
@@ -33,3 +34,4 @@ nbconvert>=5.6.1,<6.2.0
 nbformat>=5.0.4,<5.2.0
 # Test to_disk/from_disk against pathlib.Path subclasses
 pathy>=0.3.5
+black>=22.0,<23.0

setup.cfg

@@ -35,24 +35,29 @@ setup_requires =
     cymem>=2.0.2,<2.1.0
     preshed>=3.0.2,<3.1.0
     murmurhash>=1.0.2,<1.1.0
-    blis>=0.4.0,<0.8.0
+    blis>=0.7.8,<0.8.0
 install_requires =
     # Explosion-provided dependencies
-    blis>=0.4.0,<0.8.0
+    blis>=0.7.8,<0.8.0
     murmurhash>=1.0.2,<1.1.0
     cymem>=2.0.2,<2.1.0
     preshed>=3.0.2,<3.1.0
     wasabi>=0.8.1,<1.1.0
     srsly>=2.4.0,<3.0.0
     catalogue>=2.0.4,<2.1.0
+    confection>=0.0.1,<1.0.0
     # Third-party dependencies
     setuptools
     numpy>=1.15.0
-    pydantic>=1.7.4,!=1.8,!=1.8.1,<1.10.0
+    pydantic>=1.7.4,!=1.8,!=1.8.1,<1.11.0
     # Backports of modern Python features
     dataclasses>=0.6,<1.0; python_version < "3.7"
-    typing_extensions>=3.7.4.1,<4.0.0.0; python_version < "3.8"
+    typing_extensions>=3.7.4.1,<4.2.0; python_version < "3.8"
     contextvars>=2.4,<3; python_version < "3.7"
+
+[options.entry_points]
+pytest_randomly.random_seeder =
+    thinc = thinc.api:fix_random_seed
 
 [options.extras_require]
 cuda =
@@ -83,6 +88,14 @@ cuda114 =
     cupy-cuda114>=5.0.0b4
 cuda115 =
     cupy-cuda115>=5.0.0b4
+cuda116 =
+    cupy-cuda116>=5.0.0b4
+cuda117 =
+    cupy-cuda117>=5.0.0b4
+cuda11x =
+    cupy-cuda11x>=11.0.0
+cuda-autodetect =
+    cupy-wheel>=11.0.0
 datasets =
     ml_datasets>=0.2.0,<0.3.0
 torch =

setup.py

@@ -17,14 +17,15 @@
 
 PACKAGES = find_packages()
 MOD_NAMES = [
+    "thinc.backends.cblas",
     "thinc.backends.linalg",
     "thinc.backends.numpy_ops",
     "thinc.extra.search",
     "thinc.layers.sparselinear",
 ]
 COMPILE_OPTIONS = {
     "msvc": ["/Ox", "/EHsc"],
-    "other": ["-O3", "-Wno-strict-prototypes", "-Wno-unused-function"],
+    "other": ["-O3", "-Wno-strict-prototypes", "-Wno-unused-function", "-std=c++11"],
 }
 COMPILER_DIRECTIVES = {
     "language_level": -3,

thinc/about.py

@@ -1,2 +1,2 @@
-__version__ = "8.0.15"
+__version__ = "8.1.3"
 __release__ = True

thinc/api.py

@@ -16,16 +16,18 @@
 from .util import DataValidationError, data_validation
 from .util import to_categorical, get_width, get_array_module, to_numpy
 from .util import torch2xp, xp2torch, tensorflow2xp, xp2tensorflow, mxnet2xp, xp2mxnet
+from .util import get_torch_default_device
+from .compat import has_cupy
 from .backends import get_ops, set_current_ops, get_current_ops, use_ops
-from .backends import Ops, CupyOps, NumpyOps, has_cupy, set_gpu_allocator
+from .backends import Ops, CupyOps, MPSOps, NumpyOps, set_gpu_allocator
 from .backends import use_pytorch_for_gpu_memory, use_tensorflow_for_gpu_memory
 
 from .layers import Dropout, Embed, expand_window, HashEmbed, LayerNorm, Linear
 from .layers import Maxout, Mish, MultiSoftmax, Relu, softmax_activation, Softmax, LSTM
 from .layers import CauchySimilarity, ParametricAttention, Logistic
 from .layers import resizable, sigmoid_activation, Sigmoid, SparseLinear
 from .layers import ClippedLinear, ReluK, HardTanh, HardSigmoid
-from .layers import HardSwish, HardSwishMobilenet, Swish, Gelu
+from .layers import Dish, HardSwish, HardSwishMobilenet, Swish, Gelu
 from .layers import PyTorchWrapper, PyTorchRNNWrapper, PyTorchLSTM
 from .layers import TensorFlowWrapper, keras_subclass, MXNetWrapper
 from .layers import PyTorchWrapper_v2, Softmax_v2
@@ -38,6 +40,7 @@
 from .layers import with_reshape, with_getitem, strings2arrays, list2array
 from .layers import list2ragged, ragged2list, list2padded, padded2list, remap_ids
 from .layers import array_getitem, with_cpu, with_debug, with_nvtx_range
+from .layers import with_signpost_interval
 from .layers import tuplify
 
 from .layers import reduce_first, reduce_last, reduce_max, reduce_mean, reduce_sum

thinc/backends/__init__.py

@@ -5,13 +5,15 @@
 import threading
 
 from .ops import Ops
-from .cupy_ops import CupyOps, has_cupy
+from .cupy_ops import CupyOps
 from .numpy_ops import NumpyOps
+from .mps_ops import MPSOps
 from ._cupy_allocators import cupy_tensorflow_allocator, cupy_pytorch_allocator
 from ._param_server import ParamServer
 from ..util import assert_tensorflow_installed, assert_pytorch_installed
-from ..util import is_cupy_array, set_torch_tensor_type_for_ops, require_cpu
+from ..util import get_torch_default_device, is_cupy_array, require_cpu
 from .. import registry
+from ..compat import cupy, has_cupy
 
 
 context_ops: ContextVar[Optional[Ops]] = ContextVar("context_ops", default=None)
@@ -46,9 +48,11 @@ def use_pytorch_for_gpu_memory() -> None:  # pragma: no cover
     We'd like to support routing Tensorflow memory allocation via PyTorch as well
     (or vice versa), but do not currently have an implementation for it.
     """
-    import cupy.cuda
-
     assert_pytorch_installed()
+
+    if get_torch_default_device().type != "cuda":
+        return
+
     pools = context_pools.get()
     if "pytorch" not in pools:
         pools["pytorch"] = cupy.cuda.MemoryPool(allocator=cupy_pytorch_allocator)
@@ -65,8 +69,6 @@ def use_tensorflow_for_gpu_memory() -> None:  # pragma: no cover
     We'd like to support routing PyTorch memory allocation via Tensorflow as
     well (or vice versa), but do not currently have an implementation for it.
     """
-    import cupy.cuda
-
     assert_tensorflow_installed()
     pools = context_pools.get()
     if "tensorflow" not in pools:
@@ -94,7 +96,7 @@ def get_ops(name: str, **kwargs) -> Ops:
 
     cls: Optional[Callable[..., Ops]] = None
     if name == "cpu":
-        _import_extra_cpu_backends()        
+        _import_extra_cpu_backends()
         cls = ops_by_name.get("numpy")
         cls = ops_by_name.get("apple", cls)
         cls = ops_by_name.get("bigendian", cls)
@@ -137,7 +139,6 @@ def set_current_ops(ops: Ops) -> None:
     """Change the current backend object."""
     context_ops.set(ops)
     _get_thread_state().ops = ops
-    set_torch_tensor_type_for_ops(ops)
 
 
 def contextvars_eq_thread_ops() -> bool:
@@ -173,6 +174,7 @@ def _create_thread_local(
     "ParamServer",
     "Ops",
     "CupyOps",
+    "MPSOps",
     "NumpyOps",
     "has_cupy",
 ]