You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
|`AppleOps`| <iname="yes"></i> | <iname="no"></i> | <iname="no"></i> | Use AMX matrix multiplication units on Apple Silicon Macs. Added in Thinc 9.0. |
23
+
|`CupyOps`| <iname="no"></i> | <iname="yes"></i> | <iname="no"></i> | Execute via [`cupy`](https://cupy.chainer.org/) and custom CUDA. |
24
+
|`MPSOps`| <iname="yes"></i> | <iname="yes"></i> | <iname="no"></i> | Use the GPU on Apple Silicon Macs for PyTorch models, use AMX matrix multiplication units for Thinc Models. |
25
+
|`NumpyOps`| <iname="yes"></i> | <iname="no"></i> | <iname="no"></i> | Execute via `numpy`, [`blis`](https://github.com/explosion/cython-blis) (optional) and custom Cython. |
24
26
25
27
## Ops {#ops tag="class"}
26
28
27
-
The `Ops` class is typically not used directly but via `NumpyOps` or `CupyOps`,
28
-
which are subclasses of `Ops` and implement a **more efficient subset of the
29
-
methods**. You also have access to the ops via the
29
+
The `Ops` class is typically not used directly but via `NumpyOps`, `AppleOps`,
30
+
`CupyOps` or `MPSOps`, which are subclasses of `Ops` and implement a **more
31
+
efficient subset of the methods**. You also have access to the ops via the
30
32
[`Model.ops`](/docs/api-model#attributes) attribute. The documented methods
31
33
below list which backends provide optimized and more efficient versions
32
34
(indicated by <iname="yes"></i>), and which use the default implementation.
0 commit comments