mlx/python/src
Daniel Strobusch dfdb284e16
make behaviour of dtype arguments consistent and compliant to numpy (#379)
All functions that take an optional dtype should

* have a default dtype visible in the generated docs (accomplished via `"dtype"_a = std::optional{float32}`)
* behave identical when `dtype=None` or no dtype is passed

This important when passing kw args down from a numpy function like:

```
def f(x, dtype=None):
  mx.random.uniform(dtype=dtype)
  # ...
```

NumPy functions behave like this.

It also fixes a minor bug in `tri`: #378

Closes #378
2024-01-05 09:37:46 -08:00
..
array.cpp support python mlx.array creation from list of mlx.array's (#325) 2024-01-04 18:53:33 -08:00
CMakeLists.txt linalg.norm (#187) 2023-12-26 19:42:04 -08:00
device.cpp copyright + ack 2023-11-30 11:12:53 -08:00
fft.cpp copyright + ack 2023-11-30 11:12:53 -08:00
indexing.cpp Indexing bug (#233) 2023-12-20 10:44:01 -08:00
indexing.h copyright + ack 2023-11-30 11:12:53 -08:00
linalg.cpp linalg.norm (#187) 2023-12-26 19:42:04 -08:00
load.cpp Fix load compilation (#298) 2023-12-27 06:20:45 -08:00
load.h Safetensor support (#215) 2023-12-27 02:06:55 -08:00
metal.cpp copyright + ack 2023-11-30 11:12:53 -08:00
mlx.cpp linalg.norm (#187) 2023-12-26 19:42:04 -08:00
ops.cpp make behaviour of dtype arguments consistent and compliant to numpy (#379) 2024-01-05 09:37:46 -08:00
random.cpp make behaviour of dtype arguments consistent and compliant to numpy (#379) 2024-01-05 09:37:46 -08:00
stream.cpp copyright + ack 2023-11-30 11:12:53 -08:00
transforms.cpp Corrected the example of value_and_grad (#274) 2023-12-23 11:06:38 -08:00
utils.h Mlx array accessor (#128) 2023-12-11 13:42:55 -08:00