No project description provided
Project description
ml_dtypes
ml_dtypes
is a stand-alone implementation of several NumPy dtype extensions used in machine learning libraries, including:
bfloat16
: an alternative to the standardfloat16
formatfloat8_*
: several experimental 8-bit floating point representations including:float8_e4m3b11fnuz
float8_e4m3fn
float8_e4m3fnuz
float8_e5m2
float8_e5m2fnuz
int4
anduint4
: low precision integer types.
See below for specifications of these number formats.
Installation
The ml_dtypes
package is tested with Python versions 3.9-3.12, and can be installed
with the following command:
pip install ml_dtypes
To test your installation, you can run the following:
pip install absl-py pytest
pytest --pyargs ml_dtypes
To build from source, clone the repository and run:
git submodule init
git submodule update
pip install .
Example Usage
>>> from ml_dtypes import bfloat16
>>> import numpy as np
>>> np.zeros(4, dtype=bfloat16)
array([0, 0, 0, 0], dtype=bfloat16)
Importing ml_dtypes
also registers the data types with numpy, so that they may
be referred to by their string name:
>>> np.dtype('bfloat16')
dtype(bfloat16)
>>> np.dtype('float8_e5m2')
dtype(float8_e5m2)
Specifications of implemented floating point formats
bfloat16
A bfloat16
number is a single-precision float truncated at 16 bits.
Exponent: 8, Mantissa: 7, exponent bias: 127. IEEE 754, with NaN and inf.
float8_e4m3b11fnuz
Exponent: 4, Mantissa: 3, bias: 11.
Extended range: no inf, NaN represented by 0b1000'0000.
float8_e4m3fn
Exponent: 4, Mantissa: 3, bias: 7.
Extended range: no inf, NaN represented by 0bS111'1111.
The fn
suffix is for consistency with the corresponding LLVM/MLIR type, signaling this type is not consistent with IEEE-754. The f
indicates it is finite values only. The n
indicates it includes NaNs, but only at the outer range.
float8_e4m3fnuz
8-bit floating point with 3 bit mantissa.
An 8-bit floating point type with 1 sign bit, 4 bits exponent and 3 bits mantissa. The suffix fnuz
is consistent with LLVM/MLIR naming and is derived from the differences to IEEE floating point conventions. F
is for "finite" (no infinities), N
for with special NaN encoding, UZ
for unsigned zero.
This type has the following characteristics:
- bit encoding: S1E4M3 -
0bSEEEEMMM
- exponent bias: 8
- infinities: Not supported
- NaNs: Supported with sign bit set to 1, exponent bits and mantissa bits set to all 0s -
0b10000000
- denormals when exponent is 0
float8_e5m2
Exponent: 5, Mantissa: 2, bias: 15. IEEE 754, with NaN and inf.
float8_e5m2fnuz
8-bit floating point with 2 bit mantissa.
An 8-bit floating point type with 1 sign bit, 5 bits exponent and 2 bits mantissa. The suffix fnuz
is consistent with LLVM/MLIR naming and is derived from the differences to IEEE floating point conventions. F
is for "finite" (no infinities), N
for with special NaN encoding, UZ
for unsigned zero.
This type has the following characteristics:
- bit encoding: S1E5M2 -
0bSEEEEEMM
- exponent bias: 16
- infinities: Not supported
- NaNs: Supported with sign bit set to 1, exponent bits and mantissa bits set to all 0s -
0b10000000
- denormals when exponent is 0
int4
and uint4
4-bit integer types, where each element is represented unpacked (i.e., padded up to a byte in memory).
NumPy does not support types smaller than a single byte. For example, the
distance between adjacent elements in an array (.strides
) is expressed in
bytes. Relaxing this restriction would be a considerable engineering project.
The int4
and uint4
types therefore use an unpacked representation, where
each element of the array is padded up to a byte in memory. The lower four bits
of each byte contain the representation of the number, whereas the upper four
bits are ignored.
Quirks of low-precision Arithmetic
If you're exploring the use of low-precision dtypes in your code, you should be
careful to anticipate when the precision loss might lead to surprising results.
One example is the behavior of aggregations like sum
; consider this bfloat16
summation in NumPy (run with version 1.24.2):
>>> from ml_dtypes import bfloat16
>>> import numpy as np
>>> rng = np.random.default_rng(seed=0)
>>> vals = rng.uniform(size=10000).astype(bfloat16)
>>> vals.sum()
256
The true sum should be close to 5000, but numpy returns exactly 256: this is
because bfloat16
does not have the precision to increment 256
by values less than
1
:
>>> bfloat16(256) + bfloat16(1)
256
After 256, the next representable value in bfloat16 is 258:
>>> np.nextafter(bfloat16(256), bfloat16(np.inf))
258
For better results you can specify that the accumulation should happen in a
higher-precision type like float32
:
>>> vals.sum(dtype='float32').astype(bfloat16)
4992
In contrast to NumPy, projects like JAX which support low-precision arithmetic more natively will often do these kinds of higher-precision accumulations automatically:
>>> import jax.numpy as jnp
>>> jnp.array(vals).sum()
Array(4992, dtype=bfloat16)
License
This is not an officially supported Google product.
The ml_dtypes
source code is licensed under the Apache 2.0 license
(see LICENSE). Pre-compiled wheels are built with the
EIGEN project, which is released under the
MPL 2.0 license (see LICENSE.eigen).
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distributions
File details
Details for the file ml_dtypes-0.3.1.tar.gz
.
File metadata
- Download URL: ml_dtypes-0.3.1.tar.gz
- Upload date:
- Size: 691.7 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 60778f99194b4c4f36ba42da200b35ef851ce4d4af698aaf70f5b91fe70fc611 |
|
MD5 | 6adbb05530819bdb4a78d2372d187fe2 |
|
BLAKE2b-256 | 166e9a7a51ee1ca24b8e92109128260c5aec8340c8fe5572e9ceecddae559abe |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp312-cp312-win_amd64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp312-cp312-win_amd64.whl
- Upload date:
- Size: 128.1 kB
- Tags: CPython 3.12, Windows x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | a777928dcba8865ab4a8157eeb25d23aed7bc82e5fd74e1d5eca821d3f148b39 |
|
MD5 | 08707b3270b1cf85ef12fb2c400805ef |
|
BLAKE2b-256 | 299092bf12f9202f1f21c7bc2973461a7136e8bfd5ea491a1c256b18e5aaa513 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 206.6 kB
- Tags: CPython 3.12, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | d1a8dc3bac1da2a17d0e2e4cba36ee89721d0bd33ea4765af2eefb5f41409e0f |
|
MD5 | cb91a0f3600d71e003efcf12a9128112 |
|
BLAKE2b-256 | 13208fa6e50ee5a7574bb53ccc7f15c2802d247e610cca28d238da1bab71b7d1 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
- Upload date:
- Size: 201.0 kB
- Tags: CPython 3.12, manylinux: glibc 2.17+ ARM64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 4828b62fa3bf1ae35faa40f3db9a38ec72fbce02f328a1d14c3a9da4606af364 |
|
MD5 | 49f3bc1a8fa513c5f4412e1634bcae7b |
|
BLAKE2b-256 | 2e9c305a78b2087573136fd0b13c9bf4c63612eec24031294af197986be24c8a |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp312-cp312-macosx_10_9_universal2.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp312-cp312-macosx_10_9_universal2.whl
- Upload date:
- Size: 350.6 kB
- Tags: CPython 3.12, macOS 10.9+ universal2 (ARM64, x86-64)
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 3d8ca0acbd377082792d8b97081ba580abdad67c6afb7f827012c675b052f058 |
|
MD5 | 56d0828a13195acddee55d075b98c638 |
|
BLAKE2b-256 | ebabaf35d8db985a39b5bf39e65b9eac993c192c22d25942000a8f6e72f69f4b |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp311-cp311-win_amd64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp311-cp311-win_amd64.whl
- Upload date:
- Size: 127.1 kB
- Tags: CPython 3.11, Windows x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | cb0c404e0dd3e25b56362c1c1e5de0ef717f727dde59fa721af4ce6ab2acca44 |
|
MD5 | f60713ff01525029d33681a9bf345680 |
|
BLAKE2b-256 | 4f9548c66f80acb9f91c3c2fd0cc6939457b8b0c1bd0d2b96edb461a5209df80 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 206.7 kB
- Tags: CPython 3.11, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 42a8980afd8b7c8e270e8b5c260237286b5b26acd276fcb758d13cd7cb567e99 |
|
MD5 | 702b1ec00ba0046f2e3a4953767c8158 |
|
BLAKE2b-256 | 5fb052580c12377c7f9a4319a80c85e235f4eeb7f8a09d362900dd3091b246ce |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
- Upload date:
- Size: 202.1 kB
- Tags: CPython 3.11, manylinux: glibc 2.17+ ARM64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 5727effa7650f7ab10906542d137cfb3244fdc3b2b519beff42f82def8ba59be |
|
MD5 | 56d43117753abf5345fd029965ee6b29 |
|
BLAKE2b-256 | abbab3579c4a7845fdbbe4b1488fe20bb84837f65735500c2e4ecde7dacaa08a |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp311-cp311-macosx_10_9_universal2.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp311-cp311-macosx_10_9_universal2.whl
- Upload date:
- Size: 348.1 kB
- Tags: CPython 3.11, macOS 10.9+ universal2 (ARM64, x86-64)
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 5e0b0b6bb07fa5ad11bb61d174667176bee5e05857225067aabfc5adc1b51d23 |
|
MD5 | e32fb318a284b310d1c1212b73c43c2d |
|
BLAKE2b-256 | 2254f50df872e42beffe566e29b6ef6cc9daf3901160c9e5290e254aad54560f |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp310-cp310-win_amd64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp310-cp310-win_amd64.whl
- Upload date:
- Size: 126.9 kB
- Tags: CPython 3.10, Windows x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | da274599e4950a9b488d21571061f49a185537cc77f2d3f8121151d58a9e9f16 |
|
MD5 | 157c8450da4508a732e4b4f7fda8d971 |
|
BLAKE2b-256 | 58cc570cf7a14db8d30ab2edc7a0b506aa9d8fdf7780a089e4ca9bff96f50fe9 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 206.7 kB
- Tags: CPython 3.10, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | fcae2c69715410d96906e1dfe8f017d9f78a0d10e0df91aae52e91f51fdfe45e |
|
MD5 | 0633f590705de4a951f485ca2ca9847f |
|
BLAKE2b-256 | 9031ec94e33a799323a8c37d1883f44b517c38d9defa7667db97cba212384d71 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
- Upload date:
- Size: 202.1 kB
- Tags: CPython 3.10, manylinux: glibc 2.17+ ARM64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | f83ff080df8910c0f987f615b03e4f8198638e0c00c6e679ea8892dda909763b |
|
MD5 | ea40fd430caf4f1e6d3fabc78e8a4cb8 |
|
BLAKE2b-256 | f0d59cfcfb866343e770cbe190a0b7290da357f8c75a6d959e2ae935472e44f6 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp310-cp310-macosx_10_9_universal2.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp310-cp310-macosx_10_9_universal2.whl
- Upload date:
- Size: 348.1 kB
- Tags: CPython 3.10, macOS 10.9+ universal2 (ARM64, x86-64)
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 510d249a91face47211762eb294d6fe64f325356b965fb6388c1bf51bd339267 |
|
MD5 | 9fcbd55789064ae0d1ca913dacc5eb15 |
|
BLAKE2b-256 | 976a8ceeace58c85188d844cc1ab76e12ec7ad0980a769a953387554fcbbd530 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp39-cp39-win_amd64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp39-cp39-win_amd64.whl
- Upload date:
- Size: 126.7 kB
- Tags: CPython 3.9, Windows x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 979d7d196d9a17e0135ae22878f74241fbd3522cef58d7b292f1fd5b32282201 |
|
MD5 | 4bd15a5cab8dccc0c418930401f80668 |
|
BLAKE2b-256 | 2be8cc9c1736bd9ba89f5d3fe9a931770bf4c56721e1f56b18d6bcad085cc3b8 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 207.0 kB
- Tags: CPython 3.9, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 4d94b2d1bed77284694f7fd0479640fa7aa5d96433dca3cbcec407a5ef752e77 |
|
MD5 | 6dc15aba3df91af2d6c74bf68ab0e315 |
|
BLAKE2b-256 | 343618f70d9eda70be1d17a8092c287acd5deea2b916bf8f3ce8d4299045eb11 |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
- Upload date:
- Size: 202.3 kB
- Tags: CPython 3.9, manylinux: glibc 2.17+ ARM64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 70984b473db6489ec1d8c79b082a1322105155193049d08a3b0c515094e9777b |
|
MD5 | 3f23d20662dd884c1ca5bd94b55d32b3 |
|
BLAKE2b-256 | 74f62d4c6a971dd4b8ce3c43c062c04dd51673c0f45475560a7ba0680103654a |
Provenance
File details
Details for the file ml_dtypes-0.3.1-cp39-cp39-macosx_10_9_universal2.whl
.
File metadata
- Download URL: ml_dtypes-0.3.1-cp39-cp39-macosx_10_9_universal2.whl
- Upload date:
- Size: 348.1 kB
- Tags: CPython 3.9, macOS 10.9+ universal2 (ARM64, x86-64)
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.9.18
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 438437e2e614a3c91d75581653b6c40ec890e8b5994d7190a90c931740151c95 |
|
MD5 | 759a0183c81227003446114f9a2fbdc9 |
|
BLAKE2b-256 | 4e7b6bb9ea497a9575f0cf52c12324583b2eacc21efb29b4fb4f1d447018939f |