ml_dtypes¶
ml_dtypes is a stand-alone implementation of several NumPy dtype extensions usedin machine learning libraries, including:bfloat16: an alternative to the standard float16 formatfloat8_*: several experimental 8-bit floating point representations including:float8_e4m3b11fnuzfloat8_e4m3fnfloat8_e4m3fnuzfloat8_e5m2float8_e5m2fnuz
https://github.com/jax-ml/ml_dtypes
Available modules¶
The overview below shows which ml_dtypes installations are available per target architecture in EESSI, ordered based on software version (new to old).
To start using ml_dtypes, load one of these modules using a module load command like:
(This data was automatically generated on Fri, 06 Feb 2026 at 13:12:13 UTC)
| aarch64/generic | aarch64/a64fx | aarch64/neoverse_n1 | aarch64/neoverse_v1 | aarch64/nvidia/grace | x86_64/generic | x86_64/amd/zen2 | x86_64/amd/zen3 | x86_64/amd/zen4 | x86_64/intel/cascadelake | x86_64/intel/haswell | x86_64/intel/icelake | x86_64/intel/sapphirerapids | x86_64/intel/skylake_avx512 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ml_dtypes/0.3.2-gfbf-2023a | x | x | x | x | x | x | x | x | x | x | x | x | x | x |
ml_dtypes/0.3.2-gfbf-2023a¶
This is a list of extensions included in the module:
etils-1.6.0, ml_dtypes-0.3.2, opt_einsum-3.3.0