floatx

Implementation of floating point number types often used in deep neural networks (DNN).

• float8 E4M3 F8E4M3
• float8 E4M3Fn F8E4M3Fn
• float8 E5M2 F8E5M2
• float16 F16
• bfloat16 BF16
• float32 F32

See whole documentation at

No external dependency.

Has 100% code coverage: including testing the whole 256 possibilities for float8 and 65536 values for float16 types.