If you are referring to shader code, the metal shading language does not use float8 and float16. Instead, you use half for float16. There is no float8 equivalent. If you're writing to a texture, you can use a format like r8unorm or r8snorm, but those are normalized values; i.e. in [0, 1] or [-1, 1] respectively. And you would write a half or float to it where the gpu implicitly casts.

I was running Tahoe beta 4. It is fixed upon updating to Tahoe beta 5