Update mlx-c to 0.1.2 #204
Conversation
|
See also #206 -- I wasn't done with it but I set up a PR with what I have. This may go together |
Co-authored-by: DavidKoski <46639364+davidkoski@users.noreply.github.com>
Source/MLX/MLXArray+Init.swift
Outdated
| public convenience init<T: HasDType>(_ value: T) { | ||
| let floatMax = Double(Float.greatestFiniteMagnitude) | ||
| let doubleValue = value as? Double | ||
| if doubleValue != nil && doubleValue! < floatMax && doubleValue! > -floatMax { |
There was a problem hiding this comment.
The idea here is to convert a scalar to float64 only if it lies outside the range representable for float32. We could also just always cast to float32 and let the user do MLXArray(0.1, dtype: .float64) if they want a float64.
There was a problem hiding this comment.
This might be better as init(_ value: Double) -- otherwise this logic is use for bools, ints, etc. It should still work, but I think specializing it to Double expresses the intent better.
There was a problem hiding this comment.
Sounds good I'll change it.
There was a problem hiding this comment.
I remembered why I put it here:
When you initialize a MLXArray from a range e.g. in
let a = MLXRandom.uniform(0.0 ..< 1.0, [4, 10, 12, 4])
You're redirected to this initialization rather than init(_ value: Double).
Do you know if there's a better way to overload in that case as well?
Update mlx-c to 0.1.2 and mlx to 0.23.1
I added a few of the new methods:
MLXLinalg.luMLXLinalg.lu_factorMLXLinalg.solveMLXLinalg.solve_triangularflattenunflattenkron