[BUG] mlx.core.random.randint always returns the same value

[Redempt1onzzZZ]

Describe the bug
When I use ''mlx.core.random.randint'' to generate random integers from the given interval, I notice that every time I run test.py, mlx.core.random.randint always returns the same value

To Reproduce

Include code snippet

import mlx.core as mx
a = mx.random.randint(0, 65535, shape=(2, 2), dtype=mx.uint64)
print(a)

Expected behavior
Each run of the program should output a different random array

Desktop (please complete the following information):

• OS Version: MacOS 14.2.1
• Version 0.7.0

Additional context
The problem still exists when I construct multiple arrays

import mlx.core as mx
a = mx.random.randint(0, 65535, shape=(2, 2), dtype=mx.uint64)
print(a)
b = mx.random.randint(0, 255, shape=(2, 2), dtype=mx.uint8)
print(b)

[AvikantSrivastava]

@awni can I pick this up?

[awni]

Definitiely. This one should be a quick fix, we have a fixed seed in the global RNG... it probably makes sense to use a seed based on date/time or something that varies enough.

Relevant lines https://github.com/ml-explore/mlx/blob/main/mlx/random.h#L20-L22

[Redempt1onzzZZ]

Thank you very much for your response to the issue I've reported. I'm currently conducting research involving fuzz testing, and the classification of this issue could significantly impact my findings. In this context, could you kindly clarify if this issue could be considered and labeled as a bug? This would help me better understand the scope of issues that are considered bugs.

[awni]

This issue is a gray area in terms of calling it a bug. To me it's working as expected (since that is the result we expect given how we set up the RNG), but since it's not expected behavior for most people, calling it a bug is not unreasonable.

[awni]

FWIW most issues are not in such a gray area, this one is kind of unusual.

[Redempt1onzzZZ]

Okay, I have gained a deeper understanding through your answer. Thank you for your explanation and support

[AvikantSrivastava]

@awni can you take a look this PR

[AvikantSrivastava]

@Redmept1on does this look okay to you? I have raised a PR for the fix
test.py

import mlx.core as mx
a = mx.random.randint(0, 65535, shape=(2, 2), dtype=mx.uint64)
print(a)

[Redempt1onzzZZ]

Thank you for your work. After reviewing your PR, I have a question. Does this change only apply to uint64? Or is this method applicable to other data types? @AvikantSrivastava

[LeonEricsson]

Thank you for your work. After reviewing your PR, I have a question. Does this change only apply to uint64? Or is this method applicable to other data types? @AvikantSrivastava

no this change is independent of whatever dtype you want your mx.array to be

[Redempt1onzzZZ]

Thx! I'm very sorry for missing your comment. @LeonEricsson
The issue seems to have been resolved in mlx v0.10.0. I'm wondering if I can close this issue. @awni

[awni]

Yup, sorry I meant to close it, thanks!