Skip to content

Fix memory leak in transforms (#2841) #2853

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
harsh-sutariya wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix memory leak in transforms (#2841) #2853

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
294eec7
Fix memory leak in transforms (#2841)
harsh-sutariya Dec 2, 2025
63ce0c1
Merge branch 'main' into fix/2841-memory-leak-transform-container
harsh-sutariya Dec 3, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 18 additions & 2 deletions mlx/transforms.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -337,7 +337,15 @@ std::pair<std::vector<array>, std::vector<array>> vjp(
for (auto& p : primals) {
auto s = p.has_primitive() ? p.primitive().stream()
: default_stream(default_device());
primals_.push_back(copy(p, s)); // Does not do a deep copy
array source = p;
if (!p.is_tracer()) {
while (source.has_primitive() &&
typeid(source.primitive()) == typeid(Copy) &&
!source.inputs().empty()) {
source = source.inputs()[0];
}
}
primals_.push_back(copy(source, s)); // Does not do a deep copy
primals_.back().set_tracer(true);
}

Expand Down Expand Up @@ -545,7 +553,15 @@ std::pair<std::vector<array>, std::vector<array>> jvp(
for (auto& p : primals) {
auto s = p.has_primitive() ? p.primitive().stream()
: default_stream(default_device());
primals_.push_back(copy(p, s)); // Does not do a deep copy
array source = p;
if (!p.is_tracer()) {
while (source.has_primitive() &&
typeid(source.primitive()) == typeid(Copy) &&
!source.inputs().empty()) {
source = source.inputs()[0];
}
}
primals_.push_back(copy(source, s)); // Does not do a deep copy
primals_.back().set_tracer(true);
}
auto outputs = fun(primals_);
Expand Down
22 changes: 22 additions & 0 deletions python/tests/test_autograd.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -798,6 +798,28 @@ def loss_fn(model):
grad_fn(model)
self.assertEqual(model[1].item(), 2.0)

def test_grad_with_container_reuse(self):
container = [mx.array(1.0)]

def fn(p, x):
container[0] = p
return x.sum()

x = mx.ones(shape=(128,))
grad_fn = mx.grad(fn)

mx.synchronize()
gc.collect()
mem_pre = mx.get_active_memory()

for _ in range(20):
mx.eval(grad_fn(container[0], x))
gc.collect()

mx.synchronize()
mem_post = mx.get_active_memory()
self.assertLess(mem_post - mem_pre, 1024 * 1024)

def test_reduce_jvp(self):
a = mx.arange(4)
b = mx.array([3, 2, 1, 0])
Expand Down