Skip to content

fix: reduce datacube Delaunay release memory #143

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

Merged
Jammy2211 merged 1 commit into from
Jun 9, 2026
Merged

fix: reduce datacube Delaunay release memory #143

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 13 additions & 7 deletions scripts/jax_likelihood_functions/datacube/delaunay.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -235,10 +235,11 @@
print("JAX Time Taken per Likelihood:", (time.time() - start) / batch_size)

"""
Cube log-likelihood ≈ N × single-channel log-likelihood (-3165.42388511) for
identical channels. Pinned empirically below.
Cube log-likelihood ≈ N × single-channel log-likelihood for identical
channels. Pinned empirically below.
"""
EXPECTED_VMAP_LOG_LIKELIHOOD = n_channels * -3165.42388511
EXPECTED_SINGLE_CHANNEL_LOG_LIKELIHOOD = -3165.42388511
EXPECTED_VMAP_LOG_LIKELIHOOD = n_channels * EXPECTED_SINGLE_CHANNEL_LOG_LIKELIHOOD

np.testing.assert_allclose(
np.array(result),
Expand Down Expand Up @@ -278,9 +279,14 @@ def log_l_jit_fn(parameters):
"""
__Path B: TransformerNUFFT cross-check__

Re-run the same cube vmap with ``TransformerNUFFT`` and confirm the result
matches the ``TransformerDFT`` value.
Re-run a single-channel vmap with ``TransformerNUFFT`` and confirm the result
matches the single-channel ``TransformerDFT`` value. The 4-channel DFT path above
already validates datacube factor-graph summation and JIT routing; duplicating the
same Delaunay NUFFT graph across four identical channels can exceed the release
runner memory budget.
"""
nufft_channels = 1

dataset_list_nufft = [
al.Interferometer.from_fits(
data_path=path.join(dataset_path, "data.fits"),
Expand All @@ -289,7 +295,7 @@ def log_l_jit_fn(parameters):
real_space_mask=real_space_mask,
transformer_class=al.TransformerNUFFT,
)
for _ in range(n_channels)
for _ in range(nufft_channels)
]

analysis_list_nufft = [
Expand Down Expand Up @@ -329,7 +335,7 @@ def log_l_jit_fn(parameters):

np.testing.assert_allclose(
np.array(result_nufft),
EXPECTED_VMAP_LOG_LIKELIHOOD,
nufft_channels * EXPECTED_SINGLE_CHANNEL_LOG_LIKELIHOOD,
rtol=1e-4,
err_msg="datacube/delaunay: TransformerNUFFT cube vmap disagrees with TransformerDFT",
)
Expand Down
Loading