[bug#78726,5/5] gnu: python-captum: Fix tests for PyTorch 2.7.0.

Message ID 3646842b94f3ae370a54b7bf9d856c1a6b524271.1749395917.git.bvits@riseup.net
State New
Headers
Series Update PyTorch to 2.7.0 |

Commit Message

Ayan Das June 8, 2025, 3:22 p.m. UTC
  * gnu/packages/machine-learning.scm (python-captum)[arguments]: Skip
test_exp_sets_with_diffent_lengths which fails with PyTorch 2.7.0's
stricter torch.load weights_only behavior.

Signed-off-by: Ayan Das <bvits@riseup.net>
---
 gnu/packages/machine-learning.scm | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)
  

Patch

diff --git a/gnu/packages/machine-learning.scm b/gnu/packages/machine-learning.scm
index 9636e2cd19..c1621d1e0b 100644
--- a/gnu/packages/machine-learning.scm
+++ b/gnu/packages/machine-learning.scm
@@ -5611,11 +5611,14 @@  (define-public python-captum
     (arguments
      (list
       #:test-flags
-      '(list "-k"
-             ;; These two tests (out of more than 1000 tests) fail because of
-             ;; accuracy problems.
-             "not test_softmax_classification_batch_multi_target\
- and not test_softmax_classification_batch_zero_baseline")))
+      '(list "-k" (string-append
+                   ;; These two tests (out of more than 1000 tests) fail because of
+                   ;; accuracy problems.
+                   "not test_softmax_classification_batch_multi_target"
+                   " and not test_softmax_classification_batch_zero_baseline"
+                   ;; This test fails with PyTorch 2.7.0 due to stricter
+                   ;; torch.load weights_only behavior.
+                   " and not test_exp_sets_with_diffent_lengths"))))
     (propagated-inputs
      (list python-matplotlib python-numpy python-pytorch python-tqdm))
     (native-inputs (list jupyter