[https://nvbugs/5501820][fix] Add requirements for numba-cuda version to WAR mem corruption (#7992)

Signed-off-by: Pengbo Wang <[email protected]>

Author: pengbowang-nv

requirements.txt

@@ -75,3 +75,4 @@ tiktoken
 blobfile
 openai-harmony==0.0.4
 nvidia-cutlass-dsl==4.2.1; python_version >= "3.10"
+numba-cuda>=0.19.0 # WAR for nvbugs/5501820

tests/unittest/_torch/multi_gpu/test_linear.py

@@ -22,8 +22,6 @@
 
 # needed since we reuse the mpi executor pool, first test running will leak a thread
 pytestmark = pytest.mark.threadleak(enabled=False)
-pytest.skip(reason="https://nvbugspro.nvidia.com/bug/5501820",
-            allow_module_level=True)
 
 
 def rms_norm(x: torch.Tensor, weight: torch.Tensor = None, eps: float = 1e-6):