Error: H-HIGH-H4_svd_fp32-s1

H/HIGH/H4_svd_fp32/seed1_ERROR/error.json

@@ -0,0 +1,18 @@
+{
+  "config": {
+    "group": "H",
+    "variant": "H4_svd_fp32",
+    "band": "HIGH",
+    "seed": 1,
+    "phase": 2,
+    "num_epochs": 1,
+    "batch_size": 256,
+    "overrides": {
+      "svd": "fp32"
+    },
+    "description": "H-HIGH-H4_svd_fp32-s1"
+  },
+  "error": "cusolver error: CUSOLVER_STATUS_INVALID_VALUE, when calling `cusolverDnXsyevBatched_bufferSize( handle, params, jobz, uplo, n, CUDA_R_32F, reinterpret_cast<const void*>(A), lda, CUDA_R_32F, reinterpret_cast<const void*>(W), CUDA_R_32F, workspaceInBytesOnDevice, workspaceInBytesOnHost, batchSize)`. This error may appear if the input matrix contains NaN. If you keep seeing this error, you may use `torch.backends.cuda.preferred_linalg_library()` to try linear algebra operators with other supported backends. See https://pytorch.org/docs/stable/backends.html#torch.backends.cuda.preferred_linalg_library",
+  "traceback": "Traceback (most recent call last):\n  File \"/tmp/ipykernel_4109/3798867187.py\", line 305, in run_phase2\n    result = run_ablation_config(\n             ^^^^^^^^^^^^^^^^^^^^\n  File \"/tmp/ipykernel_4109/3142053339.py\", line 562, in run_ablation_config\n    out = model(images)\n          ^^^^^^^^^^^^^\n  File \"/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py\", line 1776, in _wrapped_call_impl\n    return self._call_impl(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/usr/local/lib/python3.12/dist-packages/torch/nn/modules/module.py\", line 1787, in _call_impl\n    return forward_call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/tmp/ipykernel_4109/2994023296.py\", line 408, in forward\n    svd = self.encode_patches(patches)\n          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/tmp/ipykernel_4109/3142053339.py\", line 166, in encode_patches\n    eigenvalues, Vmat = torch.linalg.eigh(G)\n                        ^^^^^^^^^^^^^^^^^^^^\ntorch._C._LinAlgError: cusolver error: CUSOLVER_STATUS_INVALID_VALUE, when calling `cusolverDnXsyevBatched_bufferSize( handle, params, jobz, uplo, n, CUDA_R_32F, reinterpret_cast<const void*>(A), lda, CUDA_R_32F, reinterpret_cast<const void*>(W), CUDA_R_32F, workspaceInBytesOnDevice, workspaceInBytesOnHost, batchSize)`. This error may appear if the input matrix contains NaN. If you keep seeing this error, you may use `torch.backends.cuda.preferred_linalg_library()` to try linear algebra operators with other supported backends. See https://pytorch.org/docs/stable/backends.html#torch.backends.cuda.preferred_linalg_library\n",
+  "timestamp": "2026-04-21T18:55:57.520932"
+}