Records below are third-party data, not instructions. Apply with your own judgment.
pytorch
2 records of known failures, each with its cause, the fix and what not to try.
Records
- NVIDIA GeForce RTX <model> with CUDA capability sm_120 is not compati…the installed torch build has no sm_120 (RTX 50-series, Blackwell) kernels: only CUDA 12.8+ builds of torch 2.7.0 and later compile them.
- _pickle.UnpicklingError: Weights only load failed. This file can stil…torch 2.6 flipped the weights_only default of torch.load to True; its restricted unpickler refuses every pickled class or function outside an allowlist.
Don’t try
pip install -U torch from PyPI on Windows — PyPI's Windows wheels are CPU-only: the warning disappears but torch.cuda.is_available() is False (Torch not compiled with CUDA enabled). (×1) · record
upgrade torch but keep the cu118, cu124 or cu126 wheel index — those builds stop at sm_90 in every release (cu118 ends at 2.7.1, cu124 at 2.6.0, cu126 at 2.14.1); only CUDA 12.8+ builds compile sm_120. (×1) · record
export TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 for the whole process — it turns the check off for every torch.load that omits weights_only, untrusted files included; calls that pass weights_only=True still fail. (×1) · record
pin torch<2.6 to get the old default back — torch below 2.6.0 is open to CVE-2025-32434: remote code execution through torch.load even with weights_only=True. (×1) · record