Ting Lu

In latest attempt, only seeing this error https://github.com/pytorch/pytorch/actions/runs/19554104324/job/56001084245 ``` 2025-11-21T02:46:33.0040241Z RuntimeError: cuDNN version incompatibility: PyTorch was compiled against (9, 13, 0) but found runtime version (9, 10, 2). PyTorch already...

In the build log for inductor-build-cuda13, https://github.com/pytorch/pytorch/actions/runs/19566532187/job/56029790307 seeing, testing if explicitly set USE_CUDNN=1 would work. ``` 2025-11-21T12:22:18.4867535Z -- USE_CUDNN is set to 0. Compiling without cuDNN support 2025-11-21T12:22:18.4868043Z -- USE_CUSPARSELT...

The mismatch seems to be an issue between runtime CUDNN lib and compiled CUDNN ver. Setting export LD_LIBRARY_PATH=/usr/local/cuda-13.0/lib64 for CUDA 13 so that it is able to find the correct...

Actual fix should be remove the hardcoded installation for cu128 wheel in the inductor setup script. In the inductor benchmark installation script [pytorch/.ci/docker/common/install_inductor_benchmark_deps.sh](https://github.com/pytorch/pytorch/blob/a5436a5e8e4ee42d1debf52c2786c7ae0043a434/.ci/docker/common/install_inductor_benchmark_deps.sh#L39) `pip_install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128 `...

@pytorchbot rebase

Error: ``` Unable to find image 'pytorch/conda-builder:cuda12.6' locally docker: Error response from daemon: manifest for pytorch/conda-builder:cuda12.6 not found: manifest unknown: manifest unknown. ``` Need to create cuda12.6 in https://hub.docker.com/r/pytorch/conda-builder/tags

Thanks @Wenaka2004! for me I got this error after upgrading to 0.43.3 ``` self = name_or_ordinal = 'cquantize_blockwise_bf16_nf4' def __getitem__(self, name_or_ordinal): > func = self._FuncPtr((name_or_ordinal, self)) E AttributeError: /usr/local/lib/python3.10/dist-packages/bitsandbytes/libbitsandbytes_cpu.so: undefined...

@pytorchbot rebase

@pytorchbot rebase

@pytorchbot rebase -b main