PYTORCH_NO_CUDA_MEMORY_CACHING

If set to1, disables caching of memory allocations in CUDA. This can be useful for debugging.

PYTORCH_CUDA_ALLOC_CONF

For a more in depth explanation of this environment variable, seeMemory management.

PYTORCH_NVML_BASED_CUDA_CHECK

If set to1, before importing PyTorch modules that check if CUDA is available, PyTorch will use NVML to check if the CUDA driver is functional instead of using the CUDA runtime. This can be helpful if forked processes fail with a CUDA initialization error.

TORCH_CUDNN_V8_API_LRU_CACHE_LIMIT

The cache limit for the cuDNN v8 API. This is used to limit the memory used by the cuDNN v8 API. The default value is 10000, which roughly corresponds to 2GiB assuming 200KiB per ExecutionPlan. Set to0 for no limit or a negative value for no caching.

TORCH_CUDNN_V8_API_DISABLED

If set to1, disables the cuDNN v8 API. And will fall back to the cuDNN v7 API.

TORCH_ALLOW_TF32_CUBLAS_OVERRIDE

If set to1, forces TF32 enablement, overridesset_float32_matmul_precision setting.

TORCH_NCCL_USE_COMM_NONBLOCKING

If set to1, enables non-blocking error handling in NCCL.

TORCH_NCCL_AVOID_RECORD_STREAMS

If set to0, enables fallback to record streams-based synchronization behavior in NCCL.

TORCH_CUDNN_V8_API_DEBUG

If set to1, sanity check whether cuDNN V8 is being used.

CUDA_VISIBLE_DEVICES

Comma-separated list of GPU device IDs that should be made available to CUDA runtime. If set to-1, no GPUs are made available.

CUDA_LAUNCH_BLOCKING

If set to1, makes CUDA calls synchronous. This can be useful for debugging.

CUBLAS_WORKSPACE_CONFIG

This environment variable is used to set the workspace configuration for cuBLAS per allocation. The format is:[SIZE]:[COUNT].As an example, the default workspace size per allocation isCUBLAS_WORKSPACE_CONFIG=:4096:2:16:8 which specifies a total size of2*4096+8*16KiB.To force cuBLAS to avoid using workspaces, setCUBLAS_WORKSPACE_CONFIG=:0:0.

CUDNN_CONV_WSCAP_DBG

Similar toCUBLAS_WORKSPACE_CONFIG, this environment variable is used to set the workspace configuration for cuDNN per allocation.

CUBLASLT_WORKSPACE_SIZE

Similar toCUBLAS_WORKSPACE_CONFIG, this environment variable is used to set the workspace size for cuBLASLT.

CUDNN_ERRATA_JSON_FILE

Can be set to a file path for an errata filter that can be passed to cuDNN to avoid specific engine configs, used primarily for debugging or to hardcode autotuning.

NVIDIA_TF32_OVERRIDE

If set to0, disables TF32 globally across all kernels, overriding all PyTorch settings.