tensorplay@1.0.0.dev20260909

Perform a memory allocation using the CUDA memory allocator.

Not exposed by this TensorPlay build.

Delete memory allocated using the CUDA memory allocator.

Check if peer access between two devices is possible.

Change the currently used memory allocator (not supported by this build).

Raise an error if the result of a CUDA runtime API call is not success.

Return the clock speed of the GPU SM in MHz (megahertz) over the past sample period as given by nvidia-smi.

Retrieves the CUDA runtime API module.

This function initializes the CUDA runtime environment if it is not already initialized and returns the CUDA runtime API module (_cudart).

Returns:

The CUDA runtime API module, or None when no ctypes runtime binding is exposed by this build.

Return type:

module or None

Return cublasHandle_t pointer to current cuBLAS handle

Return the index of a currently selected device.

Return cusolverDnHandle_t pointer to current cuSOLVER handle

Return the currently selected Stream for a given device.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns the currently selected Stream for the current device, given by current_device(), if device is None (default).

Return the default Stream for a given device.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns the default Stream for the current device, given by current_device(), if device is None (default).

Return the number of GPUs available.

Return used global (device) memory in bytes as given by nvidia-smi.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns statistic for the current device, given by current_device(), if device is None (default).

Release all unoccupied cached memory currently held by the caching allocator so that those can be used in other GPU application and visible in nvidia-smi.

Returns the active allocator backend as a string. Always native here.

Return list CUDA architectures this library was compiled for.

Get the cuda capability of a device.

Parameters:

device (tensorplay.Device or int or str, optional) – device for which to return the device capability. This function is a no-op if this argument is a negative integer. It uses the current device, given by current_device(), if device is None (default).

Returns:

the major and minor cuda capability of the device

Return type:

tuple(int, int)

Get the name of a device.

Parameters:

device (tensorplay.Device or int or str, optional) – device for which to return the name. This function is a no-op if this argument is a negative integer. It uses the current device, given by current_device(), if device is None (default).

Returns:

the name of the device

Return type:

str

Get the properties of a device.

Parameters:

device (tensorplay.Device or int or str, optional) – device for which to return the properties of the device. It uses the current device, given by current_device(), if device is None (default).

Returns:

the properties of the device

Return type:

_CudaDeviceProperties

Return NVCC gencode flags this library was compiled with.

Get memory fraction for a process. Always returns 1.0 in this build.

Return a Stream from an externally allocated CUDA stream.

Return current value of debug mode for cuda synchronizing operations.

Always returns 0 in this build (mode is not enforced).

Return the result of host_memory_stats() as a nested dictionary.

Return a dictionary of pinned (host) allocator statistics.

Initialize TensorPlay’s CUDA state.

You may need to call this explicitly if you are interacting with TensorPlay via its C API, as Python bindings for CUDA functionality will not be available until this initialization takes place. Ordinary users should not need this, as all of TensorPlay’s CUDA methods automatically initialize CUDA state on-demand.

Does nothing if the CUDA state is already initialized.

Force collects GPU memory after it has been released by CUDA IPC.

Return a bool indicating if CUDA is currently available.

Return a bool indicating if the current CUDA device supports dtype bfloat16.

Return whether GDS is available. Always False in this build.

Return whether TensorPlay’s CUDA state has been initialized.

Return a bool indicating if the current CUDA device supports dtype tf32.

Return a human-readable printout of the running processes and their GPU memory use for a given device.

Return the maximum GPU memory occupied by tensors in bytes for a given device.

By default, this returns the peak allocated memory since the beginning of this program. reset_peak_memory_stats() can be used to reset the starting point in tracking this metric.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns statistic for the current device, given by current_device(), if device is None (default).

Return the maximum GPU memory managed by the caching allocator in bytes for a given device.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns statistic for the current device, given by current_device(), if device is None (default).

Return the global free and total GPU memory for a given device using cudaMemGetInfo.

Return the current GPU memory occupied by tensors in bytes for a given device.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns statistic for the current device, given by current_device(), if device is None (default).

Return the current GPU memory managed by the caching allocator in bytes for a given device.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns statistic for the current device, given by current_device(), if device is None (default).

Return a snapshot of the CUDA memory allocator state across all devices.

Interpreting the output of this function requires familiarity with the memory allocator internals. Not exposed by this TensorPlay build.

Return the result of memory_stats() as a nested dictionary.

The native allocator reports the fragmentation-aware matrix directly (segments, free-block histogram, pending cross-stream blocks, graph pools, capture state); it is exposed under "allocator" alongside the

Return a dictionary of CUDA memory allocator statistics for a given device.

The return value of this function is a dictionary of statistics, each of the full key layout; keys that are not tracked by this TensorPlay build are always reported as zero.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns statistics for the current device, given by current_device(), if device is None (default).

Return a human-readable printout of the current memory allocator statistics for a given device.

This can be useful to display periodically during training, or when handling out-of-memory exceptions.

Parameters:

• device (tensorplay.Device or int, optional) – selected device. Returns printout for the current device, given by current_device(), if device is None (default).

• abbreviated (bool, optional) – whether to return an abbreviated summary (default: False).

Return the percent of time over the past sample period during which global (device) memory was being read or written as given by nvidia-smi.

Warning: Each sample period may be between 1 second and 1/6 second, depending on the product being queried.

Return the average power draw of the GPU sensor in mW (MilliWatts) over the past sample period as given by nvidia-smi for Fermi or newer fully supported devices.

Reset the “accumulated” (historical) stats tracked by the host memory allocator.

Reset the “accumulated” (historical) stats tracked by the CUDA memory allocator.

This TensorPlay build does not track historical totals; the call is a no-op.

Reset the “peak” stats tracked by the host memory allocator.

Reset the “peak” stats tracked by the CUDA memory allocator.

Peak stats correspond to the “peak” key in each individual stat dict.

Parameters:

device (tensorplay.Device or int, optional) – selected device. Returns statistic for the current device, given by current_device(), if device is None (default).

Set the current device.

Usage of this function is discouraged in favor of device. In most cases it’s better to use CUDA_VISIBLE_DEVICES environmental variable.

Parameters:

device (tensorplay.Device or int) – selected device. This function is a no-op if this argument is negative.

Set memory fraction for a process.

Not enforced by this TensorPlay build; the signature is retained for API compatibility.

Parameters:

• fraction (float) – Range: 0~1. Allowed memory equals total_memory * fraction.

• device (tensorplay.Device or int, optional) – selected device.

Set the current stream. This is a wrapper API to set the stream.

Usage of this function is discouraged in favor of the stream context manager.

Parameters:

stream (Stream) – selected stream. This function is a no-op if this argument is None.

Set the debug mode for cuda synchronizing operations.

Not enforced by this TensorPlay build; the signature is retained for API compatibility.

Parameters:

debug_mode (str or int) – if “default” or 0, don’t error or warn on synchronizing operations, if “warn” or 1, warn on synchronizing operations, if “error” or 2, error out synchronizing operations.

Wrap around the Context-manager StreamContext that selects a given stream.

Parameters:

stream (Stream) – selected stream. This manager is a no-op if it’s None.

Wait for all kernels in all streams on a CUDA device to complete.

Parameters:

device (tensorplay.Device or int, optional) – device for which to synchronize. It uses the current device, given by current_device(), if device is None (default).

Return the average temperature of the GPU sensor in Degrees C (Centigrades).

The average temperature is computed based on past sample period as given by nvidia-smi.

Return the percent of time over the past sample period during which one or more kernels was executing on the GPU as given by nvidia-smi.

Warning: Each sample period may be between 1 second and 1/6 second, depending on the product being queried.