# tensorplay.backends.cuda API

Source: https://www.tensorplay.cn/docs/api/tensorplay.backends.cuda.html

## Functions 4

[#](#api-tensorplay.backends.cuda.allow_fp16_bf16_reduction_math_sdp)

### allow_fp16_bf16_reduction_math_sdp

function[Full reference ↗](/docs/generated/tensorplay.backends.cuda.allow_fp16_bf16_reduction_math_sdp.html)

```python
tensorplay.backends.cuda.allow_fp16_bf16_reduction_math_sdp(enabled: bool)
```

> **Warning**
>
> This flag is beta and subject to change.

Enables or disables fp16/bf16 reduction in math scaled dot product attention.

[#](#api-tensorplay.backends.cuda.can_use_cudnn_attention)

### can_use_cudnn_attention

function[Full reference ↗](/docs/generated/tensorplay.backends.cuda.can_use_cudnn_attention.html)

```python
tensorplay.backends.cuda.can_use_cudnn_attention(params: _SDPAParams, debug: bool = False) → bool
```

Check if cudnn_attention can be utilized in scaled_dot_product_attention.

Parameters:

- params – An instance of SDPAParams containing the tensors for query, key, value, an optional attention mask, dropout rate, and a flag indicating if the attention is causal.

- debug – Whether to logging.warn debug information as to why cuDNN attention could not be run. Defaults to False.

Returns:

True if cuDNN can be used with the given parameters; otherwise, False.

> **Note**
>
> This function is dependent on a CUDA-enabled build of TensorPlay. It will return False in non-CUDA environments.

[#](#api-tensorplay.backends.cuda.can_use_efficient_attention)

### can_use_efficient_attention

function[Full reference ↗](/docs/generated/tensorplay.backends.cuda.can_use_efficient_attention.html)

```python
tensorplay.backends.cuda.can_use_efficient_attention(params: _SDPAParams, debug: bool = False) → bool
```

Check if efficient_attention can be utilized in scaled_dot_product_attention.

Parameters:

- params – An instance of SDPAParams containing the tensors for query, key, value, an optional attention mask, dropout rate, and a flag indicating if the attention is causal.

- debug – Whether to logging.warn debug information as to why efficient_attention could not be run. Defaults to False.

Returns:

True if efficient_attention can be used with the given parameters; otherwise, False.

> **Note**
>
> This function is dependent on a CUDA-enabled build of TensorPlay. It will return False in non-CUDA environments.

[#](#api-tensorplay.backends.cuda.can_use_flash_attention)

### can_use_flash_attention

function[Full reference ↗](/docs/generated/tensorplay.backends.cuda.can_use_flash_attention.html)

```python
tensorplay.backends.cuda.can_use_flash_attention(params: _SDPAParams, debug: bool = False) → bool
```

Check if FlashAttention can be utilized in scaled_dot_product_attention.

Parameters:

- params – An instance of SDPAParams containing the tensors for query, key, value, an optional attention mask, dropout rate, and a flag indicating if the attention is causal.

- debug – Whether to logging.warn debug information as to why FlashAttention could not be run. Defaults to False.

Returns:

True if FlashAttention can be used with the given parameters; otherwise, False.

> **Note**
>
> This function is dependent on a CUDA-enabled build of TensorPlay. It will return False in non-CUDA environments.

## Classes 2

[#](#api-tensorplay.backends.cuda.cuBLASModule)

### cuBLASModule

class[Full reference ↗](/docs/generated/tensorplay.backends.cuda.cuBLASModule.html)

```python
class tensorplay.backends.cuda.cuBLASModule
```

[#](#api-tensorplay.backends.cuda.cuFFTPlanCache)

### cuFFTPlanCache

class[Full reference ↗](/docs/generated/tensorplay.backends.cuda.cuFFTPlanCache.html)

```python
class tensorplay.backends.cuda.cuFFTPlanCache(device_index)
```

Represent a specific plan cache for a specific device_index.

The attributes size and max_size, and method clear, can fetch and/or
change properties of the C++ cuFFT plan cache.