Q/jUdZddlZddlmZmZddlmZddlmZ ddl m Z m Z m Z mZmZmZmZmZmZmZmZgdZeeed<d Ze Z ee _d e _d ed dfd Zej:deee e ffdZd efdZy)zx This module contains functions and classes that alter the behavior of torch.nn.functional.scaled_dot_product_attention N)ListUnion)warn) _SDPBackend) can_use_efficient_attentioncan_use_flash_attentioncudnn_sdp_enabledenable_cudnn_sdpenable_flash_sdpenable_math_sdpenable_mem_efficient_sdpflash_sdp_enabledmath_sdp_enabledmem_efficient_sdp_enabled SDPAParams) SDPBackend sdpa_kernelWARN_FOR_UNFUSED_KERNELS__all__FrparamsreturnctrFt|stdt|dt|stdt|dyyy)z If WARN_FOR_UNFUSED_KERNELS is set to True, this will raise warnings for all the reasons why the fused kernels can't be run. If using subclasses z*Efficient attention can't be used because:Tz&Flash attention can't be used because:N)rrrr)rs ^/root/repos/pocs/video-evals/.venv/lib/python3.12/site-packages/torch/nn/attention/__init__.py_raise_kernel_warningsr8sG  *62 = > ' 5&v. 9 : #FD 1/  backendsc#~Kt|ttfsJdt|tr|g}t|}t }t }t }t} tj|v}tj|v}tj|v}tj|v}t|t|t|t|it|t|t|t|y#t|t|t|t|wxYww)a Context manager to select which backend to use for scaled dot product attention. .. warning:: This function is beta and subject to change. Args: backend (Union[List[SDPBackend], SDPBackend]): A backend or list of backends for scaled dot product attention. Example: .. code-block:: python from torch.nn.functional import scaled_dot_product_attention from torch.nn.attention import SDPBackend, sdpa_kernel # Only enable flash attention backend with sdpa_kernel(SDPBackend.FLASH_ATTENTION): scaled_dot_product_attention(...) # Enable the Math or Efficient attention backends with sdpa_kernel([SDPBackend.MATH, SDPBackend.EFFICIENT_ATTENTION]): scaled_dot_product_attention(...) This context manager can be used to select which backend to use for scaled dot product attention. Upon exiting the context manager, the previous state of the flags will be restored, enabling all backends. zKBackend must be an instance of SDPBackend or a list of SDPBackend instancesN) isinstancelistrsetr rrrCUDNN_ATTENTIONFLASH_ATTENTIONEFFICIENT_ATTENTIONMATHr r r r ) rprevious_cudnnprevious_flashprevious_mem_efficient previous_math enable_cudnn enable_flashenable_mem_efficient enable_maths rrrFs6 4$ UTU (J':8}H,.N,.N#<#>*,M'!11X= !11X= )==I oo1 && !56 $(( !78 & (( !78 &sA$D='A8D -D= .D::D=cy)zEThis returns the closest matching tag for the flash attention backendz2.5.7r.rr_get_flash_versionr/s r) __doc__ contextlibtypingrrwarningsrtorch._Crrtorch.backends.cudarrr r r r r rrrrrstr__annotations__r__name__ __module__rcontextmanagerrr/r.rrr;s~.    OcN!   ! "  2: 2$ 2 5'%Z 0* <=5'5'pCr