Giy ddlZddlmZmZddlmZej eZeZ dejdejjdejfdZ ddejjdejd ejd ejd ejdzd ed edzdedzdedzdedzdejdzdeejdffdZy)N)_flash_attention_forward!flash_attn_supports_top_left_mask)loggingquerymodulereturncR|jtjk(rtjdrtjdSt |j dr|j jStd|jDjjSy)ziIf the query is in float32, return a target dtype compatible with flash attention. Return None otherwise.cuda _is_quantizedc3jK|]+}t|tjjs(|-yw)N) isinstancetorchnnLinear).0layers k/var/www/pod-logistic/pod-ai/venv/lib/python3.12/site-packages/transformers/integrations/flash_attention.py z#get_target_dtype..s&b%z%QVQYQYQ`Q`?abs)33N) dtyperfloat32is_autocast_enabledget_autocast_dtypehasattrconfignextmodulesweight)rrs rget_target_dtyper sv {{emm#  $ $V ,++F3 3 V]]O 4==&& &b6>>+;bbiioo o keyvalueattention_maskdropoutscalingsliding_windowsoftcap is_causals_auxc  8| jddrtjd|jd} t d|jDr t d|j dd}|j dd}|j dd}t||} | | n |j} t||||f| | ||||t| |jjt|dr |jnd| | j|j ndd | }|dfS) Noutput_attentionsFzFlash Attention does not support `output_attentions=True`. Please set your attention to `eager` if you want any of these features.rc3&K|] }|dk( yw)rN)rdims rrz*flash_attention_forward..0s +3!8 +szTensor query has shape with a zero dimension. FlashAttention does not support inputs with dim=0. Please check your input shapes or use SDPA instead. layer_idx) query_lengthr(r$ softmax_scaler&r'use_top_left_mask target_dtypeattn_implementationr0r))getlogger warning_onceshapeany ValueError transposerr(r_use_top_left_maskr_attn_implementationrr0tor)rrr!r"r#r$r%r&r'r(r)kwargsseq_lenr4 attn_outputs rflash_attention_forwardrCs9zz%u- W kk!nG +u{{ ++ B  OOAq !E --1 C OOAq !E$E62L'2 8H8HI*     %,!"MM>>&-fk&B&""  HHU[[ !%( )K.  r )gNNNNN)rmodeling_flash_attention_utilsrrutilsr get_logger__name__r7r=TensorrModulerrfloatintbooltuplerCr-r rrNs3 h   H %68 ELL %((// ekk & !% !!%C HHOOC <<C C << C LL4' C  CT\C$JCT\Cd{C <<$ C 5<< Cr