`io8ddlmZddlZddlmZddlmZ er ddlmZeZ n e dn #e $rZ e e ZdZ YdZ [ ndZ [ wwxYw dd ejjd ejd ejd ejd eejdedejfdZdS))OptionalN)PagedAttentionCache)is_flash_attn_2_available)flash_attn_varlen_funczFlash Attention 2 is not installed. Please refer to https://huggingface.co/docs/transformers/perf_infer_gpu_one#flashattention-2 to install itc0tdt)Nz)flash_attn_varlen_func is not available: ) Exceptionmsg)argskwargss y/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/transformers/integrations/flash_paged.pyFLASH_ATTN_VARLEN_FUNCrsICIIJJJmoduleqkvattention_maskcachereturnc t|ddsdn |jdz df} | dkrdnd} ||j|||jfi| \}}t |t r|| }| | } | t | d r| j}nt}d | vrd | d ini}|| dd  d | | | tj| tj|| f|jd | d |}t |t$r|d}|dfS)aPerform the forward pass of attention with paged key-value cache. This function handles the cache updates and performs the attention computation using the flash_attn_varlen_func for efficient processing. Args: q: (total_q, nheads, headdim), where total_q = total number of query tokens in the batch. k: (total_k, nheads_k, headdim), where total_k = total number of key tokens in the batch. but if there is a block table it can be the full k v: (total_k, nheads_k, headdim), where total_k = total number of key tokens in the batch. but if there is a block table it can be the full v cu_seq_lens_q: (batch_size + 1,), dtype torch.int32. The cumulative sequence lengths of the sequences in the batch, used to index into q. cu_seq_lens_k: (batch_size + 1,), dtype torch.int32. The cumulative sequence lengths of the sequences in the batch, used to index into kv. max_seqlen_q: int. Maximum query sequence length in the batch. max_seqlen_k: int. Maximum key sequence length in the batch. dropout_p: float. Dropout probability. softmax_scale: float. The scaling of QK^T before applying softmax. Default to 1 / sqrt(headdim). causal: bool. Whether to apply causal attention mask (e.g., for auto-regressive modeling). window_size: (left, right). If not (-1, -1), implements sliding window local attention. softcap: float. Anything > 0 activates softcapping attention. sliding_windowF)rrfull_attentionsliding_attentionNrs_auxrT) softmax_scalecausal window_size)getattrrupdate layer_idx isinstancedicthasattrrrget transposesqueeze contiguoustotorchint32clonescalingtuple)rrrrrr cu_seq_lens_q cu_seq_lens_k max_seqlen_q max_seqlen_kimplementationr r layer_typer custom_kwargs attn_outputs r paged_attention_forwardr9sH&-V5Eu%M%MqXXTZTilmTmopSqN%3x%?%?!!EXJ u|Aq&"2==f==1-&&0%j1 #J/ !gn>V&W&W!!/!F!76=6G6GWfjj1122RM(( Aq!!!$$//11  %%%%++-- n"     K+u%%%!!n  r)NNNNNNN)typingrr,generation.continuous_batchingrutilsr flash_attnrr RuntimeErrorr ereprr nnModuleTensorr9rr rEs @@@@@@------ K  "" 555555!7l ]     KKK $q''CKKKKKKKKK.2!%FF HOF |F |F | F U\* F  F \FFFFFFs7AAA