^jSU`dZddlZddlmZddlmZddlZddlmZm Z dZ eGdd Z dd ejdzd ejezezd ejd ededzf dZ dd ejdzd ejezezd ejd ededzf dZddejdej&dedzfdZddejdej&dedzfdZ dd ejezezdej&dej,d ededzdejdzf dZy)a IMPORTANT NOTICE: Every class and function in this file is deprecated in favor of using the much more general `masking_utils.py` primitives. New code should not rely on it, it is only kept for backward compatibility for now, and will be removed in the future. N) dataclass)Union)is_torchdynamo_compiling is_tracingzThe attention mask API under `transformers.modeling_attn_mask_utils` (`AttentionMaskConverter`) is deprecated and will be removed in Transformers v5.10. Please use the new API in `transformers.masking_utils`.c^eZdZUdZeed<eed<ddededzfdZ ddededed ejd e ejd fd ejdzf d Z ddejded ejdedzd ejf dZe ddej d ejd ejdededzf dZeddejd ejdedzfdZedej&defdZe d dejdzdejdededzded ef dZy)!AttentionMaskConvertera9 A utility attention mask class that allows one to: - Create a causal 4d mask - Create a causal 4d mask with slided window - Convert a 2d attention mask (batch_size, query_length) to a 4d attention mask (batch_size, 1, query_length, key_value_length) that can be multiplied with attention scores Examples: ```python >>> import torch >>> from transformers.modeling_attn_mask_utils import AttentionMaskConverter >>> converter = AttentionMaskConverter(True) >>> converter.to_4d(torch.tensor([[0, 0, 0, 1, 1]]), 5, key_value_length=5, dtype=torch.float32) tensor([[[[-3.4028e+38, -3.4028e+38, -3.4028e+38, -3.4028e+38, -3.4028e+38], [-3.4028e+38, -3.4028e+38, -3.4028e+38, -3.4028e+38, -3.4028e+38], [-3.4028e+38, -3.4028e+38, -3.4028e+38, -3.4028e+38, -3.4028e+38], [-3.4028e+38, -3.4028e+38, -3.4028e+38, 0.0000e+00, -3.4028e+38], [-3.4028e+38, -3.4028e+38, -3.4028e+38, 0.0000e+00, 0.0000e+00]]]]) ``` Parameters: is_causal (`bool`): Whether the attention mask should be a uni-directional (causal) or bi-directional mask. sliding_window (`int`, *optional*): Optionally, the sliding window masks can be created if `sliding_window` is defined to a positive integer. is_causalsliding_windowNctjtt||_||_|j )|j dkrt d|j dyy)NrzaMake sure that when passing `sliding_window` that its value is a strictly positive integer, not ``)warningswarnDEPRECATION_MESSAGE FutureWarningr r ValueError)selfr r s p/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/modeling_attn_mask_utils.py__init__zAttentionMaskConverter.__init__Fsw )=9",    *t/B/Ba/GstxuHuHtIIJK 0H * batch_size query_lengthkey_value_lengthdtypedevicestrreturnc|jstd|jd||f}||z }d}|ddkDs |j |j |||||j}|S)z Creates a causal 4D mask of (bsz, head_dim=1, query_length, key_value_length) shape and adds large negative bias to upper right hand triangular matrix (causal mask). z"Please use `to_causal_4d` only if z has `is_causal` set to True.Nrrpast_key_values_lengthr )r r __class__r _make_causal_mask) rrrrrr input_shaper!causal_4d_masks r to_causal_4dz#AttentionMaskConverter.to_causal_4dQs~~A$..AQQnop p"<0 !1L!@ r?Q $"5"5"A!33'=#22 4Nrattention_mask_2dc|jd|f}d}|ddkDs |jI|jr=| td||z }|j |||j ||j}n|j t d|j|||dj|j }|=|j|jtj|j}|} | S) a Converts 2D attention mask to 4D attention mask by expanding mask to (bsz, head_dim=1, query_length, key_value_length) shape and by adding a large negative bias to not-attended positions. If attention_mask is causal, a causal mask will be added. rNrrzpThis attention mask converter is causal. Make sure to pass `key_value_length` to correctly create a causal mask.r z?Sliding window is currently only implemented for causal masking)tgt_len)shaper r rr#rNotImplementedError _expand_maskto masked_fillbooltorchfinfomin) rr'rrrr$r%r!expanded_attn_maskexpanded_4d_masks rto_4dzAttentionMaskConverter.to_4drs))..q1<@  Oa 4#6#6#B' G&6 %D "!33(//'=#22 4N ,%&gh h"../@%Q\]_Q`.add  $ $   %!/!;!;  )y1}&:&:499R=!&LLaPwwu~ !A %99ekk'3IQV_efhlmsuvD  %->BH ::eood%**&MX`aL()zz|   lEKK,>,B,B CD$1$%,,S!WgH^>^__rrHr)ctjtt|j \}}||n|}|ddddddfj |d||j |}tjd||z }|j|j tjtj|jS)zg Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`. Nr?r:) rrrrr>rFr-r0tensorr.r/r1r2)rHrr)rGsrc_len expanded_mask inverted_masks rr,z#AttentionMaskConverter._expand_masks  )=9yy{ W$0'gQdA-.55c1gwORRSXY  S6F (()9)9%**)Eu{{SXGYG]G]^^rrO min_dtypectjtt|jt j k(r td|jt j||k(ddS)a Attend to all tokens in masked rows from the expanded attention mask, for example the relevant first rows when using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path. Details: https://github.com/pytorch/pytorch/issues/110213 `expanded_mask` is [bsz, num_masks, tgt_seq_len, src_seq_len] or [bsz, tgt_seq_len, src_seq_len]. `attention_mask` is [bsz, src_seq_len]. The dimension num_masks of `expanded_mask` is most often 1, but it can also be the number of heads in the case of alibi attention bias. For example, if `expanded_mask` is (e.g. here left-padding case) ``` [[[[0, 0, 0], [0, 0, 0], [0, 0, 1]]], [[[1, 0, 0], [1, 1, 0], [1, 1, 1]]], [[[0, 0, 0], [0, 1, 0], [0, 1, 1]]]] ``` then the modified `expanded_mask` will be ``` [[[[1, 1, 1], <-- modified [1, 1, 1], <-- modified [0, 0, 1]]], [[[1, 0, 0], [1, 1, 0], [1, 1, 1]]], [[[1, 1, 1], <-- modified [0, 1, 0], [0, 1, 1]]]] ``` z\AttentionMaskConverter._unmask_unattended expects a float `expanded_mask`, got a BoolTensor.rT)r9keepdim) rrrrrr0r/rmulall)rOrQs r_unmask_unattendedz)AttentionMaskConverter._unmask_unattendedsaR  )=9   %** ,n   %))MY,FBX\"]!]^^rattention_mask inputs_embeds is_trainingcdtjtt|jd|jd}}||z}t |}d} ||s|s|dk(s||k(r |||krd} | S|||kr?t |jdk(ry|s$tj|dk(r |dk(s||k(rd} | S)a9 Detects whether the optional user-specified attention_mask & the automatically created causal mask can be ignored in case PyTorch's SDPA is used, rather relying on SDPA's `is_causal` argument. In case no token is masked in the `attention_mask` argument, if `query_length == 1` or `key_value_length == query_length`, we rather rely on SDPA `is_causal` argument to use causal/non-causal masks, allowing to dispatch to the flash attention kernel (that can otherwise not be used if a custom `attn_mask` is passed). rrFT) rrrrr*rlenr0rU) rWrXr!r rY_rr is_tracing_ignore_causal_masks r_ignore_causal_mask_sdpaz/AttentionMaskConverter._ignore_causal_mask_sdpas"  )=9'--a0-2E2Ea2H<'*@@ / "  !K!Q&*:l*J#+/?./P%)""! #'7.'H>''(A- UYY~/B%C1$(8L(H)-&"!rN)cpurN)NF)__name__ __module__ __qualname____doc__r/__annotations__intrr0rrrTensorr&r5 staticmethodSizer#r, FloatTensorfloatrVr`rrr r #s<O $ d ".3   {{  ellE)*   L(, -  <<- - {{ - * -  - ^ '(%) !`!`{{!` !`!$ !` d !`!`F _5<< _ _cDj _ _0_((0_0_0_d &*! 8" t+8"||8"!$8"d 8"  8"  8"8"rr rWr$rXr!r ctd|}|d|z}|*>&?1&D,22 KO>NVcViVi3 ( #  #N,@,@(AQ(F%a.![^=MN %% &. 85eNtj| tj|jj } | S) a Prepares the correct `attn_mask` argument to be used by `torch.nn.functional.scaled_dot_product_attention`. In case no token is masked in the `attention_mask` argument, we simply set it to `None` for the cases `query_length == 1` and `key_value_length == query_length`, and rely instead on SDPA `is_causal` argument to use causal/non-causal masks, allowing to dispatch to the flash attention kernel (that can otherwise not be used if a custom `attn_mask` is passed). Trqr)rWrXr!r Nrr8r[)rr)cudaxpu)rQ) r rr`r&rrr9r5typerVr0r1r2) rWr$rXr!r rurr^r_r4s r*_prepare_4d_causal_attention_mask_for_sdpar|ysB14P^_"2)?? ]+K/HH%#5% I0 /  .;; NKO-=]EXEXanauau< , %    1 $- 288B#))!1 9  /66;;N5HH EKK 8K8K,L,P,P I   rrHrr)c2tj|||S) Creates a non-causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape `(batch_size, key_value_length)` Args: mask (`torch.Tensor`): A 2D attention mask of shape `(batch_size, key_value_length)` dtype (`torch.dtype`): The torch dtype the created mask shall have. tgt_len (`int`): The target length or query length the created mask shall have. rHrr))r r,rs r_prepare_4d_attention_maskrs " . .Dw . WWrctjtt|j\}}||n|}t |st j|dk(rytj|||S)r~Nrr) rrrrr*rr0rUr r,)rHrr)r]rs r#_prepare_4d_attention_mask_for_sdparsd MM%}5**A ,g2BG d  $!) 4%22ESZ2[[rrrchtd|}||dz}|j|d|d|||}|S)a/ Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` Args: input_shape (`tuple(int)` or `list(int)` or `torch.Size`): The input shape should be a tuple that defines `(batch_size, query_length)`. dtype (`torch.dtype`): The torch dtype the created mask shall have. device (`int`): The torch device the created mask shall have. sliding_window (`int`, *optional*): If the model uses windowed attention, a sliding window should be passed. Trqrrr8)r r&)r$rrr!r rurrWs r _create_4d_causal_attention_maskrsR(14P^_- B?(55A B)9v6N rrarc)rgr dataclassesrtypingrr0utils.import_utilsrrrr rjrlrtlistrirwr|rrrrrrorrrs ! Dw  ]"]" ]"J "& 1LL4'1e#d*1<<1 1 $J 1t"& 7LL4'7e#d*7<<7 7 $J 7t XU\\ X%++ XPSVZPZ X \ell\5;;\Y\_cYc\:#$!% e#d* ;; LL  $J   \\D r