o eiU@sTdZddlmZddlmZddlZddlmZddlm Z m Z e e Z dZeGd d d Z dd ejdBd ejeBeBd ejdededBf ddZ dd ejdBd ejeBeBd ejdededBf ddZddejdejdedBfddZddejdejdedBfddZ  d d ejeBeBdejdejdededBdejdBf ddZdS)!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. ) dataclass)UnionN)logging)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@sDeZdZUdZeed<eed<d'dededBfddZ d(ded ed ed ej d e ej d fdej dBf ddZ d'dej d ed ej d edBdej f ddZe  d)dejd ej d ej dededBf ddZed'dej d ej dedBfddZedejdefdd Ze  !d*d"ej dBd#ej dededBd$edef d%d&ZdS)+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_windowNcCsFttt||_||_|jdur|jdkr!td|jddSdS)NrzaMake sure that when passing `sliding_window` that its value is a strictly positive integer, not ``)logger warning_onceDEPRECATION_MESSAGE FutureWarningr r ValueError)selfr r rc/home/ubuntu/transcripts/venv/lib/python3.10/site-packages/transformers/modeling_attn_mask_utils.py__init__Is  zAttentionMaskConverter.__init__cpu batch_size query_lengthkey_value_lengthdtypedevicestrreturnc Cs\|js td|jd||f}||}d}|ddks!|jdur,|j|||||jd}|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_shapercausal_4d_maskrrr to_causal_4dTs z#AttentionMaskConverter.to_causal_4dattention_mask_2dc Cs|jd|f}d}|ddks|jdur0|jr0|durtd||}|j|||j||jd}n |jdur9td|j|||dd|j}|durW| | t |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.rz?Sliding window is currently only implemented for causal masking)tgt_len)shaper r rr!rNotImplementedError _expand_maskto masked_fillbooltorchfinfomin) rr%rrrr"r#rexpanded_attn_maskexpanded_4d_maskrrrto_4dus0  zAttentionMaskConverter.to_4drinput_ids_shaperc Csttt|\}}tj||ft|j|d}tj| d|d}| ||d | ddkd| |}|dkrLtj tj||||d|gdd}|durt||d} tjtj|tjd| d } trj|}| | t|j|ddddddf|d|||S) zJ Make causal mask used for bi-directional self-attention. )rrrrrr)dimNr)diagonal)r r rrr-fullr.r/arangesize masked_fill_viewr*catzerostril ones_liker,rcloneexpand) r3rrrr bszr&mask mask_condr7 context_maskrrrr!s "   (z(AttentionMaskConverter._make_causal_maskrDr&cCsttt|\}}|dur|n|}|ddddddf|d|||}tjd|d|}| |tj t |j S)zg Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`. Nr?r6) r r rrr:rBr*r-tensorr+r,r.r/)rDrr&rCsrc_len expanded_mask inverted_maskrrrr)s  *z#AttentionMaskConverter._expand_maskrJ min_dtypecCs<ttt|jtjkrtd|tj ||kdddS)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)r5keepdim) r r rrrr-r,rmulall)rJrLrrr_unmask_unattendeds ) z)AttentionMaskConverter._unmask_unattendedFattention_mask inputs_embeds is_trainingc Csttt|jd|jd}}||}t|}d} |dur7|s#|s5|dks+||kr5|dus3||kr5d} | S|dus?||kr[t|jdkrHdS|s[t|dkr[|dksY||kr[d} | 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). rrFNT) r r rrr'rlenr-rO) rQrRrr rS_rr is_tracing_ignore_causal_maskrrr_ignore_causal_mask_sdpa s*  z/AttentionMaskConverter._ignore_causal_mask_sdpaN)rrN)NF)__name__ __module__ __qualname____doc__r,__annotations__intrr-rrrTensorr$r2 staticmethodSizer!r) FloatTensorfloatrPrYrrrrr&s   & / #" 2rrQr"rRrr c Cstd|d}|d|}|dur%t|jdkr%|j||d||jd}|S|durdt|jdkrd|dd |d |f}t|j|krOtd t|jd |d d |}||t j t |jj }|S|j |d|d||j|jd}|S)a Creates a 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: attention_mask (`torch.Tensor` or `None`): A 2D attention mask of shape `(batch_size, key_value_length)` input_shape (`tuple(int)` or `list(int)` or `torch.Size`): The input shape should be a tuple that defines `(batch_size, query_length)`. inputs_embeds (`torch.Tensor`): The embedded inputs as a torch Tensor. past_key_values_length (`int`): The length of the key value cache. sliding_window (`int`, *optional*): If the model uses windowed attention, a sliding window should be passed. Tr r rN)rrrTrrz#Incorrect 4D attention_mask shape: z ; expected: .rGr4)rrUr'r2rtuplerr+r*r-r,r.r/r$r) rQr"rRrr attn_mask_converterrexpected_shaperKrrr!_prepare_4d_causal_attention_maskGs,  rmc Cstd|d}|d|}t|}tj||||d}|rd} | S|dur5|j|d|d||j|jd} | S|dkr>|} n |j||d|j|d } |s^| jjd vr^tj | t |jj d } | S) a Prepares the correct `attn_mask` argument to be used by `torch.nn.functional.scaled_dot_product_attention`. In case no token is masked in the `attention_mask` argument, we simply set it to `None` for the cases `query_length == 1` and `key_value_length == query_length`, and rely instead on SDPA `is_causal` argument to use causal/non-causal masks, allowing to dispatch to the flash attention kernel (that can otherwise not be used if a custom `attn_mask` is passed). Trgr)rQrRrr Nrr4rT)rr)cudaxpu)rL) rrrYr$rrr5r2typerPr-r.r/) rQr"rRrr rkrrWrXr1rrr*_prepare_4d_causal_attention_mask_for_sdpa|s<    rqrDrr&cCstj|||dS) 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. rDrr&)rr)rsrrr_prepare_4d_attention_masks rtcCsPttt|j\}}|dur|n|}t|s t|dkr dStj |||dS)rrNrrs) r r rrr'rr-rOrr))rDrr&rVrrrr#_prepare_4d_attention_mask_for_sdpas rurrcCs8td|d}||d}|j|d|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. Trgrrr4)rr$)r"rrrr rkrrQrrr _create_4d_causal_attention_masks  rvrZr[)r_ dataclassesrtypingrr-utilsrutils.import_utilsrr get_loggerr\r rrrbrdrjlistrarmrqrrtrurrvrrrrsl    '  :  :