o ۷i@sdZddlZddlmZmZddlZddlmZddlmZddl m Z ddl m Z m Z mZdd lmZdd lmZmZdd lmZdd lmZmZdd lmZddlmZmZddlmZddl m!Z!e"e#Z$Gdddej%Z&Gdddej%Z'Gdddej(Z)Gdddej(Z*GdddeZ+eGdddeZ,Gddde,Z-ed d!Gd"d#d#e,Z.ed$d!Gd%d&d&e,eZ/d&dgZ0dS)'z/PyTorch TrOCR decoder model (based on RoBERTa).N)OptionalUnion)nn)CrossEntropyLoss)ACT2FN)Cache DynamicCacheEncoderDecoderCache)GenerationMixin)_prepare_4d_attention_mask!_prepare_4d_causal_attention_mask)GradientCheckpointingLayer))BaseModelOutputWithPastAndCrossAttentions!CausalLMOutputWithCrossAttentions)PreTrainedModel)auto_docstringlogging)deprecate_kwarg) TrOCRConfigcsPeZdZdZdedeffdd Z d dejd ed eejffd d Z Z S)TrOCRLearnedPositionalEmbeddingzN This module learns positional embeddings up to a fixed maximum size. num_embeddings embedding_dimcsd|_t||j|dS)N)offsetsuper__init__)selfrr __class__^/home/ubuntu/vllm_env/lib/python3.10/site-packages/transformers/models/trocr/modeling_trocr.pyr0sz(TrOCRLearnedPositionalEmbedding.__init__rN input_idspast_key_values_length position_idscs\|dur |jdd\}}tj|||tj|jjd|d}n|d}t ||j S)z3`input_ids' shape is expected to be [bsz x seqlen].Nr)dtypedevicer) shapetorcharangelongweightr'expand unsqueezerforwardr)rr#r$r%bszseq_lenrr!r"r06s z'TrOCRLearnedPositionalEmbedding.forward)rN) __name__ __module__ __qualname____doc__intrr*Tensorrr0 __classcell__r!r!rr"r+src sLeZdZdZd dedededeeffdd Zd ej ffd d Z Z S) TrOCRScaledWordEmbeddingz\ This module overrides nn.Embeddings' forward by multiplying with embeddings scale. ?rr padding_idx embed_scalecst|||||_dSN)rrr=)rrrr<r=rr!r"rLs z!TrOCRScaledWordEmbedding.__init__r#cst||jSr>)rr0r=)rr#rr!r"r0Psz TrOCRScaledWordEmbedding.forward)r;) r3r4r5r6r7rfloatrr*r8r0r9r!r!rr"r:Gs$r:c seZdZdZddededeeffdd Zeddededeefd d Ze dd e j d efddZ dd e j ded eefddZ ZS)"TrOCRSinusoidalPositionalEmbeddingzDThis module produces sinusoidal positional embeddings of any length.N num_positionsrr<csBtd|_||_||_|||||_|dt ddS)Nr _float_tensorr) rrrrr< get_embeddingweightsregister_bufferr* FloatTensor)rrArr<rr!r"rWs z+TrOCRSinusoidalPositionalEmbedding.__init__rcCs|d}td|d}ttj|tjd| }tj|tjdd|d}tjt |t |gdd |d}|ddkrUtj|t |dgdd}|durad||ddf<| tS) z Build sinusoidal embeddings. This matches the implementation in tensor2tensor, but differs slightly from the description in Section 3.5 of "Attention Is All You Need". ri'r)r&rdimr(N)mathlogr*expr+int64r?r/catsincosviewzerostoget_default_dtype)rrr<half_dimembr!r!r"rC_s $& z0TrOCRSinusoidalPositionalEmbedding.get_embeddingrr#r$cCs|\}}|||j||j}|jd|}|jdus&||jdkr0|||j|j|_|j|j|_|j d| d ||d }|S)Nrrr() size"create_position_ids_from_input_idsr<rRr'rDrCrrB index_selectrPdetach)rr#r$r1r2r%max_posxr!r!r"r0rs "z*TrOCRSinusoidalPositionalEmbedding.forwardcCs6||}tj|dd|||}||S)z Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols are ignored. This is modified from fairseq's `utils.make_positions`. rrG)ner7r*cumsumtype_asr,)rr#r<r$maskincremental_indicesr!r!r"rWs zETrOCRSinusoidalPositionalEmbedding.create_position_ids_from_input_idsr>)r)r3r4r5r6r7rr staticmethodrCr*no_gradr8r0rWr9r!r!rr"r@Ts r@cseZdZdZ       ddededeed eed eed eed eed eedeeffdd Ze dddd      dde j dee j dee dee j dee j deedee j de e j ee j ee e j ffddZZS) TrOCRAttentionz>Multi-headed attention from 'Attention Is All You Need' paper.NFT embed_dim num_headskdimvdimdropout is_decoderbiasis_cross_attention layer_idxc st||_|dur|n||_|dur|n||_||_||_|||_|j||jks9td|jd|d|jd|_ ||_ | |_ t j |j||d|_t j |j||d|_t j |||d|_t j |||d|_dS)Nz;embed_dim must be divisible by num_heads (got `embed_dim`: z and `num_heads`: z).g࿩rk)rrrergrhrfrihead_dim ValueErrorscalingrjrmrLineark_projv_projq_projout_proj) rconfigrerfrgrhrirjrkrlrmrr!r"rs(    zTrOCRAttention.__init__past_key_valuepast_key_values4.58new_nameversion hidden_stateskey_value_statesattention_masklayer_head_maskoutput_attentionscache_positionreturncCs|du}|\} } } |||j} d} |dur1t|tr/|j|j} |r+|j}n|j }n|}|r5|n|}|rN|durN| rN|j |jj }|j |jj }nJ| |}||}|| d|j|jdd}|| d|j|jdd}|dur|s||nd}||||jd|i\}}|rt|trd|j|j<| |jd|jf}| | | |j|jdd} | j|} |j|}|j|}|d}t| |dd}|| |j| |fkrtd| |j| |fd ||dur || d| |fkr td | d| |fd ||| |j| ||}|| |j| |}tjj|dd }|dur_||jfkrDtd |jfd ||dddd|| |j| |}|| |j| |}|rv|| |j| |}|| |j| |}nd}tjj||j|jd }t||}|| |j| |jfkrtd| |j| |jfd ||| |j| |j}|dd}|| | | }||}||fS)z#Input shape: Batch x Time x ChannelNFr(rrrTz$Attention weights should be of size z , but is z!Attention mask should be of size rGz/Head mask for a single layer should be of size ptrainingz `attn_output` should be of size )rVrurq isinstancer is_updatedgetrmcross_attention_cacheself_attention_cachelayerskeysvaluesrsrtrPrfro transposeupdatereshaper*bmmrpr functionalsoftmaxrirrv)rr~rryrrrrrlr1tgt_lenre query_statesrcurr_past_key_valuecurrent_states key_states value_states proj_shapesrc_len attn_weightsattn_weights_reshaped attn_probs attn_outputr!r!r"r0s             "   zTrOCRAttention.forward)NNrdFTFN)NNNNFN)r3r4r5r6r7rr?boolrrr*r8rtupler0r9r!r!rr"rcsh    # rccseZdZddeffdd Zedddd       dd ejd eejd eejdeejdeejdeejdee dee dee deejfddZ Z S)TrOCRDecoderLayerNrwc st|j|_t||j|j|jd|d|_|j|_t |j |_ |j |_ t |j|_|jrIt||j|j|j|j|jdd|d |_t |j|_t |j|j|_t |j|j|_t |j|_dS)NT)rerfrirjrm)rerfrgrhrirjrlrm)rr hidden_sizerercdecoder_attention_headsattention_dropout self_attnriractivation_function activation_fnactivation_dropoutr LayerNormself_attn_layer_normrjcross_attention_hidden_size encoder_attnencoder_attn_layer_normrrdecoder_ffn_dimfc1fc2final_layer_norm)rrwrmrr!r"r.s<   zTrOCRDecoderLayer.__init__rxryrzr{FTr~rencoder_hidden_statesencoder_attention_maskrcross_attn_layer_head_maskr use_cacherc  Cs|} |j|||||| d\}} tjj||j|jd}| |}||}d} |durM|} |j||||||| d\}} tjj||j|jd}| |}||}|} || |}tjj||j |jd}| |}tjj||j|jd}| |}| |}|f}|r|| | f7}|S)a Args: hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)` attention_mask (`torch.FloatTensor`): attention mask of size `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values. encoder_hidden_states (`torch.FloatTensor`): cross attention input to the layer of shape `(batch, seq_len, embed_dim)` encoder_attention_mask (`torch.FloatTensor`): encoder attention mask of size `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values. layer_head_mask (`torch.FloatTensor`): mask for attention heads in a given layer of size `(encoder_attention_heads,)`. cross_attn_layer_head_mask (`torch.FloatTensor`): mask for cross-attention heads in a given layer of size *(decoder_attention_heads,)*. past_key_values (`Cache`): cached past key and value projection states output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. )r~ryrrrrrN)r~rrrryrr) rrrrirrrrrrrrr)rr~rrrrrryrrrresidualself_attn_weightscross_attn_weightsoutputsr!r!r"r0RsL          zTrOCRDecoderLayer.forwardr>) NNNNNNFTN) r3r4r5rrrr*r8rrrr0r9r!r!rr"r-sB $   rc@s,eZdZUeed<dZdZdgZddZdS)TrOCRPreTrainedModelrwmodelTrcCs|jj}t|tjtjfr%|jjjd|d|j dur#|j j dSdSt|tj rD|jjjd|d|j durF|jj|j  dSdSdS)Nrd)meanstd) rwinit_stdrrrrConv1dr-datanormal_rkzero_ Embeddingr<)rmodulerr!r!r" _init_weightss   z"TrOCRPreTrainedModel._init_weightsN) r3r4r5r__annotations__base_model_prefixsupports_gradient_checkpointing_no_split_modulesrr!r!r!r"rs  rcsJeZdZdZdeffdd Z             dddZZS) TrOCRDecoderz Transformer decoder consisting of *config.decoder_layers* layers. Each layer is a [`TrOCRDecoderLayer`] Args: config: TrOCRConfig rwcstj|_j|_j|_jrt j nd}t j j |j|d|_ jr5tjj |_ntj|jdj |j|_jrOtj |_nd|_tfddtjD|_d|_|dS)Nr;)r=rcsg|]}t|dqS))rm)r).0irwr!r" sz)TrOCRDecoder.__init__..F)rrridecoder_layerdrop layerdrop pad_token_idr<scale_embeddingrIsqrtrr: vocab_size embed_tokensuse_learned_position_embeddingsrmax_position_embeddingsembed_positionsr@layernorm_embeddingrr ModuleListrangedecoder_layersrgradient_checkpointing post_init)rrwr=rrr"rs*   zTrOCRDecoder.__init__Nc Cs| dur| n|jj} | dur| n|jj} | dur| n|jj} | dur$| n|jj} |dur4|dur4td|durD|}|d|jd}n|dur\|dd}|dddddf}ntd|j ro|j ro| rot dd} | r|dur|durt t|jdt|jdnt|jd}| rt|trt dt |}|dur|nd }|dur||}|jjr|j||d }n|j||d }||}|jdur||}tjj||j|j d }|j}t||||}|dur|durt||j|dd }| rd nd}| rd nd}| r|durd nd}t||gddgD]+\}}|durD|d t|jkrDtd|dt|jd|d dqt |jD]X\}}| rW||f7}|j rht!"g}||j#krhqK||||||durv||nd|dur||nd|| | | d }|d }| r||df7}|dur||df7}qK| r||f7}| stdd|||||fDSt$|||||dS)a; Args: input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`): Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide it. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. [What are attention masks?](../glossary#attention-mask) encoder_hidden_states (`torch.FloatTensor` of shape `(batch_size, encoder_sequence_length, hidden_size)`, *optional*): Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of the decoder. encoder_attention_mask (`torch.LongTensor` of shape `(batch_size, encoder_sequence_length)`, *optional*): Mask to avoid performing cross-attention on padding tokens indices of encoder input_ids. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. [What are attention masks?](../glossary#attention-mask) head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*): Mask to nullify selected heads of the attention modules. Mask values selected in `[0, 1]`: - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*): Mask to nullify selected heads of the attention modules in encoder to avoid performing cross-attention on hidden heads. Mask values selected in `[0, 1]`: - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`): It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache). Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see `past_key_values` input) to speed up sequential decoding. If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `decoder_input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert `input_ids` indices into associated vectors than the model's internal embedding lookup matrix. output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. output_hidden_states (`bool`, *optional*): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more detail. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. NzTYou cannot specify both decoder_input_ids and decoder_inputs_embeds at the same timer(zEYou have to specify either decoder_input_ids or decoder_inputs_embedsz^`use_cache = True` is incompatible with gradient checkpointing. Setting `use_cache = False`...FrzPassing a tuple of `past_key_values` is deprecated and will be removed in Transformers v4.58.0. You should pass an instance of `EncoderDecoderCache` instead, e.g. `past_key_values=EncoderDecoderCache.from_legacy_cache(past_key_values)`.r)r$r)rr! head_maskcross_attn_head_maskzThe `z` should be specified for z layers, but it is for .)rrrryrrrrrcss|] }|dur|VqdSr>r!)rvr!r!r" sz'TrOCRDecoder.forward..)last_hidden_stateryr~ attentionscross_attentions)%rwroutput_hidden_statesruse_return_dictrprPr)rVrrlogger warning_oncer r rrfrom_legacy_cacheget_seq_lengthrrrrrrrir r r&ziplenr enumerater*randrr)rr#rrrrrry inputs_embedsrrr return_dictrinput input_shaper$ embed_posr~all_hidden_statesall_self_attnsall_cross_attentions attn_mask mask_nameidx decoder_layerdropout_probability layer_outputsr!r!r"r0sO                  zTrOCRDecoder.forward) NNNNNNNNNNNNN)r3r4r5r6rrr0r9r!r!rr"rs"!ra The TrOCR Model with a language modeling head. Can be used for summarization. This wrapper class is a helper class to correctly load pretrained checkpoints when the causal language model is used in combination with the [`EncoderDecoderModel`] framework. ) custom_introcs$eZdZfddZddZZS)TrOCRDecoderWrappercst|t||_dSr>)rrrdecoderrrwrr!r"rs zTrOCRDecoderWrapper.__init__cOs|j|i|Sr>)r)rargskwargsr!r!r"r0szTrOCRDecoderWrapper.forward)r3r4r5rr0r9r!r!rr"rs rzy The TrOCR Decoder with a language modeling head. Can be used as the decoder part of [`EncoderDecoderModel`] and c"s eZdZdgZfddZddZddZdd Zd d Zd d Z ddZ e              d"de e jde e jde e jde e jde e jde e jde ede e jde e jde ede ede ede ede e jdeeeffd d!ZZS)#TrOCRForCausalLMzoutput_projection.weightcsDd|_d|_t|t||_tj|j|j dd|_ | dS)NTFrn) rjis_encoder_decoderrrrrrrrrroutput_projectionrrrr!r"rs    zTrOCRForCausalLM.__init__cCs |jjjSr>rrrrr!r!r"get_input_embeddings z%TrOCRForCausalLM.get_input_embeddingscCs||jj_dSr>r )rvaluer!r!r"set_input_embeddingssz%TrOCRForCausalLM.set_input_embeddingscCs|jSr>r r r!r!r"get_output_embeddingssz&TrOCRForCausalLM.get_output_embeddingscCs ||_dSr>r)rnew_embeddingsr!r!r"set_output_embeddingsrz&TrOCRForCausalLM.set_output_embeddingscCs ||j_dSr>rr)rrr!r!r" set_decoders zTrOCRForCausalLM.set_decodercCs|jjSr>rr r!r!r" get_decoderszTrOCRForCausalLM.get_decoderNr#rrrrrryrlabelsrrrrrrcCs| dur| n|jj} | dur| n|jj} | dur| n|jj} |jj||||||||| | | | |d }||d}d}| durOt}||d|jj | d}| se|f|dd}|durc|f|S|St |||j |j |j |jdS)a cross_attn_head_mask (`torch.Tensor` of shape `(decoder_layers, decoder_attention_heads)`, *optional*): Mask to nullify selected heads of the cross-attention modules. Mask values selected in `[0, 1]`: - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should either be in `[0, ..., config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. Example: ```python >>> from transformers import ( ... TrOCRConfig, ... TrOCRProcessor, ... TrOCRForCausalLM, ... ViTConfig, ... ViTModel, ... VisionEncoderDecoderModel, ... ) >>> import requests >>> from PIL import Image >>> # TrOCR is a decoder model and should be used within a VisionEncoderDecoderModel >>> # init vision2text model with random weights >>> encoder = ViTModel(ViTConfig()) >>> decoder = TrOCRForCausalLM(TrOCRConfig()) >>> model = VisionEncoderDecoderModel(encoder=encoder, decoder=decoder) >>> # If you want to start from the pretrained model, load the checkpoint with `VisionEncoderDecoderModel` >>> processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten") >>> model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten") >>> # load image from the IAM dataset >>> url = "https://fki.tic.heia-fr.ch/static/img/a01-122-02.jpg" >>> image = Image.open(requests.get(url, stream=True).raw).convert("RGB") >>> pixel_values = processor(image, return_tensors="pt").pixel_values >>> text = "industry, ' Mr. Brown commented icily. ' Let us have a" >>> # training >>> model.config.decoder_start_token_id = processor.tokenizer.eos_token_id >>> model.config.pad_token_id = processor.tokenizer.pad_token_id >>> model.config.vocab_size = model.config.decoder.vocab_size >>> labels = processor.tokenizer(text, return_tensors="pt").input_ids >>> outputs = model(pixel_values, labels=labels) >>> loss = outputs.loss >>> round(loss.item(), 2) 5.30 >>> # inference >>> generated_ids = model.generate(pixel_values) >>> generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0] >>> generated_text 'industry, " Mr. Brown commented icily. " Let us have a' ```N) r#rrrrrryrrrrrrrr(r)losslogitsryr~rr)rwrrrrrr rrPrrryr~rr)rr#rrrrrryrrrrrrrrrrloss_fctoutputr!r!r"r0sFMzTrOCRForCausalLM.forward)NNNNNNNNNNNNNN)r3r4r5_tied_weights_keysrr rrrrrrrr* LongTensorr8rFrrrrrr0r9r!r!rr"rsn        r)1r6rItypingrrr*rtorch.nnr activationsr cache_utilsrr r generationr modeling_attn_mask_utilsr r modeling_layersrmodeling_outputsrrmodeling_utilsrutilsrrutils.deprecationrconfiguration_trocrr get_loggerr3rrrr:Moduler@rcrrrrr__all__r!r!r!r"sJ          >zz