o ٷihH@sFddlZddlZddlmZddlmZeeZGdddeZdS)N) numpy_helper)BertOnnxModelTFcsteZdZfddZddZddZddZd d Zd d Zd dZ ddZ ddZ ddZ ddZ ddZZS)BertOnnxModelKerascst|||dSN)super__init__)selfmodel num_heads hidden_size __class__b/home/ubuntu/.local/lib/python3.10/site-packages/onnxruntime/transformers/onnx_model_bert_keras.pyrszBertOnnxModelKeras.__init__cCs^||gdgd}|dur|S||gdgd}|dur"|S||gdgd}|S)N)MulSubReshapeCast)Nrr)rrrSlice Unsqueeze)rrrrr)rrrrr)rNrrr)match_parent_path)radd_or_sub_before_softmax mask_nodesrrrmatch_mask_paths(z"BertOnnxModelKeras.match_mask_pathc Csg}|||fD]8}|jd}||} | |krq| jdkr,| jd|jdkr,|| qtd|d|jddgfSd|fS)NrrzCheck attention input failed:z, FT)inputop_typeoutputappendloggerdebug) rmatmul_qmatmul_kmatmul_vparentoutput_name_to_node reshape_nodesx root_input root_noderrrcheck_attention_input+s   z(BertOnnxModelKeras.check_attention_inputc,Cs||}g}d}|d}|D]}||d}|dus%|jdvrZ|jdkrJ||d}|dus9|jdvrItd|durC|jndqntd|durT|jndq ||gdgd}|durptd q|\}} } } } } td || gd gd }|durtd q|\}}}}}|| gdgd}|dur|\}}}||gdgd}|dur|\}}}}}}nA|| gdgd}|dur|| gdgd}|durtdq|\}}} }||gd gd}|dur|\}}}}}|durtdq||gd gd }!|!dur'tdq|!\}"}#}$}%}&||d}'|'dur@tdq| |'ddsOtdq| ||&|||\}(})|(r|j |'dj d}*td|jj|*||&|||$||j|j|jd| jdd }+|+durq||+|d7}|| | | g||||||!||||'||)|| ||| jd| jdqtdq|||td|dS) NrSkipLayerNormalization)r+EmbedLayerNormalizationAddrzFirst input for skiplayernorm: )r-rMatMulr Transposer.)NrrrrrzFailed to match qkv nodeszMatched qkv nodes)r/rr-rr.)rrrrrzFailed to match v path)Softmaxrr.rrr)rr/rr-rr.)rNrrrr)r0r-rr.)rrrN)r0r-Divr.zFailed to match qk path)rrrrrzFailed to match q pathzFailed to match k pathzFailed to match mask pathz;Sub node expected to have an input with constant value 1.0.zCreate an Attention node.) mask_indexq_matmulk_matmulv_matmulq_addk_addv_addr r first_inputrzRoot node not matched.zFused Attention count:)input_name_to_nodesr%get_nodes_by_op_type get_parentrrr rrhas_constant_inputr*attention_mask process_maskrattention_fusioncreate_attention_noder r radd_nodeextendrreplace_node_input remove_nodes update_graphinfo),rr%nodes_to_removeattention_countskip_layer_norm_nodesnormalize_noder$ qkv_nodesaddextra_reshape_0matmul reshape_qkv transpose_qkv matmul_qkvv_nodes transpose_v reshape_vadd_vextra_reshape_1r#qk_nodes softmax_qksub_qk matmul_qkq_nodesmul_q transpose_q reshape_qadd_qextra_reshape_2r!add_qkmul_qkk_nodes transpose_k reshape_kadd_kextra_reshape_3r"r is_same_rootr&r4attention_noderrrfuse_attention;s                                z!BertOnnxModelKeras.fuse_attentioncCs|||dSr)process_embedding fuse_mask skip_reshape)rrrr preprocesss zBertOnnxModelKeras.preprocesscCs~||d}|d}|D]}||d}|dur.|jdkr.|jd|jd<|d7}q|dkr=td|dSdS)NrrrzSkip consequent Reshape count: )r<r%r=r>rrrrI)rcountr& reshape_noder$rrrrps  zBertOnnxModelKeras.skip_reshapecCs|jdksJtd|jdd||gdgd|}|dur*tddS|\}}}||jd}|durBtd dSt|}t |j d kr_t d |j d |j |j } nt d |j d |j dS||jd} | durt| }t |j dkr|j ddkrt ||j d|j d fd} || t d| j d |j ddd} njt d| j d |j dS||ddgddg|} | durtddS| \}}||jd} | durtddSt| }t |j d krt d | j d |j | j } nt d| j d |j dS||d|}|dus3|jdkr:tddS||jd}|durNtddSt|}t |j d krlt d|j d |j |j }nt d|j d |j dSt d||| || dS)NLayerNormalizationz-start fusing embedding from node with output=rz...)r-r-Gatherr1zfailed to match word_embed_pathFzfailed to get word initializerzFound word embedding. name:z, shape:z$Failed to find word embedding. name:rposition_embeddingzFound position embedding. name:z(Failed to find position embedding. name:rurzfailed to match pos_embed_pathzfailed to get pos initializerzfailed to get gatherz!failed to get segment initializerzFound segment embedding. name:z'Failed to find segment embedding. name:zCreate Embedding nodeT)rrr rrget_initializerrrto_arraylenshaperIname from_arrayreshapeadd_initializerr>create_embedding_subgraph)rnoder%word_embed_path skip_noderD gather_nodeword_initializertempword_embeddingpos_initializertensorrxpos_embed_path pos_gather pos_slicegathersegment_initializersegment_embeddingrrrfuse_embeddingsv     " "        z!BertOnnxModelKeras.fuse_embeddingcCsFtd|}|D]}|jdkr |||rdSdSq dS)zM Automatically detect word, segment and position embeddings. z#start processing embedding layer...rtN)rrIr%nodesrr)rr%rrrrrn7s    z$BertOnnxModelKeras.process_embeddingc Csrg}|D]}|jdkr||dr||gdgd}|dur#q|\}}}}|j}|jd|krCtd|jdd|qtj j d|gd gd d gd } tj j dd gd gddgd } tj j dd gdgd} | j tj dd g|||jd d| |||g|| || || q||t|dkr|tt|dkrddSddS)Nri)rrrr)rrrrrz Cast input z is not mask input rmask_fuse_unsqueeze1_outputMask_UnSqueeze_1r)inputsoutputsr}axesmask_fuse_unsqueeze2_outputMask_UnSqueeze_2rvrmask_fuse_cast_output)rrtoz Fused maskzFailed to fuse mask)rrr?rr@get_first_maskrprintonnxhelper make_node attributerEmake_attributerFrDrGr{ prune_graphrrI) rrJr mask_pathsub_node cast_node slice_nodeunsqueeze_nodemask_input_nameunsqueeze_added_1unsqueeze_added_2 cast_node_2rrrroCsR         $zBertOnnxModelKeras.fuse_maskc Cs|d}d}|D]H}||gdgd}|durq |\ }}}}} } } } } | jd| jd<|| | jd|jd<|||jd|jd<|||d7}q |S)Nr+r) r-rr.rGelur-rr.r+) rrrrrrrrrrwr=rrr remove_node)rskiplayernorm_nodesreshape_removedskiplayernorm_nodepathadd_1 reshape_1matmul_1 reshape_2geluadd_2 reshape_3matmul_2 skiplayernormrrrremove_extra_reshapews:       z'BertOnnxModelKeras.remove_extra_reshapec Cs|d}d}|D]V}||gdgd}|durq |\ }}}}} } } } } }|jd| jd<|| | jd| jd<|| | jd|jd<|||jd|jd<|||d7}q |S)Nr+r) r-rr.rrr-rr.rr+) Nrrrrrrrrrr)rrrrrrrrrrrrr reshape_4rrrrremove_extra_reshape_2s@        z)BertOnnxModelKeras.remove_extra_reshape_2cCs.||}td|d|dS)NzRemove z Reshape nodes.)rrrrIr)rrrrr postprocesss zBertOnnxModelKeras.postprocess)__name__ __module__ __qualname__rrr*rmrqrprrnrorrr __classcell__rrr rrs L 4+4r) loggingrronnx_model_bert_tfr getLoggerrrrrrrrs