ibddlmZddlmZddlZddlmZddlmZddlm Z m Z ddl m Z dd l mZdd lmZdd lmZdd lmZdd lmZmZddlmZmZddlmZmZddlmZddl m!Z!m"Z"m#Z#ddl$m%Z%m&Z&ddl'm(Z(GddejRZ*GddejRZ+GddejRZ,dejZde.dejZfdZ/ d5dejRd ejZd!ejZd"ejZd#ejZdzd$e0d%e0d&ee!fd'Z1d(Z2d6d)Z3ee3Gd*d+ejRZ4Gd,d-eZ5e"Gd.d/eZ6e"Gd0d1e6Z7e"Gd2d3e6e Z8gd4Z9y)7)Callable)OptionalN)nn)ACT2FN)Cache DynamicCache)GenerationMixin)use_kernelized_func)create_causal_mask)FlashAttentionKwargs)GradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPast)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)TransformersKwargsauto_docstringcan_return_tuple)check_model_inputsmaybe_autocast) CohereConfigc&eZdZdfd ZdZxZS)CohereLayerNormct|tjt j ||_||_y)zcThe hidden size can be a tuple or an int. The tuple is used for QKNorm to normalize across head_dimN)super__init__r Parametertorchonesweightvariance_epsilon)self hidden_sizeepsbias __class__s t/home/obispo/Crisostomo_bridge/mision_env/lib/python3.12/site-packages/transformers/models/cohere/modeling_cohere.pyr!zCohereLayerNorm.__init__4s/ ll5::k#:; #c|j}|jtj}|j dd}||z j dj dd}||z tj ||jzz}|jjtj|z}|j|S)NT)keepdim) dtypetor#float32meanpowrsqrtr&r%)r' hidden_states input_dtyper5variances r,forwardzCohereLayerNorm.forward:s#)) %((7 !!"d!3!D(--a055b$5G&-XH]H]=]1^^  u}}5 E  ,,r-)Ngh㈵>F__name__ __module__ __qualname__r!r; __classcell__r+s@r,rr3s $ -r-rc eZdZUejed<d deffd Ze d dedzde dde dzde d e ffd Z ejed ZxZS)CohereRotaryEmbeddinginv_freqNconfigct||j|_|j|_||_|j j d|_|j}|jdk7rt|j}||j |\}|_ |jd|d|jd|jdy)N rope_typedefaultrDF) persistentoriginal_inv_freq) r r!max_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenrErope_parametersrGcompute_default_rope_parametersrattention_scalingregister_bufferclone)r'rEdevice rope_init_fnrDr+s r,r!zCohereRotaryEmbedding.__init__Gs "("@"@$*$B$B! 44[A!%!E!E >>Y &.t~~>L+7 V+L($( ZeD 0(..2BuUr-rSz torch.deviceseq_lenreturnz torch.Tensorc|jd}t|ddxs|j|jz}d}d|t j d|dtj j|tj|z zz }||fS) a Computes the inverse frequencies according to the original RoPE implementation Args: config ([`~transformers.PreTrainedConfig`]): The model configuration. device (`torch.device`): The device to use for initialization of the inverse frequencies. seq_len (`int`, *optional*): The current sequence length. Unused for this type of RoPE. Returns: Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE). rope_thetahead_dimNg?rr1r2)rSr2) rNgetattrr(num_attention_headsr#arangeint64r3float)rErSrUbasedimattention_factorrDs r,rOz5CohereRotaryEmbedding.compute_default_rope_parametersWs&%%l3fj$/c63E3EIcIc3c U\\!S!5;;?BB&X]XcXcBdgjj k )))r-c|jddddfjj|jddd}|dddddfj}t |j j tr/|j j dk7r|j j nd}t|d5|j|jzjdd}tj|dd }|j|jz}|j|jz} dddj|j  j|j fS#1swYAxYw) Nrr/rmpscpuF) device_typeenabledr1rarZ)rDr_expandshape isinstancerStypestrr transposer#repeat_interleavecosrPsinr3r2) r'x position_idsinv_freq_expandedposition_ids_expandedrffreqsembrprqs r,r;zCohereRotaryEmbedding.forwardus@!MM$4-8>>@GG HZHZ[\H]_acde ,QaZ 8 > > @'1!((--'E!((--[`J`ahhmmfk  U C 5&,,.1F1L1L1NNYYZ[]^_E))%;C'')d444C'')d444C  5 vvAGGv$cff177f&;;;  5 5s =BFF N)NNN)r=r>r?r#Tensor__annotations__rr! staticmethodrinttupler_rOno_gradrr;r@rAs@r,rCrCDsllV|V &*+/"*t#*(*t* ~u$ % **:U]]_ <