tjZddlmZddlmZddlZddlmZddlmZddlm Z m Z ddl m Z dd l mZdd lmZdd lmZdd lmZmZdd lmZmZddlmZmZddlmZddlmZm Z m!Z!ddl"m#Z#m$Z$ddl%m&Z&ddl'm(Z(Gddej)Z*Gddej)Z+Gddej)Z,dej-de.dej-fdZ/ d6d ej)d!ej-d"ej-d#ej-d$ej-dzd%e0d&e0d'eefd(Z1d)Z2d7d*Z3Gd+d,ej)Z4Gd-d.eZ5e Gd/d0eZ6e Gd1d2e6Z7e Gd3d4e6e Z8gd5Z9dS)8)Callable)OptionalN)nn)ACT2FN)Cache DynamicCache)GenerationMixin)create_causal_mask)FlashAttentionKwargs)GradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPast)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)TransformersKwargsauto_docstringcan_return_tuple)maybe_autocastmerge_with_config_defaults)capture_outputs) CohereConfigc&eZdZdfd ZdZxZS)CohereLayerNormNh㈵>Fcttjt j||_||_dS)zcThe hidden size can be a tuple or an int. The tuple is used for QKNorm to normalize across head_dimN)super__init__r Parametertorchonesweightvariance_epsilon)self hidden_sizeepsbias __class__s h/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/models/cohere/modeling_cohere.pyr"zCohereLayerNorm.__init__4sB l5:k#:#:;; #c|j}|tj}|dd}||z ddd}||z tj||jzz}|jtj|z}||S)NT)keepdim) dtypetor$float32meanpowrsqrtr'r&)r( hidden_states input_dtyper6variances r-forwardzCohereLayerNorm.forward:s#) %((77 !!"d!33!D(--a0055b$5GG&-XH]=]1^1^^  u}55 E  ,,,r.)NrF__name__ __module__ __qualname__r"r< __classcell__r,s@r-rr3sL$$$$$$ -------r.rc eZdZUejed<d deffd Ze d dedzde dde dzde d e ffd Z ejed ZxZS)CohereRotaryEmbeddinginv_freqNconfigct|j|_|j|_||_|jjd|_|j}|jdkrt|j}||j|\}|_ | d|d| d| ddS)N rope_typedefaultrEF) persistentoriginal_inv_freq) r!r"max_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenrFrope_parametersrHcompute_default_rope_parametersrattention_scalingregister_bufferclone)r(rFdevice rope_init_fnrEr,s r-r"zCohereRotaryEmbedding.__init__Gs "("@$*$B! 4[A!%!E >Y & &.t~>L+7< V+L+L($( ZeDDD 0(..2B2BuUUUUUr.rTz torch.deviceseq_lenreturnz torch.Tensorc|jd}t|ddp|j|jz}d}d|t jd|dtj|tj|z zz }||fS) a Computes the inverse frequencies according to the original RoPE implementation Args: config ([`~transformers.PreTrainedConfig`]): The model configuration. device (`torch.device`): The device to use for initialization of the inverse frequencies. seq_len (`int`, *optional*): The current sequence length. Unused for this type of RoPE. Returns: Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE). rope_thetahead_dimNg?rr2r3)rTr3) rOgetattrr)num_attention_headsr$arangeint64r4float)rFrTrVbasedimattention_factorrEs r-rPz5CohereRotaryEmbedding.compute_default_rope_parametersWs&%l3fj$//c63EIc3c U\!S!5;???BB&X]XcBddgjj k )))r.c|jddddf|jddd}|dddddf}t |jjtr|jjdkr |jjnd}t|d5||z dd}tj |dd }| |j z}||j z} dddn #1swxYwY||j | |j fS) Nrr0rmpscpuF) device_typeenabledr2rbr[)rEr`expandshape isinstancerTtypestrr transposer$repeat_interleavecosrQsinr4r3) r(x position_idsinv_freq_expandedposition_ids_expandedrgfreqsembrqrrs r-r<zCohereRotaryEmbedding.forwardus!M$4-8>>@@GG HZ[\H]_acdee ,QQQaaaZ 8 > > @ @'1!(-'E'Ek!(-[`J`J`ahmmfk  U C C C 5 5&,,..1F1L1L1N1NNYYZ[]^__E)%;;;C''))d44C''))d44C  5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 vvAGv$$cff17f&;&;;;s4B E  EEN)NNN)r>r?r@r$Tensor__annotations__rr" staticmethodrinttupler`rPno_gradrr<rArBs@r-rDrDDslVV|VVVVVV &*+/"**t#*(*t* ~u$ % ***\*:U]__ < <_ < < < <