qib$ddlmZddlmZddlZddlmZddlmZddlm Z m Z ddl m Z dd l mZdd lmZdd lmZdd lmZdd lmZmZddlmZmZddlmZmZddlmZddl m!Z!m"Z"m#Z#ddl$m%Z%m&Z&ddl'm(Z(ddl)m*Z*GddejVZ,GddejVZ-GddejVZ.dej^de0dej^fdZ1 d6d ejVd!ej^d"ej^d#ej^d$ej^dzd%e2d&e2d'ee!fd(Z3d)Z4d7d*Z5ee5Gd+d,ejVZ6Gd-d.eZ7e"Gd/d0eZ8e"Gd1d2e8Z9e"Gd3d4e8e Z:gd5Z;y)8)Callable)OptionalN)nn)ACT2FN)Cache DynamicCache)GenerationMixin)use_kernelized_func)create_causal_mask)FlashAttentionKwargs)GradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPast)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)TransformersKwargsauto_docstringcan_return_tuple)maybe_autocastmerge_with_config_defaults)capture_outputs) CohereConfigc&eZdZdfd ZdZxZS)CohereLayerNormct|tjt j ||_||_y)zcThe hidden size can be a tuple or an int. The tuple is used for QKNorm to normalize across head_dimN)super__init__r Parametertorchonesweightvariance_epsilon)self hidden_sizeepsbias __class__s \/opt/pipecat/venv/lib/python3.12/site-packages/transformers/models/cohere/modeling_cohere.pyr"zCohereLayerNorm.__init__5s/ ll5::k#:; #c|j}|jtj}|j dd}||z j dj dd}||z tj ||jzz}|jjtj|z}|j|S)NT)keepdim) dtypetor$float32meanpowrsqrtr'r&)r( hidden_states input_dtyper6variances 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-rr4s $ -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_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__Hs "("@"@$*$B$B! 44[A!%!E!E >>Y &.t~~>L+7 V+L($( ZeD 0(..2BuUr.rTz 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?rr2r3)rTr3) rOgetattrr)num_attention_headsr$arangeint64r4float)rFrTrVbasedimattention_factorrEs r-rPz5CohereRotaryEmbedding.compute_default_rope_parametersXs&%%l3fj$/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) Nrr0rmpscpuF) device_typeenabledr2rbr[)rEr`expandshape isinstancerTtypestrr transposer$repeat_interleavecosrQsinr4r3) r(x position_idsinv_freq_expandedposition_ids_expandedrgfreqsembrqrrs r-r<zCohereRotaryEmbedding.forwardvs@!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$Tensor__annotations__rr" staticmethodrinttupler`rPno_gradrr<rArBs@r-rDrDEsllV|V &*+/"*t#*(*t* ~u$ % **:U]]_ <