qiUXddlZddlmZddlmZddlZddlmZddlm Z ddl m Z m Z ddl mZddlmZdd lmZdd lmZmZmZdd lmZmZdd lmZmZdd lmZmZddl m!Z!ddl"m#Z#m$Z$m%Z%ddl&m'Z'm(Z(ddl)m*Z*ddl+m,Z,GddejZZ.GddejZZ/GddejZZ0dejbde2dejbfdZ3 d8dejZdejbd ejbd!ejbd"ejbdzd#e4d$e4d%e!e#fd&Z5d'Z6d9d(Z7ee7Gd)d*ejZZ8Gd+d,eZ9e$Gd-d.eZ:e$Gd/d0e:Z;e$Gd1d2e:eZ<Gd3d4ee:Z=Gd5d6ee:Z>gd7Z?y):N)Callable)Optional)ACT2FN)Cache DynamicCache)GenerationMixin)use_kernelized_func)create_causal_mask) GenericForSequenceClassificationGenericForTokenClassificationGradientCheckpointingLayer)BaseModelOutputWithPastCausalLMOutputWithPast)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)TransformersKwargsauto_docstringcan_return_tuple)maybe_autocastmerge_with_config_defaults)capture_outputs) HeliumConfigc,eZdZdfd ZdZdZxZS) HeliumRMSNormct|tjt j ||_||_yN)super__init__nn Parametertorchonesweightvariance_epsilon)self hidden_sizeeps __class__s \/opt/pipecat/venv/lib/python3.12/site-packages/transformers/models/helium/modeling_helium.pyr#zHeliumRMSNorm.__init__1s/ ll5::k#:; #c\|j}|jtj}|j dj dd}|tj ||jzz}|jjtj|zj|S)NT)keepdim) dtypetor&float32powmeanrsqrtr)r()r* hidden_states input_dtypevariances r.forwardzHeliumRMSNorm.forward6s#)) %((7  $$Q',,R,>% Ht?T?T4T(UU  u}}- =AA+NNr/c^t|jjd|jS)Nz, eps=)tupler(shaper))r*s r. extra_reprzHeliumRMSNorm.extra_repr=s* ))*+6$2G2G1HIIr/)gư>)__name__ __module__ __qualname__r#r=rA __classcell__r-s@r.rr0s$ OJr/rc eZdZUejed<d deffd Ze d dedzde dde dzde d e ffd Z ejed ZxZS)HeliumRotaryEmbeddinginv_freqNconfigct||j|_|j|_||_|j j d|_|j}|jdk7rt|j}||j |\}|_ |jd|d|jd|jdy)N rope_typedefaultrIF) persistentoriginal_inv_freq) r"r#max_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenrJrope_parametersrLcompute_default_rope_parametersrattention_scalingregister_bufferclone)r*rJdevice rope_init_fnrIr-s r.r#zHeliumRotaryEmbedding.__init__Ds "("@"@$*$B$B! 44[A!%!E!E >>Y &.t~~>L+7 V+L($( ZeD 0(..2BuUr/rXz 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?rr1r4)rXr4) rSgetattrr+num_attention_headsr&arangeint64r5float)rJrXrZbasedimattention_factorrIs r.rTz5HeliumRotaryEmbedding.compute_default_rope_parametersTs&%%l3fj$/c63E3EIcIc3c U\\!S!5;;?BB&X]XcXcBdgjj k )))r/cN|jddddfjj|jdddj |j }|dddddfj}t |j jtr/|j jdk7r|j jnd}t|d5|j|jzjdd}tj||fd }|j|jz}|j|jz} dddj |j  j |j fS#1swYAxYw) Nrr2rmpscpuF) device_typeenabledr1rfr_)rIrdexpandr@r5rX isinstancetypestrr transposer&catcosrUsinr4) r*x position_idsinv_freq_expandedposition_ids_expandedrkfreqsembrtrus r.r=zHeliumRotaryEmbedding.forwardrsR!MM$4-8>>@GG HZHZ[\H]_acdehhijiqiqr ,QaZ 8 > > @'1!((--'E!((--[`J`ahhmmfk  U C 5&,,.1F1L1L1NNYYZ[]^_E))UEN3C'')d444C'')d444C  5 vvAGGv$cff177f&;;;  5 5s BFF$r!)NNN)rBrCrDr&Tensor__annotations__rr# staticmethodrintr?rdrTno_gradrr=rErFs@r.rHrHAsllV|V &*+/"*t#*(*t* ~u$ % **:U]]_ <