+ NV-jT"^RIt^RIHtHt^RIHt^RIHt^RI H t !RR] 4t !RR] 4t !R R ] 4t R#) N)CallableOptional)tanh)ModulecVaa]tRt^ toRtRV3RlV3RllltRtRRltRtVt V;t #) RNNaWAn Elman recurrent layer. The input is a sequence of shape ``NLD`` or ``LD`` where: * ``N`` is the optional batch dimension * ``L`` is the sequence length * ``D`` is the input's feature dimension Concretely, for each element along the sequence length axis, this layer applies the function: .. math:: h_{t + 1} = \text{tanh} (W_{ih}x_t + W_{hh}h_t + b) The hidden state :math:`h` has shape ``NH`` or ``H``, depending on whether the input is batched or not. Returns the hidden state at each time step, of shape ``NLH`` or ``LH``. Args: input_size (int): Dimension of the input, ``D``. hidden_size (int): Dimension of the hidden state, ``H``. bias (bool, optional): Whether to use a bias. Default: ``True``. nonlinearity (callable, optional): Non-linearity to use. If ``None``, then func:`tanh` is used. Default: ``None``. c B<V^8dQhRS[RS[RS[RS[S[,/#) input_size hidden_sizebias nonlinearity)intboolrr)format __classdict__s"h/Volumes/fast/ai/experiments/ui-tars-smoke/.venv/lib/python3.14/site-packages/mlx/nn/layers/recurrent.py __annotate__RNN.__annotate__'s8      x(  c<\SV`4T;'g\Vn\ VP4'g\ RV R24hR\ P!V4, pW n\PPV)WRV3R7Vn \PPV)WRV3R7Vn V'd+\PPV)WR3R7VnR#RVnR#)z.Nonlinearity must be callable. Current value: .?lowhighshapeN)super__init__rrcallable ValueErrormathsqrtr mxrandomuniformWxhWhhr )selfr r r rscale __class__s&&&&& rr RNN.__init__'s (00D))**@aP dii ,,&99$$U *C% 99$$U *D%   II  5&uN  K  rc RVPP^, RVP RVP RVPRJ 2#) input_dims=, hidden_size=z, nonlinearity=, bias=N)r'rr rr r)s&r _extra_reprRNN._extra_reprDsT$((..+,-++,- --.gdiit6K5L N rc VPe7\P!VPWPP4pMWPP,p.p\ VP R,4FkpVe7\P!VRVR3,W PP4pM VRVR3,pVPV4pVPV4Km \P!VRR7#)N.NNNaxis) r r$addmmr'Trangerr(rappendstack)r)xhidden all_hiddenidxs&&& r__call__ RNN.__call__Ks 99 Axxzz2AHHJJA %C!!CaK.&((**E3Q;&&v.F   f % &xx ,,r)r(r'r r r)TNN __name__ __module__ __qualname____firstlineno____doc__rr2rB__static_attributes____classdictcell__ __classcell__r+rs@@rrr s#6  : --rrcVaa]tRt^]toRtRV3RlV3RllltRtRRltRtVt V;t #) GRUaA gated recurrent unit (GRU) RNN layer. The input has shape ``NLD`` or ``LD`` where: * ``N`` is the optional batch dimension * ``L`` is the sequence length * ``D`` is the input's feature dimension Concretely, for each element of the sequence, this layer computes: .. math:: \begin{aligned} r_t &= \sigma (W_{xr}x_t + W_{hr}h_t + b_{r}) \\ z_t &= \sigma (W_{xz}x_t + W_{hz}h_t + b_{z}) \\ n_t &= \text{tanh}(W_{xn}x_t + b_{n} + r_t \odot (W_{hn}h_t + b_{hn})) \\ h_{t + 1} &= (1 - z_t) \odot n_t + z_t \odot h_t \end{aligned} The hidden state :math:`h` has shape ``NH`` or ``H`` depending on whether the input is batched or not. Returns the hidden state at each time step of shape ``NLH`` or ``LH``. Args: input_size (int): Dimension of the input, ``D``. hidden_size (int): Dimension of the hidden state, ``H``. bias (bool): Whether to use biases or not. Default: ``True``. c,<V^8dQhRS[RS[RS[/#r r r r rr)rrs"rrGRU.__annotate__{s)      rc<\SV`4W nR\P!V4, p\ P PV)V^V,V3R7Vn\ P PV)V^V,V3R7Vn V'd,\ P PV)V^V,3R7MRVn V'd+\ P PV)WB3R7Vn R#RVn R#rrN) rrr r"r#r$r%r&WxWhbbhnr)r r r r*r+s&&&& rr GRU.__init__{s &dii ,,))##U1{?J*G$ ))##U1{?K*H$   II  5&uQ_r?x_rzx_nr@rArzh_proj h_proj_rzh_proj_nrzns&&& rrB GRU.__call__s 66 GGII.AGGII A))))))*d&&&(() %Cc3k"B!''))+"3(;4+;+;*;(;#;< !#(8(8'8':":;88'(H^BB88B+DACaK A!H $%DHH $ A!a%1qz1a%1   f %;&>xx ,,r)rXrWrYrZr TrDrErNs@@rrPrP]s#:  6 *-*-rrPcVaa]tRt^toRtRV3RlV3RllltRtRRltRtVt V;t #) LSTMaAn LSTM recurrent layer. The input has shape ``NLD`` or ``LD`` where: * ``N`` is the optional batch dimension * ``L`` is the sequence length * ``D`` is the input's feature dimension Concretely, for each element of the sequence, this layer computes: .. math:: \begin{aligned} i_t &= \sigma (W_{xi}x_t + W_{hi}h_t + b_{i}) \\ f_t &= \sigma (W_{xf}x_t + W_{hf}h_t + b_{f}) \\ g_t &= \text{tanh} (W_{xg}x_t + W_{hg}h_t + b_{g}) \\ o_t &= \sigma (W_{xo}x_t + W_{ho}h_t + b_{o}) \\ c_{t + 1} &= f_t \odot c_t + i_t \odot g_t \\ h_{t + 1} &= o_t \text{tanh}(c_{t + 1}) \end{aligned} The hidden state :math:`h` and cell state :math:`c` have shape ``NH`` or ``H``, depending on whether the input is batched or not. The layer returns two arrays, the hidden state and the cell state at each time step, both of shape ``NLH`` or ``LH``. Args: input_size (int): Dimension of the input, ``D``. hidden_size (int): Dimension of the hidden state, ``H``. bias (bool): Whether to use biases or not. Default: ``True``. c,<V^8dQhRS[RS[RS[/#rRrS)rrs"rrLSTM.__annotate__s)      rc<\SV`4W nR\P!V4, p\ P PV)V^V,V3R7Vn\ P PV)V^V,V3R7Vn V'd3\ P PV)V^V,3R7Vn R#RVn R#rV) rrr r"r#r$r%r&rWrXr r[s&&&& rr LSTM.__init__s &dii ,,))##U1{?J*G$ ))##U1{?K*H$   II  5&uQ_r?cellr@all_cellrAifgoifgos &&&& rrB LSTM.__call__s9 99 Awwyy1AGGII A %CS#q[>D!xxdggii8$3JA! 1 A 1 A A 1 Ax!%'u&F OOD !   f %%&(xx ,bhhxb.IIIr)rXrWr r ro)NNrErNs@@rrqrqs&@  , JJrrq)r"typingrrmlx.corecorer$mlx.nn.layers.activationsrmlx.nn.layers.baserrrPrqrrrsD %*%O-&O-di-&i-XZJ6ZJr