def __make_model(self): self.__model = wrapper.make_model( # encoder w_xi = functions.EmbedID(len(self.__src_vocab), self.__n_embed), w_ip = functions.Linear(self.__n_embed, 4 * self.__n_hidden), w_pp = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), # decoder w_pq = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), w_qj = functions.Linear(self.__n_hidden, self.__n_embed), w_jy = functions.Linear(self.__n_embed, len(self.__trg_vocab)), w_yq = functions.EmbedID(len(self.__trg_vocab), 4 * self.__n_hidden), w_qq = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), )
def __make_model(self): self.__model = wrapper.make_model( # input embedding w_xi=functions.EmbedID(len(self.__src_vocab), self.__n_embed), # forward encoder w_ia=functions.Linear(self.__n_embed, 4 * self.__n_hidden), w_aa=functions.Linear(self.__n_hidden, 4 * self.__n_hidden), # backward encoder w_ib=functions.Linear(self.__n_embed, 4 * self.__n_hidden), w_bb=functions.Linear(self.__n_hidden, 4 * self.__n_hidden), # attentional weight estimator w_aw=functions.Linear(self.__n_hidden, self.__n_hidden), w_bw=functions.Linear(self.__n_hidden, self.__n_hidden), w_pw=functions.Linear(self.__n_hidden, self.__n_hidden), w_we=functions.Linear(self.__n_hidden, 1), # decoder w_ap=functions.Linear(self.__n_hidden, self.__n_hidden), w_bp=functions.Linear(self.__n_hidden, self.__n_hidden), w_yp=functions.EmbedID(len(self.__trg_vocab), 4 * self.__n_hidden), w_pp=functions.Linear(self.__n_hidden, 4 * self.__n_hidden), w_cp=functions.Linear(self.__n_hidden, 4 * self.__n_hidden), w_dp=functions.Linear(self.__n_hidden, 4 * self.__n_hidden), w_py=functions.Linear(self.__n_hidden, len(self.__trg_vocab)), )
def __make_model(self): self.__model = wrapper.make_model( # input embedding w_xi = functions.EmbedID(len(self.__src_vocab), self.__n_embed), # forward encoder w_ia = functions.Linear(self.__n_embed, 4 * self.__n_hidden), w_aa = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), # backward encoder w_ib = functions.Linear(self.__n_embed, 4 * self.__n_hidden), w_bb = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), # attentional weight estimator w_aw = functions.Linear(self.__n_hidden, self.__n_hidden), w_bw = functions.Linear(self.__n_hidden, self.__n_hidden), w_pw = functions.Linear(self.__n_hidden, self.__n_hidden), w_we = functions.Linear(self.__n_hidden, 1), # decoder w_ap = functions.Linear(self.__n_hidden, self.__n_hidden), w_bp = functions.Linear(self.__n_hidden, self.__n_hidden), w_yp = functions.EmbedID(len(self.__trg_vocab), 4 * self.__n_hidden), w_pp = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), w_cp = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), w_dp = functions.Linear(self.__n_hidden, 4 * self.__n_hidden), w_py = functions.Linear(self.__n_hidden, len(self.__trg_vocab)), )