有没有办法在tf.GradientTape中为多个输出层应用渐变？

Question

我正在尝试对两个输出模型进行渐变应用，但是结果表明该模型没有学习并且损耗没有减少， 我需要你的支持 谢谢。

@ tf.function def train_step（inp，targ，intent，enc_hidden）：

loss = 0
intent_loss = 0

with tf.GradientTape(persistent= True) as tape:

    enc_output, enc_hidden = encoder(inp, enc_hidden)

    dec_hidden = enc_hidden




    dec_input = tf.expand_dims([targ_lang.word_index['<start>']] * BATCH_SIZE, 1)

    # Teacher forcing - feeding the target as the next input
    for t in range(1, targ.shape[1]):

        # passing enc_output to the decoder
        predictions, dec_hidden, _ =slot_decoder(dec_input, dec_hidden, enc_output)
        intent_pred, _ = intent_decoder(dec_hidden, enc_output)

        loss += loss_function(targ[:, t], predictions)
        intent_loss = loss_function(intent, intent_pred)

        # using teacher forcing
        dec_input = tf.expand_dims(targ[:, t], 1)

batch_loss = (loss / int(targ.shape[1])) + intent_loss

intent_variables = encoder.trainable_variables + intent_decoder.trainable_variables
slot_variables = encoder.trainable_variables + slot_decoder.trainable_variables

intent_gradients = tape.gradient(intent_loss, intent_variables)
slot_gradients = tape.gradient(loss, slot_variables)


optimizer.apply_gradients(zip(intent_gradients, intent_variables))
optimizer.apply_gradients(zip(slot_gradients, slot_variables))

del tape
return batch_loss + intent_loss