How to use the transformers.modeling_tf_utils.shape_list function in transformers

def split_heads(self, x):
        x_shape = shape_list(x)
        new_x_shape = x_shape[:-1] + [self.n_head, x_shape[-1] // self.n_head]
        x = tf.reshape(x, new_x_shape)
        return tf.transpose(x, (0, 2, 1, 3))  # (batch, head, seq_length, head_features)

def split_heads(self, x):
        x_shape = shape_list(x)
        new_x_shape = x_shape[:-1] + [self.n_head, x_shape[-1] // self.n_head]
        x = tf.reshape(x, new_x_shape)
        return tf.transpose(x, (0, 2, 1, 3))  # (batch, head, seq_length, head_features)

# attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        if not head_mask is None:
            raise NotImplementedError
        else:
            head_mask = [None] * self.num_hidden_layers
            # head_mask = tf.constant([0] * self.num_hidden_layers)

        position_ids = tf.reshape(position_ids, [-1, shape_list(position_ids)[-1]])

        if inputs_embeds is None:
            inputs_embeds = self.tokens_embed(input_ids, mode='embedding')
        position_embeds = self.positions_embed(position_ids)
        if token_type_ids is not None:
            token_type_ids = tf.reshape(token_type_ids, [-1, shape_list(token_type_ids)[-1]])
            token_type_embeds = self.tokens_embed(token_type_ids, mode='embedding')
        else:
            token_type_embeds = 0
        hidden_states = inputs_embeds + position_embeds + token_type_embeds
        hidden_states = self.drop(hidden_states, training=training)

        output_shape = input_shape + [shape_list(hidden_states)[-1]]

        all_attentions = []
        all_hidden_states = ()
        for i, block in enumerate(self.h):
            if self.output_hidden_states:
                all_hidden_states = all_hidden_states + (tf.reshape(hidden_states, output_shape),)

            outputs = block([hidden_states, attention_mask, head_mask[i]], training=training)
            hidden_states = outputs[0]

def _attn(self, inputs, training=False):
        q, k, v, attention_mask, head_mask = inputs
        # q, k, v have shape [batch, heads, sequence, features]
        w = tf.matmul(q, k, transpose_b=True)
        if self.scale:
            dk = tf.cast(tf.shape(k)[-1], tf.float32) # scale attention_scores
            w = w / tf.math.sqrt(dk)

        # w has shape [batch, heads, dst_sequence, src_sequence], where information flows from src to dst.
        _, _, nd, ns = shape_list(w)
        b = self.causal_attention_mask(nd, ns, dtype=w.dtype)
        b = tf.reshape(b, [1, 1, nd, ns])
        w = w * b - 1e4 * (1 - b)

        if attention_mask is not None:
            # Apply the attention mask
            w = w + attention_mask

        w = tf.nn.softmax(w, axis=-1)
        w = self.attn_dropout(w, training=training)

        # Mask heads if we want to
        if head_mask is not None:
            w = w * head_mask

        outputs = [tf.matmul(w, v)]

position_ids = inputs.get('position_ids', position_ids)
            head_mask = inputs.get('head_mask', head_mask)
            assert len(inputs) <= 6, "Too many inputs."
        else:
            input_ids = inputs

        input_shape = shape_list(input_ids)
        input_ids = tf.reshape(input_ids, [-1, input_shape[-1]])

        if past is None:
            past_length = 0
            past = [None] * len(self.h)
        else:
            past_length = shape_list(past[0][0])[-2]
        if position_ids is None:
            position_ids = tf.range(past_length, shape_list(input_ids)[-1] + past_length, dtype=tf.int32)[tf.newaxis, :]
            position_ids = tf.tile(position_ids, [shape_list(input_ids)[0], 1])

        # Attention mask.
        if attention_mask is not None:
            # We create a 3D attention mask from a 2D tensor mask.
            # Sizes are [batch_size, 1, 1, to_seq_length]
            # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
            # this attention mask is more simple than the triangular masking of causal attention
            # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
            attention_mask = attention_mask[:, tf.newaxis, tf.newaxis, :]

            # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
            # masked positions, this operation will create a tensor which is 0.0 for
            # positions we want to attend and -10000.0 for masked positions.
            # Since we are adding it to the raw scores before the softmax, this is
            # effectively the same as removing these entirely.

# input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        if not head_mask is None:
            raise NotImplementedError
        else:
            head_mask = [None] * self.num_hidden_layers
            # head_mask = tf.constant([0] * self.num_hidden_layers)

        input_shape = shape_list(input_ids)
        input_ids = tf.reshape(input_ids, [-1, input_shape[-1]])
        position_ids = tf.reshape(position_ids, [-1, shape_list(position_ids)[-1]])

        inputs_embeds = self.wte(input_ids, mode='embedding')
        position_embeds = self.wpe(position_ids)
        if token_type_ids is not None:
            token_type_ids = tf.reshape(token_type_ids, [-1, shape_list(token_type_ids)[-1]])
            token_type_embeds = self.wte(token_type_ids, mode='embedding')
        else:
            token_type_embeds = 0
        hidden_states = inputs_embeds + position_embeds + token_type_embeds
        hidden_states = self.drop(hidden_states, training=training)

        output_shape = input_shape + [shape_list(hidden_states)[-1]]

        presents = ()
        all_attentions = []
        all_hidden_states = ()
        for i, (block, layer_past) in enumerate(zip(self.h, past)):
            if self.output_hidden_states:
                all_hidden_states = all_hidden_states + (tf.reshape(hidden_states, output_shape),)

            outputs = block([hidden_states, layer_past, attention_mask, head_mask[i]], training=training)

attention_mask = (1.0 - attention_mask) * -10000.0
        else:
            attention_mask = None

        # Prepare head mask if needed
        # 1.0 in head_mask indicate we keep the head
        # attention_probs has shape bsz x n_heads x N x N
        # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
        # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
        if not head_mask is None:
            raise NotImplementedError
        else:
            head_mask = [None] * self.num_hidden_layers
            # head_mask = tf.constant([0] * self.num_hidden_layers)

        position_ids = tf.reshape(position_ids, [-1, shape_list(position_ids)[-1]])

        if inputs_embeds is None:
            inputs_embeds = self.tokens_embed(input_ids, mode='embedding')
        position_embeds = self.positions_embed(position_ids)
        if token_type_ids is not None:
            token_type_ids = tf.reshape(token_type_ids, [-1, shape_list(token_type_ids)[-1]])
            token_type_embeds = self.tokens_embed(token_type_ids, mode='embedding')
        else:
            token_type_embeds = 0
        hidden_states = inputs_embeds + position_embeds + token_type_embeds
        hidden_states = self.drop(hidden_states, training=training)

        output_shape = input_shape + [shape_list(hidden_states)[-1]]

        all_attentions = []
        all_hidden_states = ()

head_mask = inputs[5] if len(inputs) > 5 else head_mask
            mc_token_ids = inputs[6] if len(inputs) > 6 else mc_token_ids
            assert len(inputs) <= 7, "Too many inputs."
        elif isinstance(inputs, dict):
            input_ids = inputs.get('input_ids')
            past = inputs.get('past', past)
            attention_mask = inputs.get('attention_mask', attention_mask)
            token_type_ids = inputs.get('token_type_ids', token_type_ids)
            position_ids = inputs.get('position_ids', position_ids)
            head_mask = inputs.get('head_mask', head_mask)
            mc_token_ids = inputs.get('mc_token_ids', mc_token_ids)
            assert len(inputs) <= 7, "Too many inputs."
        else:
            input_ids = inputs

        input_shapes = shape_list(input_ids)

        seq_length = input_shapes[-1]

        flat_input_ids = tf.reshape(input_ids, (-1, seq_length))
        flat_attention_mask = tf.reshape(attention_mask, (-1, seq_length)) if attention_mask is not None else None
        flat_token_type_ids = tf.reshape(token_type_ids, (-1, seq_length)) if token_type_ids is not None else None
        flat_position_ids = tf.reshape(position_ids, (-1, seq_length)) if position_ids is not None else None

        flat_inputs = [flat_input_ids, past, flat_attention_mask, flat_token_type_ids, flat_position_ids, head_mask]

        transformer_outputs = self.transformer(flat_inputs, training=training)
        hidden_states = transformer_outputs[0]

        hidden_states = tf.reshape(hidden_states, input_shapes + shape_list(hidden_states)[-1:])

        lm_logits = self.transformer.wte(hidden_states, mode="linear")

if self.output_attentions:
                all_attentions.append(outputs[2])

        hidden_states = self.layernorm(hidden_states)
        hidden_states = tf.reshape(hidden_states, output_shape)
        if self.output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        outputs = (hidden_states,)
        if self.output_past:
            outputs = outputs + (presents,)
        if self.output_hidden_states:
            outputs = outputs + (all_hidden_states,)
        if self.output_attentions:
            # let the number of heads free (-1) so we can extract attention even after head pruning
            attention_output_shape = input_shape[:-1] + [-1] + shape_list(all_attentions[0])[-2:]
            all_attentions = tuple(tf.reshape(t, attention_output_shape) for t in all_attentions)
            outputs = outputs + (all_attentions,)
        return outputs