modeling_linglong.py

import math
import torch
import torch.utils.checkpoint

from torch import nn
from torch.cuda.amp import autocast
from transformers.utils import logging
from transformers.activations import ACT2FN
from transformers.pytorch_utils import Conv1D
from transformers.modeling_utils import PreTrainedModel
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutput

from .configuration_linglong import LingLongConfig

logger = logging.get_logger(__name__)


class LingLongAttention(nn.Module):

    def __init__(self, config, layer_idx=None):
        super().__init__()

        n_position = config.n_position
        self.register_buffer(
            'bias',
            torch.tril(torch.ones((n_position, n_position), dtype=torch.bool)).view(
                1, 1, n_position, n_position
            ),
            persistent=False,
        )
        self.register_buffer('masked_bias', torch.tensor(-1e4), persistent=False)

        self.n_embd = config.n_embd
        self.n_head = config.n_head
        self.head_dim = self.n_embd // self.n_head
        self.split_size = self.n_embd
        if self.head_dim * self.n_head != self.n_embd:
            raise ValueError(
                f'`embed_dim` must be divisible by num_heads (got `embed_dim`: {self.n_embd} and `num_heads`:'
                f' {self.n_head}).'
            )

        self.scale_attn_weights = config.scale_attn_weights

        # Layer-wise attention scaling, reordering, and upcasting
        self.scale_attn_by_inverse_layer_idx = config.scale_attn_by_inverse_layer_idx
        self.layer_idx = layer_idx
        self.reorder_and_upcast_attn = config.reorder_and_upcast_attn

        self.c_attn = Conv1D(3 * self.n_embd, self.n_embd)
        self.c_proj = Conv1D(self.n_embd, self.n_embd)

        self.attn_dropout = nn.Dropout(config.attn_pdrop)
        self.resid_dropout = nn.Dropout(config.resid_pdrop)

        # LingLong sparse attention.
        self.mode = config.attn_mode
        self.stride = config.attn_stride
        self.c = config.attn_c
        self.causal_mask = None

    def _causal_mask(self, query_length, key_length):
        return self.bias[:, :, key_length - query_length: key_length, :key_length]

    def _sparse_causal_mask(self, query_length, key_length):
        layout = torch.zeros([key_length, key_length], dtype=torch.bool, device=self.bias.device)
        for idx in range(self.c):
            layout[:, (self.stride - 1 - idx)::self.stride] = 1
        for q_idx in range(key_length):
            row = q_idx // self.stride
            layout[q_idx, row * self.stride:(row + 1) * self.stride] = 1
            # Any query cannot attend to keys above it.
            layout[q_idx, q_idx + 1:] = 0
        return layout[(key_length - query_length):].view(1, 1, query_length, key_length)

    def _attn(self, query, key, value, attention_mask=None):
        attn_weights = torch.matmul(query, key.transpose(-1, -2))

        if self.scale_attn_weights:
            attn_weights = attn_weights / torch.full(
                [], value.size(-1) ** 0.5, dtype=attn_weights.dtype, device=attn_weights.device
            )

        # Layer-wise attention scaling
        if self.scale_attn_by_inverse_layer_idx:
            attn_weights = attn_weights / float(self.layer_idx + 1)

        # if only "normal" attention layer implements causal mask
        query_length, key_length = query.size(-2), key.size(-2)
        if self.causal_mask is None or self.causal_mask.size() != torch.Size([1, 1, query_length, key_length]):
            if self.mode == 'sparse' and self.layer_idx % 2 != 0:
                self.causal_mask = self._sparse_causal_mask(query_length, key_length)
            else:
                self.causal_mask = self._causal_mask(query_length, key_length)
        mask_value = torch.finfo(attn_weights.dtype).min
        # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
        # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
        mask_value = torch.full([], mask_value, dtype=attn_weights.dtype, device=attn_weights.device)
        attn_weights = torch.where(self.causal_mask, attn_weights.to(attn_weights.dtype), mask_value)

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

        attn_weights = nn.functional.softmax(attn_weights, dim=-1)

        # Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op otherwise
        attn_weights = attn_weights.type(value.dtype)
        attn_weights = self.attn_dropout(attn_weights)

        attn_output = torch.matmul(attn_weights, value)

        return attn_output, attn_weights

    def _upcast_and_reordered_attn(self, query, key, value, attention_mask=None):
        # Use `torch.baddbmm` (a bit more efficient w/ alpha param for scaling -- from Megatron-LM)
        bsz, num_heads, q_seq_len, dk = query.size()
        _, _, k_seq_len, _ = key.size()

        # Preallocate attn_weights for `baddbmm`
        attn_weights = torch.empty(bsz * num_heads, q_seq_len, k_seq_len, dtype=torch.float32, device=query.device)

        # Compute Scale Factor
        scale_factor = 1.0
        if self.scale_attn_weights:
            scale_factor /= float(value.size(-1)) ** 0.5

        if self.scale_attn_by_inverse_layer_idx:
            scale_factor /= float(self.layer_idx + 1)

        # Upcast (turn off autocast) and reorder (Scale K by 1 / root(dk))
        with autocast(enabled=False):
            q, k = query.reshape(-1, q_seq_len, dk), key.transpose(-1, -2).reshape(-1, dk, k_seq_len)
            attn_weights = torch.baddbmm(attn_weights, q.float(), k.float(), beta=0, alpha=scale_factor)
            attn_weights = attn_weights.reshape(bsz, num_heads, q_seq_len, k_seq_len)

        # if only "normal" attention layer implements causal mask
        query_length, key_length = query.size(-2), key.size(-2)
        if self.causal_mask is None or self.causal_mask.size() != torch.Size([1, 1, query_length, key_length]):
            if self.mode == 'sparse' and self.layer_idx % 2 != 0:
                self.causal_mask = self._sparse_causal_mask(query_length, key_length)
            else:
                self.causal_mask = self._causal_mask(query_length, key_length)
        mask_value = torch.finfo(attn_weights.dtype).min
        # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
        # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
        mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(attn_weights.device)
        attn_weights = torch.where(self.causal_mask, attn_weights, mask_value)

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

        attn_weights = nn.functional.softmax(attn_weights, dim=-1)

        # Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op if otherwise
        if attn_weights.dtype != torch.float32:
            raise RuntimeError('Error with upcasting, attn_weights does not have dtype torch.float32.')
        attn_weights = attn_weights.type(value.dtype)
        attn_weights = self.attn_dropout(attn_weights)

        attn_output = torch.matmul(attn_weights, value)

        return attn_output, attn_weights

    @staticmethod
    def _split_heads(tensor, num_heads, attn_head_size):
        """
        Splits hidden_size dim into attn_head_size and num_heads
        """
        new_shape = tensor.size()[:-1] + (num_heads, attn_head_size)
        tensor = tensor.view(new_shape)
        return tensor.permute(0, 2, 1, 3)  # (batch, head, seq_length, head_features)

    @staticmethod
    def _merge_heads(tensor, num_heads, attn_head_size):
        """
        Merges attn_head_size dim and num_attn_heads dim into hidden_size
        """
        tensor = tensor.permute(0, 2, 1, 3).contiguous()
        new_shape = tensor.size()[:-2] + (num_heads * attn_head_size,)
        return tensor.view(new_shape)

    def forward(
            self,
            hidden_states,
            layer_past=None,
            attention_mask=None,
            use_cache=False,
            output_attentions=False,
    ):
        query, key, value = self.c_attn(hidden_states).split(self.split_size, dim=2)
        query = self._split_heads(query, self.n_head, self.head_dim)
        key = self._split_heads(key, self.n_head, self.head_dim)
        value = self._split_heads(value, self.n_head, self.head_dim)

        if layer_past is not None:
            past_key, past_value = layer_past
            key = torch.cat((past_key, key), dim=-2)
            value = torch.cat((past_value, value), dim=-2)

        if use_cache is True:
            present = (key, value)
        else:
            present = None

        if self.reorder_and_upcast_attn:
            attn_output, attn_weights = self._upcast_and_reordered_attn(query, key, value, attention_mask)
        else:
            attn_output, attn_weights = self._attn(query, key, value, attention_mask)

        attn_output = self._merge_heads(attn_output, self.n_head, self.head_dim)
        attn_output = self.c_proj(attn_output)
        attn_output = self.resid_dropout(attn_output)

        outputs = (attn_output, present)
        if output_attentions:
            outputs += (attn_weights,)

        return outputs  # a, present, (attentions)


class LingLongMLP(nn.Module):

    def __init__(self, intermediate_size, config):
        super().__init__()
        n_embd = config.n_embd
        self.c_fc = Conv1D(intermediate_size, n_embd)
        self.c_proj = Conv1D(n_embd, intermediate_size)
        self.act = ACT2FN[config.activation_function]
        self.dropout = nn.Dropout(config.resid_pdrop)

    def forward(self, hidden_states):
        hidden_states = self.c_fc(hidden_states)
        hidden_states = self.act(hidden_states)
        hidden_states = self.c_proj(hidden_states)
        hidden_states = self.dropout(hidden_states)
        return hidden_states


class LingLongBlock(nn.Module):

    def __init__(self, config, layer_idx=None):
        super().__init__()
        n_embd = config.n_embd
        inner_dim = config.n_inner if config.n_inner is not None else 4 * n_embd

        self.ln_1 = nn.LayerNorm(n_embd, eps=config.layer_norm_epsilon)
        self.attn = LingLongAttention(config, layer_idx=layer_idx)
        self.ln_2 = nn.LayerNorm(n_embd, eps=config.layer_norm_epsilon)

        self.mlp = LingLongMLP(inner_dim, config)

    def forward(
            self,
            hidden_states,
            layer_past=None,
            attention_mask=None,
            use_cache=False,
            output_attentions=False,
    ):
        residual = hidden_states
        hidden_states = self.ln_1(hidden_states)
        attn_outputs = self.attn(
            hidden_states,
            layer_past=layer_past,
            attention_mask=attention_mask,
            use_cache=use_cache,
            output_attentions=output_attentions,
        )
        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
        outputs = attn_outputs[1:]
        # residual connection
        hidden_states = attn_output + residual

        residual = hidden_states
        hidden_states = self.ln_2(hidden_states)
        feed_forward_hidden_states = self.mlp(hidden_states)
        # residual connection
        hidden_states = residual + feed_forward_hidden_states

        if use_cache:
            outputs = (hidden_states,) + outputs
        else:
            outputs = (hidden_states,) + outputs[1:]

        return outputs  # hidden_states, present, (attentions)


class LingLongPreTrainedModel(PreTrainedModel):
    """
    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
    models.
    """

    config_class = LingLongConfig
    base_model_prefix = 'transformer'
    supports_gradient_checkpointing = True
    _no_split_modules = ['LingLongBlock']
    _skip_keys_device_placement = 'past_key_values'

    def __init__(self, *inputs, **kwargs):
        super().__init__(*inputs, **kwargs)

    def _init_weights(self, module):
        """Initialize the weights."""
        if isinstance(module, (nn.Linear, Conv1D)):
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()
        elif isinstance(module, nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)

        # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
        #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
        #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
        #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
        #
        # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
        for name, p in module.named_parameters():
            if name == 'c_proj.weight':
                # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
                p.data.normal_(mean=0.0, std=(self.config.initializer_range / math.sqrt(2 * self.config.n_layer)))


class LingLongModel(LingLongPreTrainedModel):

    def __init__(self, config: LingLongConfig):
        super().__init__(config)
        self.n_embd = config.n_embd
        self.wte = nn.Embedding(config.vocab_size, self.n_embd)
        self.wpe = nn.Embedding(config.n_position, self.n_embd)
        self.drop = nn.Dropout(config.embd_pdrop)
        self.h = nn.ModuleList([LingLongBlock(config, layer_idx=i) for i in range(config.n_layer)])
        self.ln_f = nn.LayerNorm(self.n_embd, eps=config.layer_norm_epsilon)

        # Model parallel
        self.gradient_checkpointing = False

        # Initialize weights and apply final processing
        self.post_init()

    def get_input_embeddings(self):
        return self.wte

    def set_input_embeddings(self, new_embeddings):
        self.wte = new_embeddings

    def forward(
            self,
            input_ids: torch.LongTensor | None = None,
            past_key_values: tuple[tuple[torch.Tensor]] | None = None,
            attention_mask: torch.FloatTensor | None = None,
            position_ids: torch.LongTensor | None = None,
            inputs_embeds: torch.FloatTensor | None = None,
            use_cache: bool | None = None,
            output_attentions: bool | None = None,
            output_hidden_states: bool | None = None,
            return_dict: bool | None = None,
    ) -> tuple | BaseModelOutputWithPast:
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if input_ids is not None and inputs_embeds is not None:
            raise ValueError('You cannot specify both input_ids and inputs_embeds at the same time.')
        elif input_ids is not None:
            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
            input_shape = input_ids.size()
            input_ids = input_ids.view(-1, input_shape[-1])
            batch_size = input_ids.shape[0]
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
            batch_size = inputs_embeds.shape[0]
        else:
            raise ValueError('You have to specify either input_ids or inputs_embeds.')

        device = input_ids.device if input_ids is not None else inputs_embeds.device

        if past_key_values is None:
            past_length = 0
            past_key_values = tuple([None] * len(self.h))
        else:
            past_length = past_key_values[0][0].size(-2)
        if position_ids is None:
            position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
            position_ids = position_ids.unsqueeze(0)

        # LingLongAttention mask.
        if attention_mask is not None:
            if batch_size <= 0:
                raise ValueError('batch_size has to be defined and > 0.')
            attention_mask = attention_mask.view(batch_size, -1)
            # 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[:, None, None, :]

            # 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 the dtype's smallest value for masked positions.
            # Since we are adding it to the raw scores before the softmax, this is
            # effectively the same as removing these entirely.
            attention_mask = attention_mask.to(dtype=self.dtype)  # fp16 compatibility
            attention_mask = (1.0 - attention_mask) * torch.finfo(self.dtype).min

        if inputs_embeds is None:
            inputs_embeds = self.wte(input_ids)
        position_embeds = self.wpe(position_ids)
        hidden_states = inputs_embeds + position_embeds

        hidden_states = self.drop(hidden_states)

        output_shape = (-1,) + input_shape[1:] + (hidden_states.size(-1),)

        if self.gradient_checkpointing and self.training:
            if use_cache:
                # noinspection PyUnresolvedReferences
                logger.warning_once(
                    '`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...'
                )
                use_cache = False

        presents = () if use_cache else None
        all_self_attentions = () if output_attentions else None
        all_hidden_states = () if output_hidden_states else None
        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
            if output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            if self.gradient_checkpointing and self.training:
                outputs = self._gradient_checkpointing_func(
                    block.__call__,
                    hidden_states,
                    None,
                    attention_mask,
                    use_cache,
                    output_attentions,
                )
            else:
                outputs = block(
                    hidden_states,
                    layer_past=layer_past,
                    attention_mask=attention_mask,
                    use_cache=use_cache,
                    output_attentions=output_attentions,
                )

            hidden_states = outputs[0]
            if use_cache is True:
                presents = presents + (outputs[1],)

            if output_attentions:
                all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)

        hidden_states = self.ln_f(hidden_states)

        hidden_states = hidden_states.view(output_shape)
        # Add last hidden state
        if output_hidden_states:
            all_hidden_states = all_hidden_states + (hidden_states,)

        if not return_dict:
            return tuple(
                v
                for v in [hidden_states, presents, all_hidden_states, all_self_attentions]
                if v is not None
            )

        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=presents,
            hidden_states=all_hidden_states,
            attentions=all_self_attentions,
        )


class LingLongForCausalLM(LingLongPreTrainedModel):
    _tied_weights_keys = ['lm_head.weight']

    def __init__(self, config):
        super().__init__(config)
        self.transformer = LingLongModel(config)
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)

        # Initialize weights and apply final processing
        self.post_init()

    def get_output_embeddings(self):
        return self.lm_head

    def set_output_embeddings(self, new_embeddings):
        self.lm_head = new_embeddings

    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, inputs_embeds=None, **kwargs):
        # Omit tokens covered by past_key_values
        if past_key_values:
            past_length = past_key_values[0][0].shape[2]

            # Some generation methods already pass only the last input ID
            if input_ids.shape[1] > past_length:
                remove_prefix_length = past_length
            else:
                # Default to old behavior: keep only final ID
                remove_prefix_length = input_ids.shape[1] - 1

            input_ids = input_ids[:, remove_prefix_length:]

        attention_mask = kwargs.get('attention_mask', None)
        position_ids = kwargs.get('position_ids', None)

        if attention_mask is not None and position_ids is None:
            # create position_ids on the fly for batch generation
            # noinspection PyUnresolvedReferences
            position_ids = attention_mask.long().cumsum(-1) - 1
            position_ids.masked_fill_(attention_mask == 0, 1)
            if past_key_values:
                position_ids = position_ids[:, -input_ids.shape[1]:]
        else:
            position_ids = None

        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
        if inputs_embeds is not None and past_key_values is None:
            model_inputs = {'inputs_embeds': inputs_embeds}
        else:
            model_inputs = {'input_ids': input_ids}

        model_inputs.update(
            {
                'past_key_values': past_key_values,
                'use_cache': kwargs.get('use_cache'),
                'position_ids': position_ids,
                'attention_mask': attention_mask,
            }
        )

        return model_inputs

    def forward(
            self,
            input_ids: torch.LongTensor | None = None,
            past_key_values: tuple[tuple[torch.Tensor]] | None = None,
            attention_mask: torch.FloatTensor | None = None,
            position_ids: torch.LongTensor | None = None,
            inputs_embeds: torch.FloatTensor | None = None,
            labels: torch.LongTensor | None = None,
            use_cache: bool | None = None,
            output_attentions: bool | None = None,
            output_hidden_states: bool | None = None,
            return_dict: bool | None = None,
    ) -> tuple | CausalLMOutput:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        transformer_outputs = self.transformer(
            input_ids,
            past_key_values=past_key_values,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        hidden_states = transformer_outputs[0]

        lm_logits = self.lm_head(hidden_states)

        loss = None
        if labels is not None:
            # move labels to correct device to enable model parallelism
            labels = labels.to(lm_logits.device)
            # Shift so that tokens < n predict n
            shift_logits = lm_logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = nn.CrossEntropyLoss()
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))

        if not return_dict:
            output = (lm_logits,) + transformer_outputs[1:]
            return ((loss,) + output) if loss is not None else output

        return CausalLMOutput(
            loss=loss,
            logits=lm_logits,
            hidden_states=transformer_outputs.hidden_states,
            attentions=transformer_outputs.attentions,
        )

    @staticmethod
    def _reorder_cache(
            past_key_values: tuple[tuple[torch.Tensor]],
            beam_idx: torch.Tensor,
            **kwargs,
    ) -> tuple[tuple[torch.Tensor]]:
        """
        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
        beam_idx at every generation step.
        """
        # noinspection PyTypeChecker
        return tuple(
            tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
            for layer_past in past_key_values
        )