modeling_plamo.py

from typing import Any, Dict, List, NamedTuple, Optional, Tuple, Union

import numpy as np
import torch
from torch import nn
from torch.nn import functional as F
from transformers import PretrainedConfig, PreTrainedModel
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast


class DecoderInput(NamedTuple):
    hidden_states: torch.Tensor
    position_ids: torch.Tensor
    attention_mask: Optional[torch.Tensor] = None
    past_key_values: Optional[List[torch.FloatTensor]] = None
    output_hidden_states: Optional[bool] = False
    output_attentions: Optional[bool] = False
    use_cache: Optional[bool] = False
    gradient_checkpointing: bool = False


class DecoderOutput(NamedTuple):
    hidden_states: torch.Tensor
    all_hidden_states: Optional[Tuple[torch.Tensor, ...]]
    all_self_attns: Optional[Tuple[torch.Tensor, ...]]
    next_decoder_cache: Optional[Tuple[torch.Tensor, ...]]


class PlamoConfig(PretrainedConfig):  # type: ignore
    model_type: str = "plamo"

    def __init__(
        self,
        vocab_size: int = 32000,
        hidden_size: int = 4096,
        intermediate_size: int = 13312,
        num_hidden_layers: int = 32,
        num_attention_heads: int = 32,
        num_key_value_heads: Optional[int] = None,
        max_position_embeddings: int = 2048,
        initializer_range: float = 0.02,
        rms_norm_eps: float = 1e-6,
        use_cache: bool = True,
        tokenizer_class: str = "PlamoTokenizer",
        pad_token_id: Optional[int] = None,
        bos_token_id: int = 1,
        eos_token_id: int = 2,
        n_shared_head: int = 8,
        tie_word_embeddings: bool = False,
        **kwargs: Any,
    ) -> None:
        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads

        # for backward compatibility
        if num_key_value_heads is None:
            num_key_value_heads = num_attention_heads

        self.num_key_value_heads = num_key_value_heads
        self.initializer_range = initializer_range
        self.rms_norm_eps = rms_norm_eps
        self.use_cache = use_cache

        self.n_shared_head = n_shared_head

        super().__init__(
            tokenizer_class=tokenizer_class,
            pad_token_id=pad_token_id,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )


# Copied from transformers.models.bart.modeling_bart._make_causal_mask
def _make_causal_mask(
    input_ids_shape: Tuple[int, int], dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
) -> torch.Tensor:
    """
    Make causal mask used for bi-directional self-attention.
    """
    bsz, tgt_len = input_ids_shape
    mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
    mask_cond = torch.arange(mask.size(-1), device=device)
    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
    mask = mask.to(dtype)

    if past_key_values_length > 0:
        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)


# Copied from transformers.models.bart.modeling_bart._expand_mask
def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None) -> torch.Tensor:
    """
    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
    """
    bsz, src_len = mask.size()
    tgt_len = tgt_len if tgt_len is not None else src_len

    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)

    inverted_mask = 1.0 - expanded_mask

    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)  # type: ignore


class RotaryEmbedding(torch.nn.Module):
    def __init__(
        self, dim: int, max_position_embeddings: int = 2048, base: int = 10000, device: Optional[torch.device] = None
    ) -> None:
        super().__init__()

        self.dim = dim
        self.max_position_embeddings = max_position_embeddings
        self.base = base
        inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim))
        self.register_buffer("inv_freq", inv_freq, persistent=False)

        # Build here to make `torch.jit.trace` work.
        self._set_cos_sin_cache(
            seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype()
        )

    def _set_cos_sin_cache(self, seq_len: int, device: Any, dtype: Any) -> None:
        self.max_seq_len_cached = seq_len
        t = torch.arange(self.max_seq_len_cached, device=device, dtype=self.inv_freq.dtype)  # type: ignore

        freqs = torch.einsum("i,j->ij", t, self.inv_freq)
        # Different from paper, but it uses a different permutation in order to obtain the same calculation
        emb = torch.cat((freqs, freqs), dim=-1)
        self.register_buffer("cos_cached", emb.cos()[None, None, :, :].to(dtype), persistent=False)
        self.register_buffer("sin_cached", emb.sin()[None, None, :, :].to(dtype), persistent=False)

    def forward(self, x: torch.Tensor, seq_len: int) -> Tuple[torch.Tensor, torch.Tensor]:
        # x: [bs, num_attention_heads, seq_len, head_size]
        if seq_len > self.max_seq_len_cached:
            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype)

        return (
            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),  # type: ignore
            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),  # type: ignore
        )


def _rotate_half(x: torch.Tensor) -> torch.Tensor:
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)


def _rotary_pos_emb(x: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor, position_ids: torch.Tensor) -> torch.Tensor:
    # The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
    cos = cos.squeeze(1).squeeze(0)  # [seq_len, dim]
    sin = sin.squeeze(1).squeeze(0)  # [seq_len, dim]
    cos = cos[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    sin = sin[position_ids].unsqueeze(1)  # [bs, 1, seq_len, dim]
    x_embed = (x * cos) + (_rotate_half(x) * sin)
    return x_embed


class RMSNorm(nn.Module):
    def __init__(self, hidden_size: int, eps: float = 1e-6) -> None:
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
        return self.weight * hidden_states.to(input_dtype)


class Attention(torch.nn.Module):
    def __init__(self, config: PlamoConfig) -> None:
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        head_dim = self.hidden_size // config.num_attention_heads
        self.max_position_embeddings = config.max_position_embeddings

        self.q_num_heads = config.num_attention_heads
        self.qk_dim = self.v_dim = head_dim
        self.k_num_heads = self.v_num_heads = int(np.ceil(self.q_num_heads / config.n_shared_head))

        self.q_proj = nn.Linear(self.hidden_size, self.q_num_heads * self.qk_dim, bias=False)
        self.k_proj = nn.Linear(self.hidden_size, self.k_num_heads * self.qk_dim, bias=False)
        self.v_proj = nn.Linear(self.hidden_size, self.v_num_heads * self.v_dim, bias=False)
        self.o_proj = nn.Linear(self.q_num_heads * self.v_dim, self.hidden_size, bias=False)
        self.rotary_emb = RotaryEmbedding(self.qk_dim, max_position_embeddings=self.max_position_embeddings)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
        output_attentions: bool = False,
        use_cache: bool = False,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor, torch.Tensor]]]:
        bsz, q_len, _ = hidden_states.size()

        query_states = self.q_proj(hidden_states).view(bsz, q_len, self.q_num_heads, self.qk_dim).transpose(1, 2)
        key_states = self.k_proj(hidden_states).view(bsz, q_len, self.k_num_heads, self.qk_dim).transpose(1, 2)
        value_states = self.v_proj(hidden_states).view(bsz, q_len, self.v_num_heads, self.v_dim).transpose(1, 2)

        def _expand_kv(t: torch.Tensor, repeat: int, target: int) -> torch.Tensor:
            return t.repeat(1, repeat, 1, 1)[:, :target]

        # expand shared kv
        assert self.k_num_heads == self.v_num_heads
        key_states = _expand_kv(key_states, self.config.n_shared_head, self.q_num_heads)
        value_states = _expand_kv(value_states, self.config.n_shared_head, self.q_num_heads)

        kv_seq_len = key_states.shape[-2]
        if past_key_value is not None:
            kv_seq_len += past_key_value[0].shape[-2]
        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
        assert position_ids is not None
        query_states = _rotary_pos_emb(query_states, cos, sin, position_ids)
        key_states = _rotary_pos_emb(key_states, cos, sin, position_ids)
        # [bsz, nh, t, hd]

        if past_key_value is not None:
            # reuse k, v, self_attention
            key_states = torch.cat([past_key_value[0], key_states], dim=2)
            value_states = torch.cat([past_key_value[1], value_states], dim=2)

        past_key_value = (key_states, value_states) if use_cache else None

        attn_output = F.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask=attention_mask)
        attn_output = attn_output.transpose(1, 2)

        attn_output = attn_output.reshape(bsz, q_len, self.q_num_heads * self.v_dim)
        attn_output = self.o_proj(attn_output)

        if not output_attentions:
            attn_weights = None

        return attn_output, attn_weights, past_key_value


class MLP(nn.Module):
    def __init__(self, config: PlamoConfig) -> None:
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
        self.act_fn = torch.nn.functional.silu

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))  # type: ignore


class PlamoDecoderLayer(torch.nn.Module):
    def __init__(self, config: PlamoConfig) -> None:
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.self_attn = Attention(config)
        self.mlp = MLP(config)
        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
    ) -> Tuple[Any, ...]:
        # from LlamaDecoder
        residual = hidden_states

        hidden_states = self.norm(hidden_states)

        # Self Attention
        hidden_states_sa, self_attn_weights, present_key_value = self.self_attn(
            hidden_states=hidden_states,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_value=past_key_value,
            output_attentions=output_attentions,
            use_cache=use_cache,
        )

        # Fully Connected
        hidden_states_mlp = self.mlp(hidden_states)

        # Residual
        hidden_states = residual + hidden_states_sa + hidden_states_mlp

        outputs: Any = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)

        return outputs  # type: ignore


class PlamoDecoder(torch.nn.Module):
    def __init__(self, config: PlamoConfig) -> None:
        super().__init__()
        self.layers = torch.nn.ModuleList([PlamoDecoderLayer(config) for _ in range(config.num_hidden_layers)])

    def forward(self, x: DecoderInput) -> DecoderOutput:
        all_hidden_states: Optional[Tuple[torch.Tensor, ...]] = () if x.output_hidden_states else None
        all_self_attns: Optional[Tuple[torch.Tensor, ...]] = () if x.output_attentions else None
        next_decoder_cache: Optional[Tuple[torch.Tensor, ...]] = () if x.use_cache else None
        hidden_states = x.hidden_states

        for idx, decoder_layer in enumerate(self.layers):
            if x.output_hidden_states:
                assert all_hidden_states is not None
                all_hidden_states += (hidden_states,)

            past_key_value = x.past_key_values[idx] if x.past_key_values is not None else None

            if self.training and x.gradient_checkpointing:

                def create_custom_forward(module):  # type: ignore
                    def custom_forward(*inputs):  # type: ignore
                        # None for past_key_value
                        return module(*inputs, x.output_attentions, None)

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(decoder_layer),  # type: ignore
                    hidden_states,
                    x.attention_mask,
                    x.position_ids,
                    None,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=x.attention_mask,
                    position_ids=x.position_ids,
                    past_key_value=past_key_value,
                    output_attentions=x.output_attentions,
                    use_cache=x.use_cache,
                )

            hidden_states = layer_outputs[0]

            if x.use_cache:
                cache = layer_outputs[2 if x.output_attentions else 1]
                assert cache is not None
                assert next_decoder_cache is not None
                next_decoder_cache += (cache,)

            if x.output_attentions:
                assert layer_outputs[1] is not None
                assert all_self_attns is not None
                all_self_attns += (layer_outputs[1],)
        return DecoderOutput(hidden_states, all_hidden_states, all_self_attns, next_decoder_cache)


class PlamoPreTrainedModel(PreTrainedModel):  # type: ignore
    config_class = PlamoConfig
    _no_split_modules: List[str]
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["PlamoDecoderLayer"]
    _skip_keys_device_placement = "past_key_values"
    _keys_to_ignore_on_load_unexpected = [r"decoder\.version"]

    def _init_weights(self, module: torch.nn.Module) -> None:
        std = self.config.initializer_range
        if isinstance(module, nn.Linear):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.bias is not None:
                module.bias.data.zero_()
        elif isinstance(module, nn.Embedding):
            module.weight.data.normal_(mean=0.0, std=std)
            if module.padding_idx is not None:
                module.weight.data[module.padding_idx].zero_()

    def _set_gradient_checkpointing(self, module: torch.nn.Module, value: bool = False) -> None:
        module.gradient_checkpointing = value  # type: ignore


class PlamoModel(PlamoPreTrainedModel):
    def __init__(self, config: PlamoConfig):
        super().__init__(config)
        self.padding_idx = config.pad_token_id
        self.vocab_size = config.vocab_size

        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
        self.layers = PlamoDecoder(config)  # type: ignore
        self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)

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

    def get_input_embeddings(self) -> torch.nn.Embedding:
        return self.embed_tokens

    def set_input_embeddings(self, value: torch.nn.Embedding) -> None:
        self.embed_tokens = value

    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
    def _prepare_decoder_attention_mask(
        self,
        attention_mask: torch.Tensor,
        input_shape: Tuple[int, int],
        inputs_embeds: Optional[torch.FloatTensor],
        past_key_values_length: int,
    ) -> Optional[torch.Tensor]:
        # create causal mask
        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
        combined_attention_mask: Optional[torch.Tensor] = None
        if input_shape[-1] > 1:
            assert inputs_embeds is not None
            combined_attention_mask = _make_causal_mask(
                input_shape,
                inputs_embeds.dtype,
                device=inputs_embeds.device,
                past_key_values_length=past_key_values_length,
            )

        if attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            assert inputs_embeds is not None
            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
                inputs_embeds.device
            )
            combined_attention_mask = (
                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
            )

        return combined_attention_mask

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, BaseModelOutputWithPast]:
        assert input_ids is not None
        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

        # retrieve input_ids and inputs_embeds
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
        elif input_ids is not None:
            batch_size, seq_length = input_ids.shape
        else:
            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")

        seq_length_with_past = seq_length
        past_key_values_length = 0

        if past_key_values is not None:
            past_key_values_length = past_key_values[0][0].shape[2]
            seq_length_with_past = seq_length_with_past + past_key_values_length

        if position_ids is None:
            device = input_ids.device
            position_ids = torch.arange(
                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
            )
            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
        else:
            position_ids = position_ids.view(-1, seq_length).long()

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids)
        # embed positions
        if attention_mask is None:
            attention_mask = torch.ones(
                (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
            )
        attention_mask = self._prepare_decoder_attention_mask(
            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
        )

        hidden_states = inputs_embeds

        if self.gradient_checkpointing and self.training:
            if use_cache:
                use_cache = False

        # decoder layers
        out = self.layers(
            DecoderInput(
                hidden_states,
                position_ids,
                attention_mask,
                past_key_values,
                output_hidden_states,
                output_attentions,
                use_cache,
                self.gradient_checkpointing,
            )
        )
        assert isinstance(out, DecoderOutput)
        hidden_states = out.hidden_states
        all_hidden_states = out.all_hidden_states
        all_self_attns = out.all_self_attns
        next_decoder_cache = out.next_decoder_cache

        hidden_states = self.norm(hidden_states)

        # add hidden states from the last decoder layer
        if output_hidden_states:
            assert all_hidden_states is not None
            all_hidden_states += (hidden_states,)

        next_cache = next_decoder_cache if use_cache else None
        if not return_dict:
            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=next_cache,
            hidden_states=all_hidden_states,
            attentions=all_self_attns,
        )


class PlamoForCausalLM(PlamoPreTrainedModel):
    def __init__(self, config: PretrainedConfig) -> None:
        super().__init__(config)
        self.model = PlamoModel(config)

        self.lm_head: torch.nn.Module = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

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

    def get_input_embeddings(self) -> torch.nn.Embedding:
        return self.model.embed_tokens

    def set_input_embeddings(self, value: torch.nn.Embedding) -> None:
        self.model.embed_tokens = value

    def get_output_embeddings(self) -> torch.nn.Module:
        return self.lm_head

    def set_output_embeddings(self, new_embeddings: torch.nn.Module) -> None:
        self.lm_head = new_embeddings

    def set_decoder(self, decoder: PlamoModel) -> None:
        self.model = decoder

    def get_decoder(self) -> PlamoModel:
        return self.model

    def forward(  # type: ignore
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, CausalLMOutputWithPast]:
        r"""
        Args:
            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.

        Returns:

        Example:

        ```python
        >>> from transformers import AutoTokenizer, LlamaForCausalLM

        >>> model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)

        >>> prompt = "Hey, are you consciours? Can you talk to me?"
        >>> inputs = tokenizer(prompt, return_tensors="pt")

        >>> # Generate
        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
        "Hey, are you consciours? Can you talk to me?\nI'm not consciours, but I can talk to you."
        ```"""
        assert input_ids is not None

        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
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
        outputs = self.model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            past_key_values=past_key_values,
            inputs_embeds=inputs_embeds,
            use_cache=use_cache,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        hidden_states = outputs[0]
        logits = self.lm_head(hidden_states)

        loss = None
        if labels is not None:
            # Shift so that tokens < n predict n
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            # Flatten the tokens
            loss_fct = nn.CrossEntropyLoss()
            shift_logits = shift_logits.view(-1, self.config.vocab_size)
            shift_labels = shift_labels.view(-1)
            # Enable model parallelism
            shift_labels = shift_labels.to(shift_logits.device)
            loss = loss_fct(shift_logits, shift_labels)

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

        return CausalLMOutputWithPast(
            loss=loss,
            logits=logits,
            past_key_values=outputs.past_key_values,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

    def prepare_inputs_for_generation(
        self,
        input_ids: torch.Tensor,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        attention_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        **kwargs: Any,
    ) -> Dict[str, Any]:
        if past_key_values:
            input_ids = input_ids[:, -1:]

        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
            position_ids = attention_mask.long().cumsum(-1) - 1
            position_ids.masked_fill_(attention_mask == 0, 1)
            if past_key_values:
                position_ids = position_ids[:, -1].unsqueeze(-1)

        # 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: Dict[str, Any] = {"inputs_embeds": inputs_embeds}
        else:
            model_inputs = {"input_ids": input_ids}

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

    @staticmethod
    def _reorder_cache(past_key_values: List[torch.FloatTensor], beam_idx: int) -> Tuple[Any, ...]:
        reordered_past: Tuple[Any, ...] = ()
        for layer_past in past_key_values:
            reordered_past += (tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),)
        return reordered_past