modeling_tcmoe.py

# Copyright (c) The HuggingFace Inc. team. All rights reserved.
# Copyright (c) Shen Yan. All rights reserved.
# This code is built upon Huggingface's transformers repository.

import math
from typing import Optional, Tuple, Union
from transformers import PreTrainedModel
import torch
import torch.nn as nn
from transformers.utils import logging
from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
from torch.nn import CrossEntropyLoss
from configuration_tcmoe import TCMoEConfig


logger = logging.get_logger(__name__)


# Copied from transformers.models.bart.modeling_bart._make_causal_mask
def _make_causal_mask(
    input_ids_shape: torch.Size,
    dtype: torch.dtype,
    device: torch.device,
    past_key_values_length: int = 0,
):
    """Make causal mask used for bi-directional self-attention."""
    batch_size, tgt_len = input_ids_shape
    mask = torch.full((tgt_len, tgt_len), torch.finfo(torch.float16).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(batch_size, 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):
    """Expands attention_mask from `[batch_size, seq_len]` to `[batch_size, 1, tgt_seq_len, src_seq_len]`."""
    batch_size, src_len = mask.size()
    tgt_len = tgt_len if tgt_len is not None else src_len

    expanded_mask = mask[:, None, None, :].expand(batch_size, 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
    )


class RotaryEmbedding(nn.Module):
    def __init__(
        self,
        dim: int,
        max_position_embeddings: int,
        base: int = 10_000,
        device: Optional[torch.device] = 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, device=device, dtype=torch.float32) / 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: torch.device, dtype: torch.dtype):
        self.max_seq_len_cached = seq_len
        t = torch.arange(self.max_seq_len_cached, device=device, dtype=torch.float32)

        # Don't do einsum, it converts fp32 to fp16 under AMP
        # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
        freqs = torch.outer(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: Optional[int] = None):
        # x: [batch_size, num_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=torch.get_default_dtype())
        return (
            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
        )


def rotate_half(x: torch.Tensor):
    """Rotates half the hidden dims of the input."""
    x1, x2 = torch.chunk(x, 2, dim=-1)
    return torch.cat((-x2, x1), dim=-1)


def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
    # 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)  # [batch_size, 1, seq_len, dim]
    sin = sin[position_ids].unsqueeze(1)  # [batch_size, 1, seq_len, dim]
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed


class RMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        RMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states):
        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)

    def extra_repr(self):
        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
    

# Copied from transformers.models.llama.modeling_llama.repeat_kv
def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
    """
    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
    """
    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
    if n_rep == 1:
        return hidden_states
    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)


class Attention(nn.Module):
    def __init__(self, config: TCMoEConfig):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.num_heads = config.num_attention_heads
        self.head_dim = self.hidden_size // self.num_heads
        self.num_key_value_heads = config.num_key_value_heads
        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
        self.max_position_embeddings = config.max_position_embeddings

        if (self.head_dim * self.num_heads) != self.hidden_size:
            raise ValueError(
                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
                f" and `num_heads`: {self.num_heads})."
            )
        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
        self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
        self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
        self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=False)
        self.k_norm = RMSNorm(
            (self.hidden_size // self.num_heads), eps=config.norm_eps
        )

        self._init_rope()

    def _init_rope(self):
        self.rotary_ndims = int(self.head_dim * self.config.rope_pct)
        self.rotary_emb = RotaryEmbedding(
            self.rotary_ndims,
            max_position_embeddings=self.config.max_position_embeddings,
            base=self.config.rope_theta,
        )

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

        query_states = self.q_proj(hidden_states)
        key_states = self.k_proj(hidden_states)
        value_states = self.v_proj(hidden_states)

        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)

        query_rot = query_states[..., : self.rotary_ndims]
        query_pass = query_states[..., self.rotary_ndims :]
        key_rot = key_states[..., : self.rotary_ndims]
        key_pass = key_states[..., self.rotary_ndims :]

        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)
        query_states, key_states = apply_rotary_pos_emb(query_rot, key_rot, cos, sin, position_ids)

        key_states = self.k_norm(key_states)

        # [batch_size, num_heads, seq_len, head_dim]
        query_states = torch.cat((query_states, query_pass), dim=-1)
        key_states = torch.cat((key_states, key_pass), dim=-1)

        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

        # Repeat k/v heads if n_kv_heads < n_heads
        key_states = repeat_kv(key_states, self.num_key_value_groups)
        value_states = repeat_kv(value_states, self.num_key_value_groups)

        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)

        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
            raise ValueError(
                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
                f" {attn_weights.size()}"
            )

        if attention_mask is not None:
            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
                raise ValueError(
                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
                )
            attn_weights = attn_weights + attention_mask

        # Upcast attention to fp32
        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
        attn_output = torch.matmul(attn_weights, value_states)

        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
            raise ValueError(
                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
                f" {attn_output.size()}"
            )

        # Merge heads
        attn_output = attn_output.transpose(1, 2).contiguous()
        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)

        # Final linear projection
        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: TCMoEConfig):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.gate_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
        self.up_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
        self.down_proj = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
        self.act_fn = nn.SiLU()

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

class TCMoEBlock(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.num_experts = config.num_experts
        self.top_k = config.moe_topk
        self.num_null_experts = config.num_null_experts
        self.gate = nn.Linear(config.hidden_size, self.num_experts * 2 + self.num_null_experts, bias=False)
        self.experts = nn.ModuleList([MLP(config) for _ in range(self.num_experts)])

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        self.gate.float()
        batch_size, sequence_length, hidden_dim = hidden_states.shape
        hidden_states = hidden_states.view(-1, hidden_dim)
        
        logits = self.gate(hidden_states.float())

        # Add bias to the null experts and negative experts
        logits[:, self.num_experts * 2:] = logits[:, self.num_experts * 2:] - 10.0
        logits[:, self.num_experts:self.num_experts * 2] = logits[:, self.num_experts:self.num_experts * 2] - 1.0

        gates = torch.nn.functional.softmax(logits, dim=1)

        # Select Top-k experts
        weights, selected_experts = torch.topk(gates, k=self.top_k, dim=-1, sorted=False)

        # Calculate the weight sum for the activated non-null experts
        weights_from_non_null_experts = weights * (selected_experts < 2 * self.num_experts).float()
        weightsum_from_non_null_experts = weights_from_non_null_experts.sum(1, keepdim=True)

        # Calculate the weight sum for all null experts (since all null experts are activated)
        weightsum_from_null_experts = gates[:, 2 * self.num_experts:].sum(1, keepdim=True)

        # Normalize the weights of all activated experts
        weightsum = weightsum_from_non_null_experts + weightsum_from_null_experts
        gates = gates / weightsum
        weights = torch.gather(gates, 1, selected_experts)
        weights = weights.to(hidden_states.dtype)

        final_hidden_states = torch.zeros(
            (batch_size * sequence_length, hidden_dim), dtype=hidden_states.dtype, device=hidden_states.device
        )

        # One hot encode the selected experts to create an expert mask
        # this will be used to easily index which expert is going to be selected
        expert_mask = torch.nn.functional.one_hot(selected_experts, num_classes=self.num_experts * 2 + self.num_null_experts).permute(2, 1, 0)

        # Loop over all available experts in the model and perform the computation on each expert
        for expert_idx in range(self.num_experts * 2):
            expert_layer = self.experts[expert_idx if expert_idx < self.num_experts else expert_idx - self.num_experts]
            idx, top_x = torch.where(expert_mask[expert_idx])

            # Index the correct hidden states and compute the expert hidden state for
            # the current expert. We need to make sure to multiply the output hidden
            # states by `routing_weights` on the corresponding tokens (top-1 and top-2)
            current_state = hidden_states[None, top_x].reshape(-1, hidden_dim)
            if expert_idx < self.num_experts:
                current_hidden_states = expert_layer(current_state) * weights[top_x, idx, None]
            else:
                current_hidden_states = expert_layer(current_state) * weights[top_x, idx, None] * -1.0
            current_hidden_states = expert_layer(current_state) * weights[top_x, idx, None]

            # However `index_add_` only support torch tensors for indexing so we'll use
            # the `top_x` tensor here.
            final_hidden_states.index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
        final_hidden_states = final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
        return final_hidden_states


class TCMoEDecoderLayer(nn.Module):
    def __init__(self, config: TCMoEConfig):
        super().__init__()
        self.self_attn = Attention(config)
        self.mlp = TCMoEBlock(config)
        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.norm_eps)
        self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.norm_eps)

    def forward(
        self,
        hidden_states: Optional[torch.FloatTensor],
        attention_mask: Optional[torch.FloatTensor] = 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,
    ) -> Union[Tuple[torch.Tensor], Optional[Tuple[torch.Tensor, Tuple[torch.FloatTensor, ...]]]]:
        residual = hidden_states

        hidden_states = self.input_layernorm(hidden_states)

        # Self Attention
        hidden_states, 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,
        )
        hidden_states = residual + hidden_states

        # Fully Connected
        residual = hidden_states
        hidden_states = self.post_attention_layernorm(hidden_states)
        hidden_states = self.mlp(hidden_states)
        hidden_states = residual + hidden_states

        outputs = (hidden_states,)

        if output_attentions:
            outputs += (self_attn_weights,)

        if use_cache:
            outputs += (present_key_value,)

        return outputs


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

    config_class = TCMoEConfig
    base_model_prefix = "transformer"
    supports_gradient_checkpointing = True
    _no_split_modules = ["TCMoEDecoderLayer"]
    _skip_keys_device_placement = "past_key_values"

    def _init_weights(self, module):
        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_()
        elif isinstance(module, RMSNorm):
            module.weight.data.fill_(1.0)
        elif isinstance(module, nn.LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)


class TCMoEModel(TCMoEPreTrainedModel):
    def __init__(self, config: TCMoEConfig):
        super().__init__(config)
        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, config.pad_token_id)
        self.layers = nn.ModuleList([TCMoEDecoderLayer(config) for _ in range(config.num_hidden_layers)])
        self.norm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps)

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

    def get_input_embeddings(self):
        return self.embed_tokens

    def set_input_embeddings(self, value):
        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: torch.Size,
        inputs_embeds: torch.Tensor,
        past_key_values_length: int,
    ):
        # Create causal mask
        # [batch_size, seq_len] -> [batch_size, 1, tgt_seq_len, src_seq_len]
        combined_attention_mask = None
        if input_shape[-1] > 1:
            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:
            # [batch_size, seq_len] -> [batch_size, 1, tgt_seq_len, src_seq_len]
            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.FloatTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Tuple[Tuple[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]:
        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
        elif inputs_embeds is not None:
            batch_size, seq_length, _ = inputs_embeds.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 if input_ids is not None else inputs_embeds.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:
                logger.warning(
                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
                )
                use_cache = False

        # Decoder layers
        all_hidden_states = () if output_hidden_states else None
        all_self_attns = () if output_attentions else None
        next_decoder_cache = () if use_cache else None

        for idx, decoder_layer in enumerate(self.layers):
            if output_hidden_states:
                all_hidden_states += (hidden_states,)

            past_key_value = (
                past_key_values[idx] if past_key_values is not None else None
            )

            if self.gradient_checkpointing and self.training:

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

                    return custom_forward

                layer_outputs = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(decoder_layer),
                    hidden_states,
                    attention_mask,
                    position_ids,
                )
            else:
                layer_outputs = decoder_layer(
                    hidden_states,
                    attention_mask=attention_mask,
                    position_ids=position_ids,
                    past_key_value=past_key_value,
                    output_attentions=output_attentions,
                    use_cache=use_cache,
                )

            hidden_states = layer_outputs[0]

            if use_cache:
                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)

            if output_attentions:
                all_self_attns += (layer_outputs[1],)

        hidden_states = self.norm(hidden_states)

        # Add hidden states from the last decoder layer
        if output_hidden_states:
            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 TCMoEForCausalLM(PreTrainedModel):
    _tied_weights_keys = ["lm_head.weight"]

    def __init__(self, config):
        super().__init__(config)
        self.model = TCMoEModel(config)
        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)

        # Initialize weights and apply final processing
        self.post_init()
    
    def get_input_embeddings(self):
        return self.model.embed_tokens

    def set_input_embeddings(self, value):
        self.model.embed_tokens = value

    def get_output_embeddings(self):
        return self.lm_head

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

    def get_decoder(self):
        return self.model

    def set_decoder(self, decoder):
        self.model = decoder
    
    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[Tuple[Tuple[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]:
        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,
            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).float()

        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 = 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,
        past_key_values: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        **kwargs,
    ):
        # Trim decoder_input_ids if past is used
        if past_key_values and past_key_values[0] is not None:
            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 = {"inputs_embeds": inputs_embeds}
        else:
            model_inputs = {"input_ids": input_ids}

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

    @staticmethod
    def _reorder_cache(past_key_values, beam_idx):
        reordered_past = ()
        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