# modeling_theta.py

import torch
import torch.nn as nn
from transformers.modeling_outputs import BaseModelOutputWithPast
from transformers import PreTrainedModel
from configuration_theta import ThetaConfig

class ThetaAttention(nn.Module):
    def __init__(self, config: ThetaConfig):
        super().__init__()
        self.num_heads = config.num_attention_heads
        self.head_dim = config.hidden_size // config.num_attention_heads
        self.scale = self.head_dim ** -0.5

        self.q_proj = nn.Linear(config.hidden_size, config.hidden_size)
        self.k_proj = nn.Linear(config.hidden_size, config.hidden_size)
        self.v_proj = nn.Linear(config.hidden_size, config.hidden_size)
        self.out_proj = nn.Linear(config.hidden_size, config.hidden_size)

    def forward(self, hidden_states):
        batch_size, seq_length, embed_dim = hidden_states.size()
        
        query = self.q_proj(hidden_states)
        key = self.k_proj(hidden_states)
        value = self.v_proj(hidden_states)

        query = query.view(batch_size, seq_length, self.num_heads, self.head_dim).transpose(1, 2)
        key = key.view(batch_size, seq_length, self.num_heads, self.head_dim).transpose(1, 2)
        value = value.view(batch_size, seq_length, self.num_heads, self.head_dim).transpose(1, 2)

        attn_scores = torch.matmul(query, key.transpose(-2, -1)) * self.scale
        attn_probs = nn.functional.softmax(attn_scores, dim=-1)
        attn_output = torch.matmul(attn_probs, value)

        attn_output = attn_output.transpose(1, 2).contiguous().view(batch_size, seq_length, embed_dim)
        attn_output = self.out_proj(attn_output)

        return attn_output

class ThetaMLP(nn.Module):
    def __init__(self, config: ThetaConfig):
        super().__init__()
        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
        self.act = nn.SiLU()
        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)

    def forward(self, hidden_states):
        hidden_states = self.fc1(hidden_states)
        hidden_states = self.act(hidden_states)
        hidden_states = self.fc2(hidden_states)
        return hidden_states

class ThetaBlock(nn.Module):
    def __init__(self, config: ThetaConfig):
        super().__init__()
        self.attention = ThetaAttention(config)
        self.mlp = ThetaMLP(config)
        self.norm1 = nn.LayerNorm(config.hidden_size, eps=config.rms_norm_eps)
        self.norm2 = nn.LayerNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(self, hidden_states):
        hidden_states = hidden_states + self.attention(self.norm1(hidden_states))
        hidden_states = hidden_states + self.mlp(self.norm2(hidden_states))
        return hidden_states

class ThetaModel(PreTrainedModel):
    config_class = ThetaConfig

    def __init__(self, config: ThetaConfig):
        super().__init__(config)
        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
        self.layers = nn.ModuleList([ThetaBlock(config) for _ in range(config.num_hidden_layers)])
        self.norm = nn.LayerNorm(config.hidden_size, eps=config.rms_norm_eps)

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        **kwargs,
    ):
        hidden_states = self.embed_tokens(input_ids)

        for layer in self.layers:
            hidden_states = layer(hidden_states)

        hidden_states = self.norm(hidden_states)

        return BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            hidden_states=None,
            past_key_values=None,
        )

class ThetaForCausalLM(PreTrainedModel):
    config_class = ThetaConfig

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

    def forward(
        self,
        input_ids=None,
        labels=None,
        **kwargs,
    ):
        outputs = self.model(input_ids=input_ids, **kwargs)
        logits = self.lm_head(outputs.last_hidden_state)

        loss = None
        if labels is not None:
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss_fct = nn.CrossEntropyLoss()
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))

        return {"loss": loss, "logits": logits} if loss is not None else {"logits": logits}