"""
Chess Transformer Model - The "Nuclear Patch" Edition
"""
from __future__ import annotations
import math
from typing import Optional, Tuple, Union
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import PretrainedConfig, PreTrainedModel
from transformers.modeling_outputs import CausalLMOutputWithPast

class ChessConfig(PretrainedConfig):
    model_type = "chess_transformer"
    
    def __init__(
        self,
        vocab_size=1200,
        n_embd=128,
        n_layer=6,
        n_head=4,
        n_ctx=256,
        n_inner=None,
        dropout=0.1,
        layer_norm_epsilon=1e-5,
        tie_weights=True,
        pad_token_id=0,
        bos_token_id=1,
        eos_token_id=2,
        unk_token_id=3,
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.n_embd = n_embd
        self.n_layer = n_layer
        self.n_head = n_head
        self.n_ctx = n_ctx
        self.n_inner = n_inner if n_inner is not None else 3 * n_embd
        self.dropout = dropout
        self.layer_norm_epsilon = layer_norm_epsilon
        self.tie_weights = tie_weights
        
        # On passe les IDs vitaux à kwargs pour le parent
        kwargs["pad_token_id"] = pad_token_id
        kwargs["bos_token_id"] = bos_token_id
        kwargs["eos_token_id"] = eos_token_id
        kwargs["unk_token_id"] = unk_token_id
        
        super().__init__(**kwargs)

class MultiHeadAttention(nn.Module):
    def __init__(self, config: ChessConfig):
        super().__init__()
        self.n_head = config.n_head
        self.n_embd = config.n_embd
        self.head_dim = config.n_embd // config.n_head
        self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd)
        self.c_proj = nn.Linear(config.n_embd, config.n_embd)
        self.dropout = nn.Dropout(config.dropout)
        self.register_buffer("bias", torch.tril(torch.ones(config.n_ctx, config.n_ctx)).view(1, 1, config.n_ctx, config.n_ctx), persistent=False)
    
    def forward(self, x, attention_mask=None):
        B, T, C = x.size()
        qkv = self.c_attn(x)
        q, k, v = qkv.split(self.n_embd, dim=2)
        q = q.view(B, T, self.n_head, self.head_dim).transpose(1, 2)
        k = k.view(B, T, self.n_head, self.head_dim).transpose(1, 2)
        v = v.view(B, T, self.n_head, self.head_dim).transpose(1, 2)
        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
        att = att.masked_fill(self.bias[:,:,:T,:T] == 0, float('-inf'))
        if attention_mask is not None:
            att = att.masked_fill(attention_mask.view(B, 1, 1, T) == 0, float('-inf'))
        att = F.softmax(att, dim=-1)
        att = self.dropout(att)
        y = att @ v
        y = y.transpose(1, 2).contiguous().view(B, T, C)
        return self.c_proj(y)

class FeedForward(nn.Module):
    def __init__(self, config: ChessConfig):
        super().__init__()
        self.c_fc = nn.Linear(config.n_embd, config.n_inner)
        self.c_proj = nn.Linear(config.n_inner, config.n_embd)
        self.dropout = nn.Dropout(config.dropout)
    def forward(self, x):
        return self.dropout(self.c_proj(F.gelu(self.c_fc(x))))

class TransformerBlock(nn.Module):
    def __init__(self, config: ChessConfig):
        super().__init__()
        self.ln_1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
        self.attn = MultiHeadAttention(config)
        self.ln_2 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
        self.mlp = FeedForward(config)
    def forward(self, x, attention_mask=None):
        x = x + self.attn(self.ln_1(x), attention_mask)
        x = x + self.mlp(self.ln_2(x))
        return x

class ChessForCausalLM(PreTrainedModel):
    config_class = ChessConfig
    base_model_prefix = "transformer"
    
    def __init__(self, config: ChessConfig):
        super().__init__(config)
        self.wte = nn.Embedding(config.vocab_size, config.n_embd)
        self.wpe = nn.Embedding(config.n_ctx, config.n_embd)
        self.drop = nn.Dropout(config.dropout)
        self.h = nn.ModuleList([TransformerBlock(config) for _ in range(config.n_layer)])
        self.ln_f = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
        if config.tie_weights: self.post_init()

    def get_input_embeddings(self): return self.wte
    def set_input_embeddings(self, new_embeddings): self.wte = new_embeddings
    def get_output_embeddings(self): return self.lm_head
    def set_output_embeddings(self, new_embeddings): self.lm_head = new_embeddings

    def forward(self, input_ids, attention_mask=None, position_ids=None, labels=None, return_dict=None, **kwargs):
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        if return_dict is None: return_dict = True

        device = input_ids.device
        b, t = input_ids.size()
        if position_ids is None: position_ids = torch.arange(t, device=device).unsqueeze(0)
        
        x = self.wte(input_ids) + self.wpe(position_ids)
        x = self.drop(x)
        for block in self.h: x = block(x, attention_mask)
        x = self.ln_f(x)
        logits = self.lm_head(x)
        if labels is None:

            nuclear_bad_ids = [0, 1, 2, 3] 
            
            logits[:, :, nuclear_bad_ids] = float("-inf")
        loss = None
        if labels is not None:
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss = nn.CrossEntropyLoss(ignore_index=self.config.pad_token_id)(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))

        if not return_dict:
            return ((loss,) + (logits,)) if loss is not None else (logits,)
        
        return CausalLMOutputWithPast(loss=loss, logits=logits)

from transformers import AutoConfig, AutoModelForCausalLM
AutoConfig.register("chess_transformer", ChessConfig)
AutoModelForCausalLM.register(ChessConfig, ChessForCausalLM)