model.py

import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import PretrainedConfig, PreTrainedModel
from transformers.modeling_outputs import CausalLMOutputWithPast

def rotate_half(x):
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)

def apply_rope(q, k):
    dim = q.shape[-1]
    device = q.device
    seq_len = q.shape[-2]
    theta = 1.0 / (10000 ** (torch.arange(0, dim, 2, device=device).float() / dim))
    pos = torch.arange(seq_len, device=device).float()
    freqs = torch.einsum('i,j->ij', pos, theta)
    emb = torch.cat((freqs, freqs), dim=-1)
    cos = emb.cos()[None, None, :, :]
    sin = emb.sin()[None, None, :, :]
    q = (q * cos) + (rotate_half(q) * sin)
    k = (k * cos) + (rotate_half(k) * sin)
    return q, k

class ChessConfig(PretrainedConfig):
    model_type = "chess_transformer"
    def __init__(self, vocab_size=1682, n_embd=96, n_layer=8, n_head=8, n_ctx=256, n_inner=288, dropout=0.15, layer_norm_epsilon=1e-5, tie_weights=True, pad_token_id=0, bos_token_id=1, eos_token_id=2, **kwargs):
        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **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
        self.dropout = dropout
        self.layer_norm_epsilon = layer_norm_epsilon
        self.tie_weights = tie_weights
        self.tie_word_embeddings = True

class MultiHeadAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        assert config.n_embd % config.n_head == 0
        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):
        batch_size, seq_len, _ = x.size()
        qkv = self.c_attn(x)
        q, k, v = qkv.split(self.n_embd, dim=2)
        q = q.view(batch_size, seq_len, self.n_head, self.head_dim).transpose(1, 2)
        k = k.view(batch_size, seq_len, self.n_head, self.head_dim).transpose(1, 2)
        v = v.view(batch_size, seq_len, self.n_head, self.head_dim).transpose(1, 2)
        q, k = apply_rope(q, k)
        attn_weights = torch.matmul(q, k.transpose(-2, -1)) / math.sqrt(self.head_dim)
        causal_mask = self.bias[:, :, :seq_len, :seq_len]
        attn_weights = attn_weights.masked_fill(causal_mask == 0, float("-inf"))
        if attention_mask is not None:
            attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
            attn_weights = attn_weights.masked_fill(attention_mask == 0, float("-inf"))
        attn_weights = F.softmax(attn_weights, dim=-1)
        attn_weights = self.dropout(attn_weights)
        attn_output = torch.matmul(attn_weights, v)
        attn_output = attn_output.transpose(1, 2).contiguous().view(batch_size, seq_len, self.n_embd)
        attn_output = self.c_proj(attn_output)
        return attn_output

class FeedForward(nn.Module):
    def __init__(self, config):
        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):
        x = self.c_fc(x)
        x = F.gelu(x)
        x = self.c_proj(x)
        x = self.dropout(x)
        return x

class TransformerBlock(nn.Module):
    def __init__(self, config):
        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=attention_mask)
        x = x + self.mlp(self.ln_2(x))
        return x

class ChessForCausalLM(PreTrainedModel):
    config_class = ChessConfig
    base_model_prefix = "transformer"
    supports_gradient_checkpointing = True
    keys_to_ignore_on_load_missing = ["lm_head.weight"]
    def __init__(self, config):
        super().__init__(config)
        self.wte = nn.Embedding(config.vocab_size, 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._tied_weights_keys = ["lm_head.weight"]
        self.post_init()
        if config.tie_weights:
            self.tie_weights()
    def get_input_embeddings(self):
        return self.wte
    def set_input_embeddings(self, new_embeddings):
        self.wte = new_embeddings
        if getattr(self.config, "tie_weights", False):
            self.tie_weights()
    def get_output_embeddings(self):
        return self.lm_head
    def set_output_embeddings(self, new_embeddings):
        self.lm_head = new_embeddings
    def tie_weights(self):
        if getattr(self.config, "tie_weights", False) or getattr(self.config, "tie_word_embeddings", False):
            self._tie_or_clone_weights(self.lm_head, self.wte)
    def _init_weights(self, module):
        if isinstance(module, nn.Linear):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
            if module.bias is not None:
                torch.nn.init.zeros_(module.bias)
        elif isinstance(module, nn.Embedding):
            torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
        elif isinstance(module, nn.LayerNorm):
            torch.nn.init.ones_(module.weight)
            torch.nn.init.zeros_(module.bias)
    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
        batch_size, seq_len = input_ids.size()
        token_embeds = self.wte(input_ids)
        hidden_states = self.drop(token_embeds)
        for block in self.h:
            hidden_states = block(hidden_states, attention_mask=attention_mask)
        hidden_states = self.ln_f(hidden_states)
        logits = self.lm_head(hidden_states)
        loss = None
        if labels is not None:
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss_fct = nn.CrossEntropyLoss(ignore_index=-100)
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
        if not return_dict:
            output = (logits,)
            return ((loss,) + output) if loss is not None else output
        return CausalLMOutputWithPast(loss=loss, logits=logits, past_key_values=None, hidden_states=None, attentions=None)