Chess Challenge submission by clmrie

README.md

@@ -0,0 +1,26 @@
+---
+library_name: transformers
+tags:
+- chess
+- llm-course
+- chess-challenge
+license: mit
+---
+
+# chess-clmrie
+
+Chess model submitted to the LLM Course Chess Challenge.
+
+## Submission Info
+
+- Submitted by: clmrie
+- Parameters: 991,168
+- Organization: LLM-course
+
+## Model Details
+
+- Architecture: chess_transformer (custom)
+- Vocab size: 85
+- Embedding dim: 136
+- Layers: 5
+- Heads: 8

config.json

@@ -0,0 +1,26 @@
+{
+  "architectures": [
+    "ChessForCausalLM"
+  ],
+  "bos_token_id": 1,
+  "dropout": 0.1,
+  "dtype": "float32",
+  "eos_token_id": 2,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "chess_transformer",
+  "n_ctx": 256,
+  "n_embd": 136,
+  "n_head": 8,
+  "n_inner": 408,
+  "n_layer": 5,
+  "pad_token_id": 0,
+  "tie_weights": false,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.57.6",
+  "vocab_size": 85,
+  "auto_map": {
+    "AutoConfig": "model.ChessConfig",
+    "AutoModelForCausalLM": "model.ChessForCausalLM"
+  },
+  "unk_token_id": 3
+}

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "pad_token_id": 0,
+  "transformers_version": "4.57.6"
+}

model.py

@@ -0,0 +1,502 @@
+from __future__ import annotations
+
+import json
+import math
+import os
+from dataclasses import dataclass
+from typing import Dict, List, Optional, Tuple
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from transformers import PretrainedConfig, PreTrainedModel
+from transformers.modeling_outputs import CausalLMOutputWithPast
+from transformers.utils.hub import cached_file
+
+
+def _is_square(tok: str) -> bool:
+    return len(tok) == 2 and tok[0] in "abcdefgh" and tok[1] in "12345678"
+
+
+def _resolve_file(name_or_path: str, filename: str) -> str:
+    if isinstance(name_or_path, str) and os.path.isdir(name_or_path):
+        p = os.path.join(name_or_path, filename)
+        if os.path.exists(p):
+            return p
+    return cached_file(name_or_path, filename)
+
+
+def _load_vocab(name_or_path: str) -> Tuple[Dict[str, int], Dict[int, str]]:
+    vocab_path = _resolve_file(name_or_path, "vocab.json")
+    with open(vocab_path, "r", encoding="utf-8") as f:
+        tok2id = json.load(f)
+    id2tok = {int(i): t for t, i in tok2id.items()}
+    return tok2id, id2tok
+
+
+@dataclass
+class TokenScheme:
+    W: str
+    B: str
+    pieces: Dict[str, str]
+    sep: Optional[str]
+    suffix: Dict[str, str]
+    prom: Dict[str, str]
+    pad_id: int
+    bos_id: int
+    eos_id: int
+    unk_id: int
+
+
+def _detect_scheme(tok2id: Dict[str, int], config) -> TokenScheme:
+    W = "W" if "W" in tok2id else None
+    B = "B" if "B" in tok2id else None
+    if W is None or B is None:
+        raise ValueError("Cannot find W/B tokens in vocab")
+
+    pieces = {}
+    for p in ["P", "N", "B", "R", "Q", "K"]:
+        if p in tok2id:
+            pieces[p] = p
+        else:
+            raise ValueError(f"Cannot find piece token {p} in vocab")
+
+    sep = " " if " " in tok2id else None
+
+    suffix = {}
+    for k, v in [
+        ("cap", "(x)"),
+        ("cap_check", "(x*)"),
+        ("cap_mate", "(x+*)"),
+        ("check", "(+)"),
+        ("mate", "(+*)"),
+        ("o", "(o)"),
+        ("O", "(O)"),
+    ]:
+        if v in tok2id:
+            suffix[k] = v
+
+    prom = {}
+    for p, v in [("Q", "(Q)"), ("R", "(R)"), ("B", "(B)"), ("N", "(N)")]:
+        if v in tok2id:
+            prom[p] = v
+
+    pad_id = int(getattr(config, "pad_token_id", 0))
+    bos_id = int(getattr(config, "bos_token_id", 1))
+    eos_id = int(getattr(config, "eos_token_id", 2))
+    unk_id = int(getattr(config, "unk_token_id", 3))
+
+    return TokenScheme(W=W, B=B, pieces=pieces, sep=sep, suffix=suffix, prom=prom,
+                       pad_id=pad_id, bos_id=bos_id, eos_id=eos_id, unk_id=unk_id)
+
+
+class ChessConfig(PretrainedConfig):
+    model_type = "chess_transformer"
+
+    def __init__(
+        self,
+        vocab_size: int = 85,
+        n_embd: int = 128,
+        n_layer: int = 5,
+        n_head: int = 4,
+        n_ctx: int = 256,
+        n_inner: Optional[int] = None,
+        dropout: float = 0.1,
+        layer_norm_epsilon: float = 1e-5,
+        tie_weights: bool = False,
+        pad_token_id: int = 0,
+        bos_token_id: int = 1,
+        eos_token_id: int = 2,
+        unk_token_id: int = 3,
+        **kwargs,
+    ):
+        self.vocab_size = int(vocab_size)
+        self.n_embd = int(n_embd)
+        self.n_layer = int(n_layer)
+        self.n_head = int(n_head)
+        self.n_ctx = int(n_ctx)
+        self.n_inner = int(n_inner) if n_inner is not None else 3 * int(n_embd)
+        self.dropout = float(dropout)
+        self.layer_norm_epsilon = float(layer_norm_epsilon)
+        self.tie_weights = bool(tie_weights)
+
+        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 MLP(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: torch.Tensor) -> torch.Tensor:
+        x = self.c_fc(x)
+        x = F.gelu(x)
+        x = self.c_proj(x)
+        x = self.dropout(x)
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(self, config: ChessConfig):
+        super().__init__()
+        assert config.n_embd % config.n_head == 0
+        self.n_head = config.n_head
+        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)
+
+        bias = torch.tril(torch.ones(config.n_ctx, config.n_ctx)).view(1, 1, config.n_ctx, config.n_ctx)
+        self.register_buffer("bias", bias, persistent=False)
+
+    def forward(self, x: torch.Tensor, attention_mask: Optional[torch.Tensor] = None) -> torch.Tensor:
+        B, T, C = x.size()
+        qkv = self.c_attn(x)
+        q, k, v = qkv.split(C, 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)) / math.sqrt(self.head_dim)
+        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)
+        y = self.c_proj(y)
+        y = self.dropout(y)
+        return y
+
+
+class Block(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 = MLP(config)
+
+    def forward(self, x: torch.Tensor, attention_mask: Optional[torch.Tensor] = None) -> torch.Tensor:
+        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 = ""
+
+    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([Block(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 getattr(config, "tie_weights", False):
+            self.lm_head.weight = self.wte.weight
+
+        self.post_init()
+
+        self._tok2id = None
+        self._id2tok = None
+        self._scheme = None
+
+    def _ensure_vocab(self):
+        if self._tok2id is None or self._id2tok is None:
+            name_or_path = getattr(self.config, "_name_or_path", None) or getattr(self, "name_or_path", None)
+            if not name_or_path:
+                raise ValueError("Cannot resolve model path to load vocab.json")
+            self._tok2id, self._id2tok = _load_vocab(name_or_path)
+
+    def _get_scheme(self) -> TokenScheme:
+        if self._scheme is None:
+            self._ensure_vocab()
+            self._scheme = _detect_scheme(self._tok2id, self.config)
+        return self._scheme
+
+    def forward(self, input_ids, attention_mask=None, labels=None, return_dict=True, **kwargs):
+        B, T = input_ids.shape
+        if T > self.config.n_ctx:
+            input_ids = input_ids[:, -self.config.n_ctx :]
+            if attention_mask is not None:
+                attention_mask = attention_mask[:, -self.config.n_ctx :]
+            if labels is not None:
+                labels = labels[:, -self.config.n_ctx :]
+            B, T = input_ids.shape
+
+        pos = torch.arange(0, T, device=input_ids.device).unsqueeze(0)
+        x = self.wte(input_ids) + self.wpe(pos)
+        x = self.drop(x)
+
+        for block in self.h:
+            x = block(x, attention_mask=attention_mask)
+
+        x = self.ln_f(x)
+        logits = self.lm_head(x)
+
+        loss = None
+        if labels is not None:
+            shift_logits = logits[:, :-1].contiguous()
+            shift_labels = labels[:, 1:].contiguous()
+            loss = F.cross_entropy(
+                shift_logits.view(-1, shift_logits.size(-1)),
+                shift_labels.view(-1),
+                ignore_index=-100,
+            )
+
+        if not return_dict:
+            return (logits, loss)
+        return CausalLMOutputWithPast(logits=logits, loss=loss)
+
+    def _ids_to_tokens(self, ids: List[int]) -> List[str]:
+        self._ensure_vocab()
+        return [self._id2tok.get(int(i), "[UNK]") for i in ids]
+
+    def _parse_history_to_board(self, input_ids_1d: List[int]):
+        import chess
+        scheme = self._get_scheme()
+        toks = self._ids_to_tokens(input_ids_1d)
+
+        specials = {"[PAD]", "[BOS]", "[EOS]", "[UNK]"}
+        toks = [t for t in toks if t not in specials]
+
+        b = chess.Board()
+        i = 0
+        while i < len(toks):
+            while i < len(toks) and toks[i] not in (scheme.W, scheme.B):
+                i += 1
+            if i >= len(toks):
+                break
+
+            i += 1
+
+            while i < len(toks) and scheme.sep is not None and toks[i] == scheme.sep:
+                i += 1
+
+            if i >= len(toks) or toks[i] not in scheme.pieces.values():
+                break
+            i += 1
+
+            while i < len(toks) and scheme.sep is not None and toks[i] == scheme.sep:
+                i += 1
+
+            if i >= len(toks) or not _is_square(toks[i]):
+                break
+            src = toks[i]
+            i += 1
+
+            while i < len(toks) and scheme.sep is not None and toks[i] == scheme.sep:
+                i += 1
+
+            if i >= len(toks) or not _is_square(toks[i]):
+                break
+            dst = toks[i]
+            i += 1
+
+            suffixes = []
+            while i < len(toks) and toks[i] not in (scheme.W, scheme.B):
+                if scheme.sep is not None and toks[i] == scheme.sep:
+                    i += 1
+                    continue
+                suffixes.append(toks[i])
+                i += 1
+
+            uci = f"{src}{dst}"
+            promo = None
+            for p, ptok in scheme.prom.items():
+                if ptok in suffixes:
+                    promo = p.lower()
+                    break
+            if promo is not None:
+                uci += promo
+
+            try:
+                mv = chess.Move.from_uci(uci)
+                if mv in b.legal_moves:
+                    b.push(mv)
+                else:
+                    break
+            except Exception:
+                break
+
+        return b
+
+    def _move_to_ids(self, board, move_uci: str) -> List[int]:
+        import chess
+
+        scheme = self._get_scheme()
+        self._ensure_vocab()
+        tok2id = self._tok2id
+
+        mv = chess.Move.from_uci(move_uci)
+
+        color_tok = scheme.W if board.turn == chess.WHITE else scheme.B
+        piece = board.piece_at(mv.from_square)
+        pl = piece.symbol().upper() if piece is not None else "P"
+        if pl not in scheme.pieces:
+            pl = "P"
+
+        src = chess.square_name(mv.from_square)
+        dst = chess.square_name(mv.to_square)
+
+        toks = [color_tok, pl]
+        if scheme.sep is not None:
+            toks += [scheme.sep, src, scheme.sep, dst]
+        else:
+            toks += [src, dst]
+
+        is_capture = board.is_capture(mv)
+        board.push(mv)
+        is_mate = board.is_checkmate()
+        is_check = board.is_check()
+        board.pop()
+
+        suffix_tok = None
+        if is_capture and is_mate:
+            suffix_tok = scheme.suffix.get("cap_mate")
+        elif is_capture and is_check:
+            suffix_tok = scheme.suffix.get("cap_check")
+        elif is_capture:
+            suffix_tok = scheme.suffix.get("cap")
+        elif is_mate:
+            suffix_tok = scheme.suffix.get("mate")
+        elif is_check:
+            suffix_tok = scheme.suffix.get("check")
+
+        if suffix_tok is not None:
+            toks.append(suffix_tok)
+
+        if mv.promotion is not None:
+            prom = chess.piece_symbol(mv.promotion).upper()
+            if prom in scheme.prom:
+                toks.append(scheme.prom[prom])
+
+        if scheme.sep is not None:
+            toks.append(scheme.sep)
+
+        return [tok2id.get(t, scheme.unk_id) for t in toks]
+
+    @torch.no_grad()
+    def _score_candidates(self, prefix_ids, cand_ids_list, attention_mask, temperature, batch_size=64):
+        device = prefix_ids.device
+        T0 = prefix_ids.size(1)
+        scores = torch.empty(len(cand_ids_list), device=device, dtype=torch.float32)
+        pad_id = int(self.config.pad_token_id)
+
+        for start in range(0, len(cand_ids_list), batch_size):
+            batch = cand_ids_list[start : start + batch_size]
+            max_c = max(len(c) for c in batch)
+
+            input_ids_list = []
+            attn_list = []
+
+            for c in batch:
+                c_ids = torch.tensor(c, device=device, dtype=torch.long).unsqueeze(0)
+                seq = torch.cat([prefix_ids, c_ids], dim=1)
+                pad_len = (T0 + max_c) - seq.size(1)
+                if pad_len > 0:
+                    pad = torch.full((1, pad_len), pad_id, device=device, dtype=torch.long)
+                    seq = torch.cat([seq, pad], dim=1)
+                input_ids_list.append(seq)
+
+                if attention_mask is None:
+                    a = torch.ones((1, seq.size(1)), device=device, dtype=torch.long)
+                else:
+                    a = attention_mask
+                    if a.size(1) != T0:
+                        a = a[:, -T0:]
+                    ones = torch.ones((1, len(c)), device=device, dtype=torch.long)
+                    zeros = torch.zeros((1, max_c - len(c)), device=device, dtype=torch.long)
+                    a = torch.cat([a, ones, zeros], dim=1)
+                attn_list.append(a)
+
+            input_ids = torch.cat(input_ids_list, dim=0)
+            attn_mask = torch.cat(attn_list, dim=0)
+
+            out = self.forward(input_ids=input_ids, attention_mask=attn_mask, return_dict=True)
+            logits = out.logits / float(max(1e-6, temperature))
+            logp = torch.log_softmax(logits, dim=-1)
+
+            for bi, c in enumerate(batch):
+                lp = 0.0
+                for j in range(len(c)):
+                    pos = T0 + j - 1
+                    if pos < 0:
+                        continue
+                    tok_id = int(c[j])
+                    lp += float(logp[bi, pos, tok_id].item())
+                scores[start + bi] = lp
+
+        return scores
+
+    def generate(self, input_ids=None, attention_mask=None, max_new_tokens=16, temperature=1.0, do_sample=False, **kwargs):
+        import chess
+
+        if input_ids is None:
+            raise ValueError("generate() requires input_ids")
+        if input_ids.dim() == 1:
+            input_ids = input_ids.unsqueeze(0)
+
+        if input_ids.size(0) != 1:
+            return super().generate(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                max_new_tokens=max_new_tokens,
+                temperature=temperature,
+                do_sample=do_sample,
+                **kwargs,
+            )
+
+        try:
+            board = self._parse_history_to_board(input_ids[0].tolist())
+        except Exception:
+            board = None
+
+        if board is None or board.is_game_over():
+            return super().generate(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                max_new_tokens=max_new_tokens,
+                temperature=temperature,
+                do_sample=do_sample,
+                **kwargs,
+            )
+
+        legal = list(board.legal_moves)
+        if not legal:
+            return input_ids
+
+        cand_ids_list = [self._move_to_ids(board, mv.uci()) for mv in legal]
+
+        scores = self._score_candidates(
+            prefix_ids=input_ids,
+            cand_ids_list=cand_ids_list,
+            attention_mask=attention_mask,
+            temperature=float(temperature),
+            batch_size=64,
+        )
+
+        best = int(torch.argmax(scores).item())
+        best_ids = torch.tensor(cand_ids_list[best], device=input_ids.device, dtype=torch.long).unsqueeze(0)
+
+        if best_ids.size(1) > int(max_new_tokens):
+            best_ids = best_ids[:, : int(max_new_tokens)]
+
+        return torch.cat([input_ids, best_ids], dim=1)

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b66b88123670200735099e5106816a20fc5f05d2aae87bc01ee747ac7f1f2fc
+size 3970192

special_tokens_map.json

@@ -0,0 +1,6 @@
+{
+  "bos_token": "[BOS]",
+  "eos_token": "[EOS]",
+  "pad_token": "[PAD]",
+  "unk_token": "[UNK]"
+}

tokenizer.py

@@ -0,0 +1,190 @@
+"""
+Custom Chess Tokenizer for the Chess Challenge.
+
+This tokenizer treats each move as a single token using the extended UCI notation
+from the Lichess dataset (e.g., WPe2e4, BNg8f6).
+
+The dataset format uses:
+- W/B prefix for White/Black
+- Piece letter: P=Pawn, N=Knight, B=Bishop, R=Rook, Q=Queen, K=King
+- Source and destination squares (e.g., e2e4)
+- Special suffixes: (x)=capture, (+)=check, (+*)=checkmate, (o)/(O)=castling
+"""
+
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path
+from typing import Dict, List, Optional
+
+from transformers import PreTrainedTokenizer
+
+
+class ChessTokenizer(PreTrainedTokenizer):
+    """
+    A custom tokenizer for chess moves using extended UCI notation.
+    
+
+    """
+    
+    model_input_names = ["input_ids", "attention_mask"]
+    vocab_files_names = {"vocab_file": "vocab.json"}
+    
+    # Special tokens
+    PAD_TOKEN = "[PAD]"
+    BOS_TOKEN = "[BOS]"
+    EOS_TOKEN = "[EOS]"
+    UNK_TOKEN = "[UNK]"
+    
+    def __init__(
+        self,
+        vocab_file: Optional[str] = None,
+        vocab: Optional[Dict[str, int]] = None,
+        **kwargs,
+    ):
+        """
+        Initialize the chess tokenizer.
+        
+        Args:
+            vocab_file: Path to a JSON file containing the vocabulary mapping.
+            vocab: Dictionary mapping tokens to IDs (alternative to vocab_file).
+            **kwargs: Additional arguments passed to PreTrainedTokenizer.
+        """
+        # Initialize special tokens
+        self._pad_token = self.PAD_TOKEN
+        self._bos_token = self.BOS_TOKEN
+        self._eos_token = self.EOS_TOKEN
+        self._unk_token = self.UNK_TOKEN
+
+        # Remove any duplicate special-token entries passed through kwargs
+        # to avoid "multiple values for keyword" errors when loading from disk.
+        kwargs.pop("pad_token", None)
+        kwargs.pop("bos_token", None)
+        kwargs.pop("eos_token", None)
+        kwargs.pop("unk_token", None)
+        
+        # Load or create vocabulary
+        if vocab is not None:
+            self._vocab = vocab
+        elif vocab_file is not None and os.path.exists(vocab_file):
+            with open(vocab_file, "r", encoding="utf-8") as f:
+                self._vocab = json.load(f)
+        else:
+            # Create a minimal vocabulary with just special tokens
+            # The full vocabulary should be built from the dataset
+            self._vocab = self._create_fixed_vocab()
+        
+        # Create reverse mapping
+        self._ids_to_tokens = {v: k for k, v in self._vocab.items()}
+        
+        # Call parent init AFTER setting up vocab
+        super().__init__(
+            pad_token=self._pad_token,
+            bos_token=self._bos_token,
+            eos_token=self._eos_token,
+            unk_token=self._unk_token,
+            **kwargs,
+        )
+    
+    def _create_fixed_vocab(self) -> Dict[str, int]:
+
+        vocab_list = [self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN]
+
+        # Separate moves
+        vocab_list.append(" ")
+        
+        # Colors
+        vocab_list.extend(["W", "B"])
+        
+        # Pieces
+        vocab_list.extend(["P", "N", "B", "R", "Q", "K"])
+        
+        # 3. Squares
+        files = "abcdefgh"
+        ranks = "12345678"
+        squares = [f"{f}{r}" for f in files for r in ranks]
+        vocab_list.extend(sorted(squares))
+
+        # Suffixes
+        suffixes = ["(x)", "(+)", "(+*)", "(o)", "(O)", "(Q)", "(K)", "(x*)", "(x+*)"]
+        vocab_list.extend(suffixes)
+
+        vocab_list = list(dict.fromkeys(vocab_list))
+        
+        
+        return {token: idx for idx, token in enumerate(vocab_list)}
+    
+    
+    @property
+    def vocab_size(self) -> int:
+        """Return the size of the vocabulary."""
+        return len(self._vocab)
+    
+    def get_vocab(self) -> Dict[str, int]:
+        """Return the vocabulary as a dictionary."""
+        return dict(self._vocab)
+    
+    def _tokenize(self, text: str) -> List[str]:
+        """
+        Tokenize a string of moves into atomic components.
+        """
+        import re
+        tokens = []
+        moves = text.strip().split()
+        
+        pattern = re.compile(r"^([WB])([PNBRQK])([a-h][1-8])([a-h][1-8])(.*)$")
+        
+        for move in moves:
+            match = pattern.match(move)
+            if match:
+                for i in range(1,6):
+                    if  match.group(i) in self._vocab:
+                        tokens.append(match.group(i))
+                tokens.append(' ')
+            else:
+                tokens.append(self.UNK_TOKEN)
+                
+        return tokens
+    
+    def _convert_token_to_id(self, token: str) -> int:
+        """Convert a token to its ID."""
+        return self._vocab.get(token, self._vocab.get(self.UNK_TOKEN, 0))
+    
+    def _convert_id_to_token(self, index: int) -> str:
+        """Convert an ID to its token."""
+        return self._ids_to_tokens.get(index, self.UNK_TOKEN)
+    
+    def convert_tokens_to_string(self, tokens: List[str]) -> str:
+        special = {self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN}
+        return "".join(t for t in tokens if t not in special)
+    
+    def save_vocabulary(
+        self,
+        save_directory: str,
+        filename_prefix: Optional[str] = None,
+    ) -> tuple:
+        """
+        Save the vocabulary to a JSON file.
+        
+        Args:
+            save_directory: Directory to save the vocabulary.
+            filename_prefix: Optional prefix for the filename.
+        
+        Returns:
+            Tuple containing the path to the saved vocabulary file.
+        """
+        if not os.path.isdir(save_directory):
+            os.makedirs(save_directory, exist_ok=True)
+        
+        vocab_file = os.path.join(
+            save_directory,
+            (filename_prefix + "-" if filename_prefix else "") + "vocab.json",
+        )
+        
+        with open(vocab_file, "w", encoding="utf-8") as f:
+            json.dump(self._vocab, f, ensure_ascii=False, indent=2)
+        
+        return (vocab_file,)
+
+

tokenizer_config.json

@@ -0,0 +1,50 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "[BOS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "[EOS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenizer.ChessTokenizer",
+      null
+    ]
+  },
+  "bos_token": "[BOS]",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "[EOS]",
+  "extra_special_tokens": {},
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "[PAD]",
+  "tokenizer_class": "ChessTokenizer",
+  "unk_token": "[UNK]"
+}

vocab.json

@@ -0,0 +1,87 @@
+{
+  "[PAD]": 0,
+  "[BOS]": 1,
+  "[EOS]": 2,
+  "[UNK]": 3,
+  " ": 4,
+  "W": 5,
+  "B": 6,
+  "P": 7,
+  "N": 8,
+  "R": 9,
+  "Q": 10,
+  "K": 11,
+  "a1": 12,
+  "a2": 13,
+  "a3": 14,
+  "a4": 15,
+  "a5": 16,
+  "a6": 17,
+  "a7": 18,
+  "a8": 19,
+  "b1": 20,
+  "b2": 21,
+  "b3": 22,
+  "b4": 23,
+  "b5": 24,
+  "b6": 25,
+  "b7": 26,
+  "b8": 27,
+  "c1": 28,
+  "c2": 29,
+  "c3": 30,
+  "c4": 31,
+  "c5": 32,
+  "c6": 33,
+  "c7": 34,
+  "c8": 35,
+  "d1": 36,
+  "d2": 37,
+  "d3": 38,
+  "d4": 39,
+  "d5": 40,
+  "d6": 41,
+  "d7": 42,
+  "d8": 43,
+  "e1": 44,
+  "e2": 45,
+  "e3": 46,
+  "e4": 47,
+  "e5": 48,
+  "e6": 49,
+  "e7": 50,
+  "e8": 51,
+  "f1": 52,
+  "f2": 53,
+  "f3": 54,
+  "f4": 55,
+  "f5": 56,
+  "f6": 57,
+  "f7": 58,
+  "f8": 59,
+  "g1": 60,
+  "g2": 61,
+  "g3": 62,
+  "g4": 63,
+  "g5": 64,
+  "g6": 65,
+  "g7": 66,
+  "g8": 67,
+  "h1": 68,
+  "h2": 69,
+  "h3": 70,
+  "h4": 71,
+  "h5": 72,
+  "h6": 73,
+  "h7": 74,
+  "h8": 75,
+  "(x)": 76,
+  "(+)": 77,
+  "(+*)": 78,
+  "(o)": 79,
+  "(O)": 80,
+  "(Q)": 81,
+  "(K)": 82,
+  "(x*)": 83,
+  "(x+*)": 84
+}