""" Custom Chess Tokenizer for the Chess1MChallenge. Goal: maximize legal-move rate in the evaluation. Key idea: - The evaluator only needs to recover the UCI move (e.g. "e2e4") from the model output. It extracts squares like [a-h][1-8] and builds a move from the first 2 squares. - So we normalize dataset tokens like "WPe2e4(x+)" to plain UCI "e2e4" (plus promotion suffix "q/r/b/n"). - We use a FIXED UCI vocabulary so there is (almost) no OOV -> far fewer [UNK] -> higher legal-move rate. Vocabulary: - All from-to square pairs: "a1a2", ..., excluding from==to. - All promotion moves: e7e8[qrbn], a2a1[qrbn], including capture-promotions (still covered by from-to). """ from __future__ import annotations import json import os import re from typing import Dict, List, Optional from transformers import PreTrainedTokenizer _SQUARE_RE = re.compile(r"[a-h][1-8]") _PROMO_RE = re.compile(r"=([QRBN])") # dataset often uses "=Q" class ChessTokenizer(PreTrainedTokenizer): """ Tokenizer that maps each chess move to a single token. It is compatible with Hugging Face `AutoTokenizer(..., trust_remote_code=True)`. Notes: - Input text may contain "extended UCI" tokens from the Lichess dataset (e.g. "WPe2e4", "BKe8g8(O)", "WPe7e8=Q(+)" ...). - We normalize those tokens to plain UCI: "e2e4", "e8g8", "e7e8q", ... """ 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 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 # Avoid duplicate special-token args 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: self._vocab = self._create_default_vocab() self._ids_to_tokens = {v: k for k, v in self._vocab.items()} 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_default_vocab(self) -> Dict[str, int]: special = [self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN] return {tok: i for i, tok in enumerate(special)} @staticmethod def _normalize_one_token(tok: str) -> str: """ Convert an extended token to plain UCI. Examples: "WPe2e4" -> "e2e4" "BKe8g8(O)" -> "e8g8" "WPe7e8=Q(+)" -> "e7e8q" "WPe5d6(x)" -> "e5d6" """ squares = _SQUARE_RE.findall(tok) if len(squares) < 2: return ChessTokenizer.UNK_TOKEN uci = squares[0] + squares[1] m = _PROMO_RE.search(tok) if m: uci += m.group(1).lower() # Q->q etc. return uci @classmethod def build_fixed_uci_vocab(cls) -> "ChessTokenizer": """ Build a FIXED vocabulary of (almost) all possible UCI moves. This dramatically reduces OOV compared to building vocab from the dataset with a high min_frequency. """ files = "abcdefgh" ranks = "12345678" tokens: List[str] = [] # All from-to square pairs (excluding from==to) for f1 in files: for r1 in ranks: for f2 in files: for r2 in ranks: if f1 == f2 and r1 == r2: continue tokens.append(f"{f1}{r1}{f2}{r2}") # Promotions: white (7->8) and black (2->1), with q/r/b/n promos = "qrbn" # White promotions for f in files: fr = f + "7" for df in (-1, 0, 1): j = files.index(f) + df if 0 <= j < 8: to = files[j] + "8" base = fr + to for p in promos: tokens.append(base + p) # Black promotions for f in files: fr = f + "2" for df in (-1, 0, 1): j = files.index(f) + df if 0 <= j < 8: to = files[j] + "1" base = fr + to for p in promos: tokens.append(base + p) tokens = sorted(set(tokens)) special = [cls.PAD_TOKEN, cls.BOS_TOKEN, cls.EOS_TOKEN, cls.UNK_TOKEN] vocab = {tok: i for i, tok in enumerate(special + tokens)} return cls(vocab=vocab) @classmethod def build_vocab_from_iterator(cls, iterator, min_frequency: int = 1) -> "ChessTokenizer": """ Optional: build vocabulary from an iterator of strings. We normalize tokens to UCI before counting. """ from collections import Counter counts = Counter() for game in iterator: raw = game.strip().split() norm = [cls._normalize_one_token(t) for t in raw] counts.update(norm) tokens = [t for t, c in counts.items() if c >= min_frequency] tokens = sorted(set(tokens)) special = [cls.PAD_TOKEN, cls.BOS_TOKEN, cls.EOS_TOKEN, cls.UNK_TOKEN] vocab = {tok: i for i, tok in enumerate(special + tokens)} return cls(vocab=vocab) @property def vocab_size(self) -> int: return len(self._vocab) def get_vocab(self) -> Dict[str, int]: return dict(self._vocab) def _tokenize(self, text: str) -> List[str]: raw = text.strip().split() return [self._normalize_one_token(t) for t in raw] def _convert_token_to_id(self, token: str) -> int: return self._vocab.get(token, self._vocab.get(self.UNK_TOKEN, 0)) def _convert_id_to_token(self, index: int) -> str: 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: 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,)