Chess Challenge submission by velmen

README.md

@@ -0,0 +1,26 @@
+---
+library_name: transformers
+tags:
+- chess
+- llm-course
+- chess-challenge
+license: mit
+---
+
+# velmen-chess-model_v3
+
+Chess model submitted to the LLM Course Chess Challenge.
+
+## Submission Info
+
+- **Submitted by**: [velmen](https://huggingface.co/velmen)
+- **Parameters**: 992,216
+- **Organization**: LLM-course
+
+## Model Details
+
+- **Architecture**: Chess Transformer (GPT-style)
+- **Vocab size**: 79
+- **Embedding dim**: 128
+- **Layers**: 6
+- **Heads**: 8

config.json

@@ -0,0 +1,20 @@
+{
+  "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": 128,
+  "n_head": 8,
+  "n_inner": 356,
+  "n_layer": 6,
+  "pad_token_id": 0,
+  "tie_weights": true,
+  "transformers_version": "4.57.6",
+  "vocab_size": 79
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c9c4e154d997da71131adf2b797d7e24f4f3d923a23f963dc38c1c34b9cebe9
+size 3975312

special_tokens_map.json

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

tokenizer.py

@@ -0,0 +1,272 @@
+"""
+Custom Chess Tokenizer V3 for the Chess Challenge.
+
+Enhanced version with additional chess-specific tokens for:
+- Castling moves (O-O, O-O-O)
+- Check/checkmate indicators (+, #)
+- Capture indicator (x)
+- Turn indicators ([WHITE], [BLACK])
+
+This provides richer context while keeping vocabulary minimal (81 tokens total).
+"""
+
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path
+from typing import Dict, List, Optional
+import re
+
+from transformers import PreTrainedTokenizer
+
+
+class ChessTokenizer(PreTrainedTokenizer):
+    """
+    Enhanced chess tokenizer with special chess notation tokens.
+    
+    Vocabulary (79 tokens):
+    - 4 special tokens: [PAD], [BOS], [EOS], [UNK]
+    - 64 square tokens: a1-h8
+    - 4 promotion tokens: q, r, b, n
+    - 2 castling tokens: O-O, O-O-O
+    - 3 modifier tokens: +, #, x (check, checkmate, capture)
+    - 2 turn tokens: [WHITE], [BLACK]
+    """
+    
+    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]"
+    WHITE_TOKEN = "[WHITE]"
+    BLACK_TOKEN = "[BLACK]"
+    
+    def __init__(
+        self,
+        vocab_file: Optional[str] = None,
+        vocab: Optional[Dict[str, int]] = None,
+        **kwargs,
+    ):
+        self._pad_token = self.PAD_TOKEN
+        self._bos_token = self.BOS_TOKEN
+        self._eos_token = self.EOS_TOKEN
+        self._unk_token = self.UNK_TOKEN
+
+        kwargs.pop("pad_token", None)
+        kwargs.pop("bos_token", None)
+        kwargs.pop("eos_token", None)
+        kwargs.pop("unk_token", None)
+        
+        # Enhanced regex pattern for chess notation
+        # Matches: squares, promotions, castling, modifiers, turn indicators
+        self.token_pattern = re.compile(
+            r'O-O-O|O-O|'  # Castling (match O-O-O first!)
+            r'\[WHITE\]|\[BLACK\]|'  # Turn indicators
+            r'[a-h][1-8]|'  # Squares
+            r'[qrbn]|'  # Promotions
+            r'[+#x]'  # Check, checkmate, capture
+        )
+
+        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]:
+        """
+        Create the complete vocabulary with all chess-specific tokens.
+        
+        Total: 79 tokens
+        """
+        vocab = {}
+        idx = 0
+        
+        # Special tokens (0-3)
+        for token in [self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN]:
+            vocab[token] = idx
+            idx += 1
+        
+        # Squares (4-67)
+        for f in 'abcdefgh':
+            for r in '12345678':
+                vocab[f"{f}{r}"] = idx
+                idx += 1
+        
+        # Promotions (68-71)
+        for p in ['q', 'r', 'b', 'n']:
+            vocab[p] = idx
+            idx += 1
+        
+        # Castling (72-73)
+        vocab["O-O"] = idx
+        idx += 1
+        vocab["O-O-O"] = idx
+        idx += 1
+        
+        # Modifiers (74-76)
+        vocab["+"] = idx  # Check
+        idx += 1
+        vocab["#"] = idx  # Checkmate
+        idx += 1
+        vocab["x"] = idx  # Capture
+        idx += 1
+        
+        # Turn indicators (77-78)
+        vocab[self.WHITE_TOKEN] = idx
+        idx += 1
+        vocab[self.BLACK_TOKEN] = idx
+        idx += 1
+        
+        return vocab
+    
+    def _tokenize(self, text: str) -> List[str]:
+        """
+        Enhanced tokenization with preprocessing for common chess notation variants.
+        
+        Handles:
+        - Lichess format: (Q) → q, (x) → x, (+) → +, (#) → #
+        - Standard notation: keeps O-O, O-O-O, +, #, x as-is
+        - Extracts squares, promotions, castling, and modifiers
+        """
+        # Normalize Lichess-style parentheses notation
+        text = (text.replace("(Q)", "q")
+                    .replace("(R)", "r")
+                    .replace("(B)", "b")
+                    .replace("(N)", "n")
+                    .replace("(x)", "x")
+                    .replace("(+)", "+")
+                    .replace("(#)", "#")
+                    .replace("(+*)", "#")  # Checkmate variant
+                    .replace("(o)", "O-O")  # Kingside castling
+                    .replace("(O)", "O-O-O"))  # Queenside castling
+        
+        # Extract all chess tokens
+        return self.token_pattern.findall(text)
+    
+    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:
+        """
+        Reconstructs chess moves in standard UCI format with modifiers.
+        
+        Intelligently groups tokens:
+        - Combines squares into moves: e2, e4 → e2e4
+        - Attaches promotions: a7, a8, q → a7a8q
+        - Keeps modifiers separate: e2e4, x, + → e2e4x+
+        - Preserves castling and turn indicators
+        """
+        special = {self.PAD_TOKEN, self.BOS_TOKEN, self.EOS_TOKEN, self.UNK_TOKEN}
+        clean_tokens = [t for t in tokens if t not in special]
+        
+        output = []
+        modifiers = {'+', '#', 'x'}
+        promotions = {'q', 'r', 'b', 'n'}
+        
+        for token in clean_tokens:
+            # Castling and turn indicators stay as-is
+            if token in ["O-O", "O-O-O", self.WHITE_TOKEN, self.BLACK_TOKEN]:
+                output.append(token)
+            # Promotions attach to previous move
+            elif token in promotions and output and len(output[-1]) == 4:
+                output[-1] += token
+            # Modifiers can attach or stay separate (flexible)
+            elif token in modifiers and output:
+                output[-1] += token
+            # Square: either start new move or complete previous
+            elif len(token) == 2 and token[0] in 'abcdefgh':
+                if output and len(output[-1]) == 2 and output[-1][0] in 'abcdefgh':
+                    # Complete the move
+                    output[-1] += token
+                else:
+                    # Start new move
+                    output.append(token)
+            else:
+                output.append(token)
+        
+        return " ".join(output)
+    
+    def add_turn_indicators(self, text: str, add_white_indicator: bool = True) -> str:
+        """
+        Add turn indicators to help the model understand whose turn it is.
+        
+        Args:
+            text: Game string (space-separated moves)
+            add_white_indicator: If True, add [WHITE] at start (white moves first)
+        
+        Returns:
+            Game string with turn indicators
+        """
+        moves = text.strip().split()
+        result = []
+        
+        # White starts (by convention)
+        is_white = add_white_indicator
+        
+        for move in moves:
+            turn_token = self.WHITE_TOKEN if is_white else self.BLACK_TOKEN
+            result.append(turn_token)
+            result.append(move)
+            is_white = not is_white
+        
+        return " ".join(result)
+    
+    def save_vocabulary(
+        self,
+        save_directory: str,
+        filename_prefix: Optional[str] = None,
+    ) -> tuple:
+        """Save the vocabulary to a JSON 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,)
+    
+    @classmethod
+    def build_vocab_from_iterator(cls, iterator, min_frequency=1):
+        """Returns tokenizer with fixed vocabulary (doesn't depend on data)."""
+        return cls()
+    
+    @classmethod
+    def build_vocab_from_dataset(cls, **kwargs):
+        """Returns tokenizer with fixed vocabulary (doesn't depend on data)."""
+        return cls()
+    
+    @property
+    def vocab_size(self) -> int:
+        """Return the size of the vocabulary (79 tokens)."""
+        return len(self._vocab)
+    
+    def get_vocab(self) -> Dict[str, int]:
+        """Return the vocabulary as a dictionary."""
+        return dict(self._vocab)

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