File size: 5,677 Bytes
70830d6
 
 
 
d2c1e33
70830d6
 
 
683d749
 
 
 
 
 
 
 
 
70830d6
 
 
 
 
 
 
 
683d749
 
 
 
70830d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
683d749
 
 
 
70830d6
 
 
 
 
d2c1e33
70830d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
683d749
 
 
 
70830d6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d2c1e33
 
70830d6
 
 
 
683d749
 
 
70830d6
 
 
 
 
 
 
 
 
 
 
 
 
 
683d749
 
 
70830d6
 
 
 
 
 
 
 
683d749
 
 
 
70830d6
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
from arena.llm import LLM
from arena.board import pieces, cols
import json
import random
import logging


class Player:
    """
    This class represents one AI player in the game, and is responsible for managing the prompts
    Delegating to an LLM instance to connect to the LLM
    """

    def __init__(self, model: str, color: int):
        """
        Set up this instance for the given model and player color
        """
        self.color = color
        self.model = model
        self.llm = LLM.create(self.model)
        self.evaluation = ""
        self.threats = ""
        self.opportunities = ""
        self.strategy = ""

    def system(self, board, legal_moves: str, illegal_moves: str) -> str:
        """
        Return the system prompt for this move
        """
        return f"""You are playing the board game Connect 4.
Players take turns to drop counters into one of 7 columns A, B, C, D, E, F, G.
The winner is the first player to get 4 counters in a row in any direction.
You are {pieces[self.color]} and your opponent is {pieces[self.color * -1]}.
You must pick a column for your move. You must pick one of the following legal moves: {legal_moves}.
You should respond in JSON according to this spec:

{{
    "evaluation": "my assessment of the board",
    "threats": "any threats from my opponent that I should block",
    "opportunities": "my best chances to win",
    "strategy": "my thought process",
    "move_column": "one letter from this list of legal moves: {legal_moves}"
}}

You must pick one of these letters for your move_column: {legal_moves}{illegal_moves}"""

    def user(self, board, legal_moves: str, illegal_moves: str) -> str:
        """
        Return the user prompt for this move
        """
        return f"""It is your turn to make a move as {pieces[self.color]}.
Here is the current board, with row 1 at the bottom of the board:

{board.json()}

Here's another way of looking at the board visually, where R represents a red counter, Y for a yellow counter, and _ represents an empty square.

{board.alternative()}

Your final response should be only in JSON strictly according to this spec:

{{
    "evaluation": "my assessment of the board",
    "threats": "any threats from my opponent that I should block",
    "opportunities": "my best chances to win",
    "strategy": "my thought process",
    "move_column": "one of {legal_moves} which are the legal moves"
}}

For example, the following could be a response:

{{
    "evaluation": "the board is equally balanced but I have a slight advantage",
    "threats": "my opponent has a threat but I can block it",
    "opportunities": "I've developed several promising 3 in a row opportunities",
    "strategy": "I must first block my opponent, then I can continue to develop",
    "move_column": "{random.choice(board.legal_moves())}"
}}

And this is another example of a well formed response:

{{
    "evaluation": "although my opponent has more threats, I can win immediately",
    "threats": "my opponent has several threats",
    "opportunities": "I can immediately win the game by making a diagonal 4",
    "strategy": "I will take the winning move",
    "move_column": "{random.choice(board.legal_moves())}"
}}


Now make your decision.
You must pick one of these letters for your move_column: {legal_moves}{illegal_moves}
"""

    def process_move(self, reply: str, board):
        """
        Interpret the reply and make the move; if the move is illegal, then the current player loses
        """
        try:
            if len(reply) == 3 and reply[0] == "{" and reply[2] == "}":
                reply = f'{{"move_column": "{reply[1]}"}}'
            result = json.loads(reply)
            move = result.get("move_column") or "missing"
            move = move.upper()
            col = cols.find(move)
            if not (0 <= col <= 6) or board.height(col) == 6:
                raise ValueError("Illegal move")
            board.move(col)
            self.evaluation = result.get("evaluation") or ""
            self.threats = result.get("threats") or ""
            self.opportunities = result.get("opportunities") or ""
            self.strategy = result.get("strategy") or ""
        except Exception as e:
            logging.error(f"Exception {e}")
            logging.exception(e)
            board.forfeit = True
            board.winner = -1 * board.player

    def move(self, board):
        """
        Have the underlying LLM make a move, and process the result
        """
        legal_moves = ", ".join(board.legal_moves())
        if illegal := board.illegal_moves():
            illegal_moves = (
                "\nYou must NOT make any of these moves which are ILLEGAL: "
                + ", ".join(illegal)
            )
        else:
            illegal_moves = ""
        system = self.system(board, legal_moves, illegal_moves)
        user = self.user(board, legal_moves, illegal_moves)
        reply = self.llm.send(system, user)
        self.process_move(reply, board)

    def thoughts(self):
        """
        Return HTML to describe the inner thoughts
        """
        result = '<div style="text-align: left;font-size:14px"><br/>'
        result += f"<b>Evaluation:</b><br/>{self.evaluation}<br/><br/>"
        result += f"<b>Threats:</b><br/>{self.threats}<br/><br/>"
        result += f"<b>Opportunities:</b><br/>{self.opportunities}<br/><br/>"
        result += f"<b>Strategy:</b><br/>{self.strategy}"
        result += "</div>"
        return result

    def switch_model(self, new_model_name: str):
        """
        Change the underlying LLM to the new model
        """
        self.llm = LLM.create(new_model_name)