Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,11 +3,13 @@ import gradio as gr
|
|
| 3 |
from llama_cpp import Llama
|
| 4 |
from huggingface_hub import hf_hub_download
|
| 5 |
import numpy as np
|
|
|
|
|
|
|
| 6 |
|
| 7 |
model = Llama(
|
| 8 |
model_path=hf_hub_download(
|
| 9 |
-
repo_id=os.environ.get("REPO_ID", "Lyte/QuadConnect2.5-0.5B-v0.0.6b")
|
| 10 |
-
filename=os.environ.get("MODEL_FILE", "quadconnect.Q8_0.gguf")
|
| 11 |
),
|
| 12 |
n_ctx=16384
|
| 13 |
)
|
|
@@ -35,11 +37,95 @@ Specify the column letter (a–g) for your next move.
|
|
| 35 |
</move>
|
| 36 |
"""
|
| 37 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
class ConnectFour:
|
| 39 |
def __init__(self):
|
| 40 |
self.board = np.zeros((6, 7))
|
| 41 |
self.current_player = 1 # 1 for player (X), 2 for AI (O)
|
| 42 |
self.game_over = False
|
|
|
|
|
|
|
| 43 |
|
| 44 |
def make_move(self, col):
|
| 45 |
if self.game_over:
|
|
@@ -47,9 +133,17 @@ class ConnectFour:
|
|
| 47 |
|
| 48 |
# Find the lowest empty row in the selected column
|
| 49 |
for row in range(5, -1, -1):
|
| 50 |
-
if self.board[row][col] == 0:
|
| 51 |
-
self.board[row][col] = self.current_player
|
| 52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
return False, -1
|
| 54 |
|
| 55 |
def check_winner(self):
|
|
@@ -96,7 +190,17 @@ class ConnectFour:
|
|
| 96 |
row_num = str(6 - row) # Convert to 1-based indexing
|
| 97 |
piece = "X" if self.board[row][col] == 1 else "O"
|
| 98 |
moves.append(f"{col_letter}{row_num}({piece})")
|
| 99 |
-
return ", ".join(moves) if moves else ""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
|
| 101 |
def parse_ai_move(self, move_str):
|
| 102 |
# Parse move like 'a', 'b', etc.
|
|
@@ -107,6 +211,25 @@ class ConnectFour:
|
|
| 107 |
return -1
|
| 108 |
except:
|
| 109 |
return -1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 110 |
|
| 111 |
def create_interface():
|
| 112 |
game = ConnectFour()
|
|
@@ -228,15 +351,15 @@ def create_interface():
|
|
| 228 |
|
| 229 |
# AI move
|
| 230 |
game.current_player = 2
|
| 231 |
-
board_state = game.board_to_string()
|
| 232 |
|
| 233 |
-
|
|
|
|
| 234 |
|
| 235 |
# Get AI response
|
| 236 |
response = model.create_chat_completion(
|
| 237 |
messages=[
|
| 238 |
{"role": "system", "content": SYSTEM_PROMPT},
|
| 239 |
-
{"role": "user", "content":
|
| 240 |
],
|
| 241 |
temperature=0.8,
|
| 242 |
top_p=0.95,
|
|
@@ -249,7 +372,11 @@ def create_interface():
|
|
| 249 |
# Extract reasoning and move
|
| 250 |
try:
|
| 251 |
reasoning = ai_response.split("<reasoning>")[1].split("</reasoning>")[0].strip()
|
| 252 |
-
move_str = ai_response
|
|
|
|
|
|
|
|
|
|
|
|
|
| 253 |
ai_col = game.parse_ai_move(move_str)
|
| 254 |
|
| 255 |
if ai_col == -1:
|
|
@@ -298,6 +425,8 @@ def create_interface():
|
|
| 298 |
game.board = np.zeros((6, 7))
|
| 299 |
game.current_player = 1
|
| 300 |
game.game_over = False
|
|
|
|
|
|
|
| 301 |
return [
|
| 302 |
render_board(),
|
| 303 |
"Your turn! Click a button to drop your piece!",
|
|
|
|
| 3 |
from llama_cpp import Llama
|
| 4 |
from huggingface_hub import hf_hub_download
|
| 5 |
import numpy as np
|
| 6 |
+
from typing import List
|
| 7 |
+
import re
|
| 8 |
|
| 9 |
model = Llama(
|
| 10 |
model_path=hf_hub_download(
|
| 11 |
+
repo_id=os.environ.get("REPO_ID", "Lyte/QuadConnect2.5-0.5B-v0.0.6b"),
|
| 12 |
+
filename=os.environ.get("MODEL_FILE", "quadconnect.Q8_0.gguf"),
|
| 13 |
),
|
| 14 |
n_ctx=16384
|
| 15 |
)
|
|
|
|
| 37 |
</move>
|
| 38 |
"""
|
| 39 |
|
| 40 |
+
def extract_xml_move(text: str) -> str:
|
| 41 |
+
"""
|
| 42 |
+
Extracts the move (a single column letter a–g) from the XML format
|
| 43 |
+
using regex.
|
| 44 |
+
"""
|
| 45 |
+
match = re.search(r'<move>\s*([a-g])\s*</move>', text)
|
| 46 |
+
if match:
|
| 47 |
+
return match.group(1)
|
| 48 |
+
return ""
|
| 49 |
+
|
| 50 |
+
def convert_moves_to_coordinate_list(moves_list: List[str]) -> str:
|
| 51 |
+
"""
|
| 52 |
+
Converts a list of moves to a coordinate list representation.
|
| 53 |
+
Each move is formatted as <column><row>(<piece>).
|
| 54 |
+
Returns "Empty Board" if no moves are present.
|
| 55 |
+
"""
|
| 56 |
+
# Create an empty 6x7 grid (row 1 is at index 0)
|
| 57 |
+
grid = [['.' for _ in range(7)] for _ in range(6)]
|
| 58 |
+
|
| 59 |
+
for i, move in enumerate(moves_list):
|
| 60 |
+
if not move:
|
| 61 |
+
continue
|
| 62 |
+
col = ord(move[0]) - ord('a')
|
| 63 |
+
# Find the lowest available row in this column:
|
| 64 |
+
for row in range(6):
|
| 65 |
+
if grid[row][col] == '.':
|
| 66 |
+
grid[row][col] = 'X' if i % 2 == 0 else 'O'
|
| 67 |
+
break
|
| 68 |
+
|
| 69 |
+
# Build coordinate list: Only include cells with a piece
|
| 70 |
+
coords = []
|
| 71 |
+
for row in range(6):
|
| 72 |
+
for col in range(7):
|
| 73 |
+
if grid[row][col] != '.':
|
| 74 |
+
# Convert row index to board row number (row 0 -> 1, etc.)
|
| 75 |
+
coords.append(f"{chr(col + ord('a'))}{row+1}({grid[row][col]})")
|
| 76 |
+
|
| 77 |
+
return ", ".join(coords) if coords else "Empty Board"
|
| 78 |
+
|
| 79 |
+
def parse_coordinate_list(board_str: str) -> List[List[str]]:
|
| 80 |
+
"""
|
| 81 |
+
Converts a coordinate list representation (e.g., "a1(O), a2(X), b1(O)")
|
| 82 |
+
into a 6x7 grid (list of lists) with row index 0 as the bottom.
|
| 83 |
+
"""
|
| 84 |
+
grid = [['.' for _ in range(7)] for _ in range(6)]
|
| 85 |
+
if not board_str.strip() or board_str == "Empty Board":
|
| 86 |
+
return grid
|
| 87 |
+
|
| 88 |
+
coords = board_str.split(",")
|
| 89 |
+
for coord in coords:
|
| 90 |
+
coord = coord.strip()
|
| 91 |
+
# Expecting format: a1(O)
|
| 92 |
+
if len(coord) < 4:
|
| 93 |
+
continue
|
| 94 |
+
col_letter = coord[0]
|
| 95 |
+
try:
|
| 96 |
+
row_number = int(coord[1])
|
| 97 |
+
except ValueError:
|
| 98 |
+
continue
|
| 99 |
+
piece = coord[3] # The piece inside the parentheses
|
| 100 |
+
col = ord(col_letter) - ord('a')
|
| 101 |
+
row = row_number - 1
|
| 102 |
+
if 0 <= row < 6 and 0 <= col < 7:
|
| 103 |
+
grid[row][col] = piece
|
| 104 |
+
return grid
|
| 105 |
+
|
| 106 |
+
def get_available_positions(grid: List[List[str]]) -> str:
|
| 107 |
+
"""Returns the next available position (lowest empty spot) for each column."""
|
| 108 |
+
available = []
|
| 109 |
+
for col in range(7):
|
| 110 |
+
col_letter = chr(ord('a') + col)
|
| 111 |
+
# Search from bottom (row 0) to top (row 5)
|
| 112 |
+
next_pos = None
|
| 113 |
+
for row in range(6):
|
| 114 |
+
if grid[row][col] == '.': # Found empty spot
|
| 115 |
+
next_pos = f"{col_letter}{row + 1}" # +1 because grid is 0-based
|
| 116 |
+
break
|
| 117 |
+
if next_pos: # Only add if column isn't full
|
| 118 |
+
available.append(f"{col_letter}: {next_pos}")
|
| 119 |
+
|
| 120 |
+
return "\n ".join(available) # Format with newlines and indentation
|
| 121 |
+
|
| 122 |
class ConnectFour:
|
| 123 |
def __init__(self):
|
| 124 |
self.board = np.zeros((6, 7))
|
| 125 |
self.current_player = 1 # 1 for player (X), 2 for AI (O)
|
| 126 |
self.game_over = False
|
| 127 |
+
self.player_moves = []
|
| 128 |
+
self.ai_moves = []
|
| 129 |
|
| 130 |
def make_move(self, col):
|
| 131 |
if self.game_over:
|
|
|
|
| 133 |
|
| 134 |
# Find the lowest empty row in the selected column
|
| 135 |
for row in range(5, -1, -1):
|
| 136 |
+
if self.board[5-row][col] == 0: # Flip the row index to match the display
|
| 137 |
+
self.board[5-row][col] = self.current_player
|
| 138 |
+
|
| 139 |
+
# Record the move
|
| 140 |
+
move = f"{chr(ord('a') + col)}{row+1}"
|
| 141 |
+
if self.current_player == 1:
|
| 142 |
+
self.player_moves.append(move)
|
| 143 |
+
else:
|
| 144 |
+
self.ai_moves.append(move)
|
| 145 |
+
|
| 146 |
+
return True, 5-row
|
| 147 |
return False, -1
|
| 148 |
|
| 149 |
def check_winner(self):
|
|
|
|
| 190 |
row_num = str(6 - row) # Convert to 1-based indexing
|
| 191 |
piece = "X" if self.board[row][col] == 1 else "O"
|
| 192 |
moves.append(f"{col_letter}{row_num}({piece})")
|
| 193 |
+
return ", ".join(moves) if moves else "Empty Board"
|
| 194 |
+
|
| 195 |
+
def get_grid_from_board(self):
|
| 196 |
+
grid = [['.' for _ in range(7)] for _ in range(6)]
|
| 197 |
+
for row in range(6):
|
| 198 |
+
for col in range(7):
|
| 199 |
+
if self.board[row][col] == 1:
|
| 200 |
+
grid[5-row][col] = 'X' # Convert to the format used by helper functions
|
| 201 |
+
elif self.board[row][col] == 2:
|
| 202 |
+
grid[5-row][col] = 'O'
|
| 203 |
+
return grid
|
| 204 |
|
| 205 |
def parse_ai_move(self, move_str):
|
| 206 |
# Parse move like 'a', 'b', etc.
|
|
|
|
| 211 |
return -1
|
| 212 |
except:
|
| 213 |
return -1
|
| 214 |
+
|
| 215 |
+
def format_game_state(self):
|
| 216 |
+
board_str = self.board_to_string()
|
| 217 |
+
grid = self.get_grid_from_board()
|
| 218 |
+
available_positions = get_available_positions(grid)
|
| 219 |
+
|
| 220 |
+
# Format the player and AI moves
|
| 221 |
+
player_moves_str = ", ".join(self.player_moves) if self.player_moves else ""
|
| 222 |
+
ai_moves_str = ", ".join(self.ai_moves) if self.ai_moves else ""
|
| 223 |
+
|
| 224 |
+
return f"""Game State:
|
| 225 |
+
- You are playing as: O
|
| 226 |
+
- Your previous moves: {ai_moves_str}
|
| 227 |
+
- Opponent's moves: {player_moves_str}
|
| 228 |
+
- Current board state: {board_str}
|
| 229 |
+
- Next available position per column:
|
| 230 |
+
{available_positions}
|
| 231 |
+
|
| 232 |
+
Make your move."""
|
| 233 |
|
| 234 |
def create_interface():
|
| 235 |
game = ConnectFour()
|
|
|
|
| 351 |
|
| 352 |
# AI move
|
| 353 |
game.current_player = 2
|
|
|
|
| 354 |
|
| 355 |
+
# Format game state using the new format
|
| 356 |
+
game_state = game.format_game_state()
|
| 357 |
|
| 358 |
# Get AI response
|
| 359 |
response = model.create_chat_completion(
|
| 360 |
messages=[
|
| 361 |
{"role": "system", "content": SYSTEM_PROMPT},
|
| 362 |
+
{"role": "user", "content": game_state}
|
| 363 |
],
|
| 364 |
temperature=0.8,
|
| 365 |
top_p=0.95,
|
|
|
|
| 372 |
# Extract reasoning and move
|
| 373 |
try:
|
| 374 |
reasoning = ai_response.split("<reasoning>")[1].split("</reasoning>")[0].strip()
|
| 375 |
+
move_str = extract_xml_move(ai_response)
|
| 376 |
+
|
| 377 |
+
if not move_str:
|
| 378 |
+
raise ValueError("Invalid move format from AI")
|
| 379 |
+
|
| 380 |
ai_col = game.parse_ai_move(move_str)
|
| 381 |
|
| 382 |
if ai_col == -1:
|
|
|
|
| 425 |
game.board = np.zeros((6, 7))
|
| 426 |
game.current_player = 1
|
| 427 |
game.game_over = False
|
| 428 |
+
game.player_moves = []
|
| 429 |
+
game.ai_moves = []
|
| 430 |
return [
|
| 431 |
render_board(),
|
| 432 |
"Your turn! Click a button to drop your piece!",
|