mirror of
https://github.com/InternLM/InternBootcamp.git
synced 2026-04-24 17:05:00 +00:00
a8249acc18
* feat(run_eval): add checkpoint resume functionality and update example documentation; - update new bootcamp benchmark dataset * refactor(data_pipeline): optimize data generation pipeline; add multiple preset configurations for data generation * docs: update bootcamp list and add new scripts - Update Fulllist_InternBootcamp.md with new bootcamps and categories - Add new scripts to .gitignore: - examples/pipelines/filter_autogen_configs.py - examples/pipelines/quickgen_data_configs_from_eval_meta.py - Update dependencies in setup.py: - Add scipy and scikit-learn * refactor(internbootcamp): update bootcamp modules and improve error handling - Update import statements in __init__.py files - Add timestamp to target directory name in verl_data_preprocess.py - Improve error handling and scoring logic in bootcamp_judger.py - Remove unnecessary comments and update puzzle descriptions in multiple files
241 lines
8.5 KiB
Python
Executable file
241 lines
8.5 KiB
Python
Executable file
"""### 谜题描述
|
||
|
||
Sudoku is a logic-based number placement puzzle played on a square grid divided into smaller subgrids (called \"regions\" or \"blocks\"). The rules are as follows:
|
||
|
||
1. **Grid Structure**:
|
||
- The puzzle grid is a square of size N×N, where N is a perfect square (e.g., 9×9, 16×16).
|
||
- The grid is subdivided into N smaller rectangular regions, each of size √N×√N. For example, a 9×9 grid has nine 3×3 regions.
|
||
|
||
2. **Number Placement**:
|
||
- The grid is partially filled with numbers (or symbols) at the start.
|
||
- The solver must fill every empty cell with a number from **1 to N** (or symbols equivalent in count to N).
|
||
|
||
3. **Core Rules**:
|
||
- **Row Constraint**: Each number must appear **exactly once** in every row.
|
||
- **Column Constraint**: Each number must appear **exactly once** in every column.
|
||
- **Region Constraint**: Each number must appear **exactly once** in every subgrid/region.
|
||
|
||
The puzzle is solved when all cells are filled while satisfying all three constraints. No arithmetic or guessing is required—only logical deduction.
|
||
|
||
|
||
请完成上述谜题的训练场环境类实现，包括所有必要的方法。
|
||
"""
|
||
|
||
from bootcamp import Basebootcamp
|
||
import math
|
||
import random
|
||
from typing import List, Optional
|
||
|
||
class Sudokubootcamp(Basebootcamp):
|
||
def __init__(self, size: int = 9):
|
||
"""
|
||
初始化数独训练场参数
|
||
|
||
参数:
|
||
size: 数独尺寸（必须为完全平方数，默认9）
|
||
"""
|
||
sqrt_n = math.isqrt(size)
|
||
if sqrt_n * sqrt_n != size:
|
||
raise ValueError("Size必须为完全平方数")
|
||
self.size = size
|
||
self.sqrt_n = sqrt_n
|
||
|
||
def case_generator(self) -> dict:
|
||
"""
|
||
生成数独谜题实例
|
||
|
||
返回包含以下信息的字典:
|
||
- puzzle: 数独初始网格（0表示空格）
|
||
- size: 数独尺寸
|
||
- region_rows: 子区域行数
|
||
- region_cols: 子区域列数（同region_rows）
|
||
"""
|
||
# 生成完整解
|
||
solution = self._generate_full_sudoku()
|
||
|
||
# 挖空50%的格子（可根据需求调整比例）
|
||
puzzle = self._dig_holes(solution.copy(), dig_prob=0.5)
|
||
|
||
return {
|
||
"puzzle": [row.copy() for row in puzzle],
|
||
"size": self.size,
|
||
"region_rows": self.sqrt_n,
|
||
"region_cols": self.sqrt_n
|
||
}
|
||
|
||
@staticmethod
|
||
def prompt_func(question_case: dict) -> str:
|
||
"""
|
||
将数独实例转换为自然语言描述的问题
|
||
|
||
参数:
|
||
question_case: case_generator生成的谜题实例
|
||
|
||
返回:
|
||
包含规则说明和当前谜题状态的格式化字符串
|
||
"""
|
||
puzzle = question_case["puzzle"]
|
||
size = question_case["size"]
|
||
region_size = question_case["region_rows"]
|
||
|
||
prompt = f"""你是一个专业数独玩家，请解决以下{size}x{size}的数独谜题。规则要求：
|
||
1. 每行必须包含1-{size}所有数字，无重复
|
||
2. 每列必须包含1-{size}所有数字，无重复
|
||
3. 每个{region_size}x{region_size}的子区域必须包含1-{size}所有数字，无重复
|
||
|
||
当前谜题状态（0表示空格）：
|
||
"""
|
||
for i, row in enumerate(puzzle):
|
||
prompt += f"第{i+1}行：" + " ".join(str(n) if n != 0 else "▢" for n in row) + "\n"
|
||
|
||
prompt += "\n请将完整解答按如下格式放在[answer]标记之间：\n[answer]\n1 2 3 ...\n4 5 6 ...\n...\n[/answer]"
|
||
return prompt
|
||
|
||
@staticmethod
|
||
def extract_output(output: str) -> Optional[List[List[int]]]:
|
||
"""
|
||
从模型输出中提取最后一个数独解
|
||
|
||
参数:
|
||
output: 包含[answer]标记的完整输出文本
|
||
|
||
返回:
|
||
二维整数矩阵，解析失败时返回None
|
||
"""
|
||
import re
|
||
matches = re.findall(r'\[answer\](.*?)\[/answer\]', output, re.DOTALL)
|
||
if not matches:
|
||
return None
|
||
|
||
try:
|
||
solution = []
|
||
for line in matches[-1].strip().split('\n'):
|
||
nums = list(map(int, line.strip().split()))
|
||
if nums:
|
||
solution.append(nums)
|
||
return solution
|
||
except:
|
||
return None
|
||
|
||
@classmethod
|
||
def _verify_correction(cls, solution: List[List[int]], identity: dict) -> bool:
|
||
"""
|
||
验证解的正确性
|
||
|
||
参数:
|
||
solution: 用户提交的解
|
||
identity: 谜题实例信息
|
||
|
||
返回:
|
||
布尔值表示解的正确性
|
||
"""
|
||
def is_valid_region(grid, row_start, col_start, size, region_size) -> bool:
|
||
"""验证子区域有效性"""
|
||
nums = set()
|
||
for i in range(row_start, row_start+region_size):
|
||
for j in range(col_start, col_start+region_size):
|
||
num = grid[i][j]
|
||
if num < 1 or num > size or num in nums:
|
||
return False
|
||
nums.add(num)
|
||
return True
|
||
|
||
puzzle = identity["puzzle"]
|
||
size = identity["size"]
|
||
region_size = identity["region_rows"]
|
||
|
||
# 基本维度检查
|
||
if len(solution) != size or any(len(row) != size for row in solution):
|
||
return False
|
||
|
||
# 验证初始条件
|
||
for i in range(size):
|
||
for j in range(size):
|
||
if puzzle[i][j] != 0 and solution[i][j] != puzzle[i][j]:
|
||
return False
|
||
|
||
# 验证行、列、子区域
|
||
valid_range = set(range(1, size+1))
|
||
for i in range(size):
|
||
if set(solution[i]) != valid_range: # 行验证
|
||
return False
|
||
if set(solution[j][i] for j in range(size)) != valid_range: # 列验证
|
||
return False
|
||
|
||
# 子区域验证
|
||
for i in range(0, size, region_size):
|
||
for j in range(0, size, region_size):
|
||
if not is_valid_region(solution, i, j, size, region_size):
|
||
return False
|
||
|
||
return True
|
||
|
||
def _generate_full_sudoku(self) -> List[List[int]]:
|
||
"""生成有效完整数独的核心算法"""
|
||
size = self.size
|
||
region_size = self.sqrt_n
|
||
grid = [[0]*size for _ in range(size)]
|
||
|
||
# 填充对角线子区域
|
||
for i in range(0, size, region_size):
|
||
nums = list(range(1, size+1))
|
||
random.shuffle(nums)
|
||
for x in range(region_size):
|
||
for y in range(region_size):
|
||
grid[i+x][i+y] = nums[x*region_size + y]
|
||
|
||
# 解数独
|
||
self._solve_sudoku(grid)
|
||
return grid
|
||
|
||
def _solve_sudoku(self, grid: List[List[int]]) -> bool:
|
||
"""回溯法解数独"""
|
||
size = self.size
|
||
region_size = self.sqrt_n
|
||
empty = self._find_empty(grid)
|
||
|
||
if not empty:
|
||
return True
|
||
|
||
row, col = empty
|
||
for num in random.sample(range(1, size+1), size): # 随机尝试增加多样性
|
||
if self._is_safe(grid, row, col, num):
|
||
grid[row][col] = num
|
||
if self._solve_sudoku(grid):
|
||
return True
|
||
grid[row][col] = 0
|
||
return False
|
||
|
||
def _find_empty(self, grid: List[List[int]]) -> Optional[tuple]:
|
||
"""寻找下一个空单元格"""
|
||
for i in range(self.size):
|
||
for j in range(self.size):
|
||
if grid[i][j] == 0:
|
||
return (i, j)
|
||
return None
|
||
|
||
def _is_safe(self, grid: List[List[int]], row: int, col: int, num: int) -> bool:
|
||
"""检查数字是否可以安全填入"""
|
||
size = self.size
|
||
region_size = self.sqrt_n
|
||
|
||
# 检查行和列
|
||
if num in grid[row] or num in [grid[i][col] for i in range(size)]:
|
||
return False
|
||
|
||
# 检查子区域
|
||
start_row, start_col = row - row%region_size, col - col%region_size
|
||
for i in range(region_size):
|
||
for j in range(region_size):
|
||
if grid[start_row+i][start_col+j] == num:
|
||
return False
|
||
return True
|
||
|
||
def _dig_holes(self, grid: List[List[int]], dig_prob: float) -> List[List[int]]:
|
||
"""挖洞生成谜题（保证至少有一个解）"""
|
||
size = self.size
|
||
for i in range(size):
|
||
for j in range(size):
|
||
if random.random() < dig_prob:
|
||
grid[i][j] = 0
|
||
return grid
|