import json import math import random from internbootcamp.bootcamp.base import Basebootcamp def remove_boxed(s): if "\\boxed " in s: left = "\\boxed " assert s[:len(left)] == left return s[len(left):] left = "\\boxed{" assert s[:len(left)] == left assert s[-1] == "}" return s[len(left):-1] def last_boxed_only_string(string): idx = string.rfind("\\boxed") if "\\boxed " in string: return "\\boxed " + string.split("\\boxed ")[-1].split("$")[0] if idx < 0: idx = string.rfind("\\fbox") if idx < 0: return None i = idx right_brace_idx = None num_left_braces_open = 0 while i < len(string): if string[i] == "{": num_left_braces_open += 1 if string[i] == "}": num_left_braces_open -= 1 if num_left_braces_open == 0: right_brace_idx = i break i += 1 if right_brace_idx is None: retval = None else: retval = string[idx:right_brace_idx + 1] return retval class Medcalculatorbootcamp(Basebootcamp): # 医学计算器类 def __init__(self, conf_file="./internbootcamp/libs/med_calculator/med_calculator.json", seed=None, add_rule_ratio=0.25): random.seed(seed) self.add_rule_ratio = add_rule_ratio with open(conf_file, "r", encoding="utf-8") as f: self.config = json.load(f) self.rules = self._construct_rules() def _gen_a_case(self, category, name): details = self.config[category][name] indicators = self.config["indicator"] inputs = [] match category: case 'equation': while 1: formula = details["formula"] try: for i in details["inputs"]: match indicators[i]["type"]: case "int": k = v = random.randint(*indicators[i]["range"]) case "float": v = random.uniform(*indicators[i]["range"]) if 'precision' in indicators[i]: v = round(v, indicators[i]["precision"]) k = v case "choice": rg = indicators[i]["range"] if (t := type(rg)) is list: k = v = random.choice(indicators[i]["range"]) elif t is dict: k = random.choice(list(rg.keys())) v = rg[k] inp = i.split('[')[0] + str(k) + indicators[i].get("unit", "") inputs.append(inp) formula = formula.replace(i, str(v)) target = eval(formula) break except (ZeroDivisionError, ValueError): pass except Exception as e: e.args=(*e.args, name, '公式：'+details["formula"], '带入数值：'+formula) raise e out_k = name.split('—')[-1] if 'precision' in indicators[out_k]: target = round(target, indicators[out_k]["precision"]) case 'scale': target = 0 try: for title, options in details["points"].items(): if options['type'] == '单选': selected_option = random.choice(list(options['items'].keys())) inputs.append(f'{title}: {selected_option}') target += options['items'][selected_option] elif options['type'] == '多选': selected_options = random.sample(list(options['items'].keys()), k=random.randint(1, len(options['items']))) inputs.append(f'{title}: {"、".join(selected_options)}') target += sum(options['items'][opt] for opt in selected_options) except Exception as e: e.args=(*e.args, name) raise e ret = { "category": category, "name": name, "inputs": inputs, "add_rule": random.random() < self.add_rule_ratio, "target": target, } return ret def case_generator(self): category = random.choice(['equation', 'scale']) name = random.choice(list(self.config[category].keys())) return self._gen_a_case(category, name) def _construct_rules(self): rules = {} for name, details in self.config["equation"].items(): formula = details["formula"] explanation_arr = [] for i in details["inputs"]: indicator = self.config["indicator"][i] if indicator["type"] == "choice": j = i.split('[')[0] t = f'{j}：'+ '，'.join([f'{k}：{v}' for k,v in indicator["range"].items()]) formula = formula.replace(i, j) explanation_arr.append(t) formula = formula.replace('math.', '') explanation = '' if explanation_arr: explanation = "公式说明：" + '；'.join(explanation_arr) + "\n" rules[name] = f"{name}的计算公式：{formula}\n{explanation}\n" for name, details in self.config["scale"].items(): rules[name] = f"{name}量表的评分标准：\n" for title, options in details["points"].items(): rules[name] += f"[{title}][{options['type']}]\n" rules[name] += '\n'.join([f'{k}（{v}分）' for k, v in options['items'].items()]) + "\n" rules[name] += "\n" return rules def prompt_func(self, case): indicators = self.config["indicator"] random.shuffle(case["inputs"]) inp_items = '，'.join(case["inputs"]) out_item = case["name"] other_item = '' match case["category"]: case 'equation': out_name = out_item.split('—')[-1] if 'precision' in indicators[out_name]: other_item = f"，保留{indicators[out_name]['precision']}位小数" rule = self.rules[case["name"]] if case["add_rule"] else "" instruction = f"{rule}患者信息：{inp_items}。请计算{out_item}{other_item}。" instruction_following = """Let's think step by step and output the final answer within \\boxed{xxx:xxx}. For example "\\boxed{BMI: 20.5}".""" prompt = instruction + '\n' + instruction_following return prompt @staticmethod def extract_output(output): output = last_boxed_only_string(output) if output is None: return None return remove_boxed(output) @classmethod def _verify_correction(cls, solution, identity): if ':' in solution: solution = solution.split(':')[-1].strip() elif '：' in solution: solution = solution.split('：')[-1].strip() return solution.strip() == str(identity['target']) def gen_all_case(self, k=1): cases = [] for category in ['equation', 'scale']: for name in self.config[category]: for _ in range(k): case = self._gen_a_case(category, name) cases.append(case) return cases if __name__ == '__main__': bootcamp = Medcalculatorbootcamp(seed=42) identity = bootcamp.case_generator() print(f'{identity = }') prompt = bootcamp.prompt_func(identity) print(f"Prompt: \n{prompt}") response = "...some reasoning process...\\boxed{BMI: 134.7}" print(f"Response: \n{response}") score = bootcamp.verify_score(response, identity, short_penalty=False, format_penalty=False) print(f"Score: {score}")