hkust-nlp/gsm8k-fix

---
dataset_info:
features:
- name: id
dtype: string
- name: query
dtype: string
- name: query4test
dtype: bool
- name: query_src_agent
dtype: string
- name: query_src_method
dtype: string
- name: domain
dtype: string
- name: stage
dtype: string
- name: gt_ans
dtype: string
- name: query_metadata
struct:
- name: n_step
dtype: int64
- name: resp
dtype: string
- name: resp_src_agent
dtype: string
- name: resp_src_method
dtype: string
- name: ans
dtype: string
- name: ans_correct
dtype: bool
- name: ration_type
dtype: string
- name: proc_correct_prob
dtype: float64
- name: resp_metadata
struct:
- name: n_step
dtype: int64
splits:
- name: train
num_bytes: 4970081.335646042
num_examples: 7473
download_size: 2218640
dataset_size: 4970081.335646042
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: mit
task_categories:
- text-generation
language:
- en
tags:
- mathematics
pretty_name: GSM8K (Fixed)
size_categories:
- 1K<n<10K
---

# GSM8K (Fixed)

Some **erroneous labels** exist in the GSM8K dataset.

This dataset is fixed from https://github.com/openai/grade-school-math/blob/master/grade_school_math/data/train.jsonl with the code appended at the end.

The errors are located by delving into **unreasonably low pass rates by the strong DeepSeekMath-7B-RL** and hopefully should be exhaustive.

This dataset is used by [the **🎯DART-Math** project](https://github.com/hkust-nlp/dart-math) to synthesize data.

> [!WARNING]
> ⚠️ Only the **training** set has been fixed so far.

```python
for dp in collected_dps:
if dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:167":
# Janet filmed a new movie that is 60% longer than her previous 2-hour long movie. Her previous movie cost $50 per minute to film, and the newest movie cost twice as much per minute to film as the previous movie. What was the total amount of money required to film Janet's entire newest film?
dp["resp"] = (
"The first movie was 2*60=120 minutes\nSo this movie is 120*.6=72 minutes longer\nSo this movie is 192 minutes\nIt also cost 50*2=$100 per minute to film\nSo it cost 192*100=$19200"
)
dp["ans"] = dp["gt_ans"] = "19200"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:474":
# A club with 30 members ordered fruit juices. Two-fifths of them ordered lemon juice. One-third of the remaining members ordered mango juice, and the rest ordered orange juice. How many members ordered orange juice?
dp["resp"] = (
"30 x 2/5 = 12 members ordered lemon juice.\nSo, 30 - 12 = 18 members did not order lemon juice.\nSince 1/3 of the remaining ordered mango juice, then 18 x 1/3 = 6 members ordered mango juice.\nTherefore, 18 - 6 = 12 members ordered orange juice."
)
dp["ans"] = dp["gt_ans"] = "12"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:1081":
# John has 2 hives of bees. One of the hives has 1000 bees and produces 500 liters of honey. The second has 20% fewer bees but each bee produces 40% more honey. How much honey does he produce?
dp["resp"] = (
"The second hive has 20/100*1000 = 200 fewer bees.\nThis translates to 1000-200 = 800 bees.\nEach bee in the first hive produces 500/1000 = 0.5 liters\nThe second hive has bees each producing 1.4*0.5 = 0.7 liters\nThe total amount of honey produces by the bees in the second hive is 0.7*800 = 560\nThe total honey produced is 500+560 = 1060 liters"
)
dp["ans"] = dp["gt_ans"] = "1060"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:2338":
# A bottle can hold 2 cups of water. How many cups of water are needed to fill up 10 whole bottles and 5 half-capacity bottles?
dp[
"resp"
] = r"""For 10 whole bottles, you will need 10*2=20 cups of water.
With 5 half-capacity bottles, it requires 5*1=5 cups of water.
In total, you need to have 20+5=25 cups of water to fill them all."""
dp["ans"] = dp["gt_ans"] = "25"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:2620":
# Ryan started with 36 tokens at the arcade. Ryan wasted a third of his tokens on Pac-Man, a fourth of his tokens on Candy Crush, and 7 on Ski-ball. Then, his parents bought him seven times as many tokens as he spent on Ski-ball. How many tokens did Ryan end up with?
dp[
"resp"
] = r"""Ryan used 36/3 = 12 tokens on Pac-Man.
Ryan used 36/4 = 9 tokens on Candy Crush.
Ryan used a total of 12+9+7 = 28 tokens on all three games.
Ryan had 36-28 = 8 tokens left.
Ryan’s parents bought him 7*7 = 49 more tokens.
Ryan had 8+49 = 57 tokens after his parents bought him some."""
dp["ans"] = dp["gt_ans"] = "57"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:2770":
# Robert and Teddy are planning to buy snacks for their friends. Robert orders five boxes of pizza at $10 each box and ten cans of soft drinks at $2 each. Teddy buys six hamburgers at $3 each and an additional ten cans of soft drinks. How much do they spend in all?
dp["ans"] = dp["gt_ans"] = "108"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:3263":
# Andy gets a cavity for every 4 candy canes he eats. He gets 2 candy canes from his parents and 3 candy canes each from 4 teachers. Then he uses his allowance to buy 1/7 as many candy canes as he was given. How many cavities does he get from eating all his candy canes?
dp["ans"] = dp["gt_ans"] = "4"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:3529":
# During one game, a total of 50 people attended a baseball team’s games. Forty percent and thirty-four percent of the audiences are supporters of the first and second teams, respectively. How many people attended the game did not support either of the teams?
dp[
"resp"
] = r"""50 x 40/100 = 20 people support the first team.
50 x 34/100 = 17 people support the second team.
So, a total of 20 + 17 = 37 people supported the first and second teams.
Thus, 50 - 37 = 13 people did not support either team."""
dp["ans"] = dp["gt_ans"] = "13"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:3995":
# Kimiko is retiling her kitchen floor. Each tile is 6 square inches. If her kitchen is 48 inches by 72 inches, how many tiles does Kimiko need to buy?
# "First figure out how many tiles are in one row by dividing the kitchen's width by each tile's width: 48 inches / 6 inches/tile = 8 tiles in one row\nThen figure out how many rows of tiles there are by dividing the kitchen's height by each tile's height: 72 inches / 6 inches/tile = 12 rows of tiles\nThen multiply the number of rows by the number of tiles per row to find the total number of tiles Kimiko needs to buy: 8 tiles/row * 12 rows = 96 tiles"
dp[
"resp"
] = r"""To find out how many tiles Kimiko needs, we first need to calculate the total area of her kitchen floor in square inches.
The area of the kitchen floor is given by the length multiplied by the width:
Area = Length × Width
So, for Kimiko's kitchen:
Area = 48 inches × 72 inches
Now, let's calculate:
Area = 48 inches × 72 inches = 3456 square inches
Now, we need to find out how many 6-square-inch tiles can fit into this total area. We can do this by dividing the total area by the area of each tile:
Number of tiles = Total area / Area of each tile
Number of tiles = 3456 square inches / 6 square inches per tile
Number of tiles ≈ 576
So, Kimiko needs to buy approximately 576 tiles to retile her kitchen floor."""
dp["ans"] = dp["gt_ans"] = "576"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:4099":
# Big Dig Mining Company mines three different types of ore: copper, iron, and nickel. Across all their mines, 10% of their output is nickel, 60% is iron, and the rest is copper. They mine 720 tons of nickel a day. How many tons of copper does Big Dig Mining Company mine daily?
# 'Let R be the total ore output of the company.\nOf Big Dig’s output, 100 - 10 - 60 = 30% is nickel.\nSince 720 is 60% of 100% of their output, 720 / R = 60 / 100.\nThus, Big Dig mines R = 100 * 720 / 60 = 1200 tons of ore daily.\nTherefore, Big Dig mines 1200 * 30 / 100 = 360 tons of copper daily.'
dp[
"resp"
] = r"""Let R be the total ore output of the company.
Of Big Dig’s output, 100 - 10 - 60 = 30% is nickel.
Since 720 is 10% of 100% of their output, 720 / R = 10 / 100.
Thus, Big Dig mines R = 100 * 720 / 10 = 7200 tons of ore daily.
Therefore, Big Dig mines 7200 * 30 / 100 = 360 tons of copper daily."""
dp["ans"] = dp["gt_ans"] = "2160"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:4105":
# Andrew eats 14 donuts on Monday, and half as many on Tuesday. On Wednesday Andrew eats 4 times as many as he did on Monday. How many donuts did Andrew eat total in the three days?
# 'Monday:14\nTuesday:14/2=7\nWednesday:4(7)=28\nTotal:14+7+28=49 donuts'
dp[
"resp"
] = r"""Monday:14
Tuesday:14/2=7
Wednesday:4*14=56
Total:14+7+56=77 donuts"""
dp["ans"] = dp["gt_ans"] = "77"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:4796":
# In a yard, the number of tanks is five times the number of trucks. If there are 20 trucks in the yard, calculate the total number of tanks and trucks in the yard.
# 'There are 5*20 = 100 tanks in the yard.\nAltogether, there are 100+20 = 120 trucks and tanks in the yard.'
dp["ans"] = dp["gt_ans"] = "120"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:5598":
# Mary just arrived at the beach. She has 4 times as many towels as Frances does. The total weight of their towels is 60 pounds. If Mary has 24 towels, how much do Frances's towels weigh in ounces?
# 'Frances has 24/4 = 6 towels.\nThey have 24+6=30 towels.\nEach towel weighs 60/30=2 pounds.\nFrances’s towels weigh a total of 2*4 = 8 pounds\nFrances’s towels weigh a total of 8*16 = 128 ounces'
dp[
"resp"
] = r"""Frances has 24/4 = 6 towels.
They have 24+6=30 towels.
Each towel weighs 60/30=2 pounds.
Frances’s towels weigh a total of 2*6 = 12 pounds
Frances’s towels weigh a total of 12*16 = 192 ounces"""
dp["ans"] = dp["gt_ans"] = "192"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:6768":
# Kate's hair is half as long as Emily's hair. Emily’s hair is 6 inches longer than Logan's hair. If Logan hair is 20 inches, how many inches is Kate’s hair?
# "Emily’s hair is 20-6 = 14 inches long.\nKate's hair 14/2= 7 inches long."
dp[
"resp"
] = r"""Emily’s hair is 20+6 = 26 inches long.
Kate's hair 26/2= 13 inches long."""
dp["ans"] = dp["gt_ans"] = "13"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:7182":
# If you double a number and add 5 to the result, then that's 20 more than half of the original number. What's the original number?
# 'Let x be the original number.\n2*x+5=20+x/2\n2*x-x/2=15\n4*x-x=30\n3*x=30\nx=10'
dp["ans"] = dp["gt_ans"] = "10"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:7401":
# Mr. Finnegan has 3 tanks with a capacity of 7000 gallons, 5000 gallons, and 3000 gallons, respectively. If he fills the first tank up to 3/4 full, the second tank with water up to 4/5 of its capacity, and the third tank up to half of its capacity, how many gallons in total are in the tanks?
# 'The capacity of the first tank is 7000 gallons, and if it is filled up to 3/4 full, it carries 3/4*7000 = 5250 gallons.\nWhen the second tank is filled up to 4/5 of its capacity, it carries 4/5*5000 = 4000 gallons.\nThe total amount of water in the first two tanks is 5250+4000 = 9250 gallons.\nIf Mr. Finnegan fills the third tank with water up to half its capacity, the tank fills up with 1/2*3000 = 1500 gallons.\nIn total, the three tanks have 9350+1500 = 10850 gallons of water.'
dp[
"resp"
] = r"""The capacity of the first tank is 7000 gallons, and if it is filled up to 3/4 full, it carries 3/4*7000 = 5250 gallons.
When the second tank is filled up to 4/5 of its capacity, it carries 4/5*5000 = 4000 gallons.
The total amount of water in the first two tanks is 5250+4000 = 9250 gallons.
If Mr. Finnegan fills the third tank with water up to half its capacity, the tank fills up with 1/2*3000 = 1500 gallons.
In total, the three tanks have 9250+1500 = 10750 gallons of water."""
dp["ans"] = dp["gt_ans"] = "10750"
```

数据集信息：
特征：
- 名称：id，数据类型：字符串
- 名称：query，数据类型：字符串
- 名称：query4test，数据类型：布尔型
- 名称：query_src_agent（AI智能体，AI Agent），数据类型：字符串
- 名称：query_src_method，数据类型：字符串
- 名称：domain，数据类型：字符串
- 名称：stage，数据类型：字符串
- 名称：gt_ans（标准答案，ground truth answer），数据类型：字符串
- 名称：query_metadata，结构体类型：
- 名称：n_step，数据类型：64位整型
- 名称：resp，数据类型：字符串
- 名称：resp_src_agent（AI智能体，AI Agent），数据类型：字符串
- 名称：resp_src_method，数据类型：字符串
- 名称：ans，数据类型：字符串
- 名称：ans_correct，数据类型：布尔型
- 名称：ration_type，数据类型：字符串
- 名称：proc_correct_prob，数据类型：64位浮点型
- 名称：resp_metadata，结构体类型：
- 名称：n_step，数据类型：64位整型
数据集划分：
- 名称：训练集（train），占用字节数：4970081.335646042，样本数量：7473
下载大小：2218640
数据集总大小：4970081.335646042
配置项：
- 配置名称：默认配置（default），数据文件：
- 划分：训练集，文件路径：data/train-*
许可证：MIT许可证（mit）
任务类别：文本生成
语言：英语（en）
标签：数学
展示名称：修正版GSM8K（GSM8K (Fixed)）
样本规模类别：1千至1万个样本（1K<n<10K）

# 修正版GSM8K
原始GSM8K数据集中存在**错误标签**。
本数据集基于https://github.com/openai/grade-school-math/blob/master/grade_school_math/data/train.jsonl 修正而来，修正代码附于文档末尾。
本次错误排查通过追踪**强模型DeepSeekMath-7B-RL出现异常偏低的通过率**完成，且已尽可能覆盖全部错误。
本数据集被用于[**🎯DART-Math项目**](https://github.com/hkust-nlp/dart-math)以完成数据合成。

> [!WARNING]
> ⚠️ 截至目前仅修复了**训练集**的标签错误。

python
for dp in collected_dps:
if dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:167":
# 珍妮特拍摄了一部新电影，时长比她此前2小时的旧电影长60%。她的旧电影每分钟拍摄成本为50美元，新电影的每分钟拍摄成本是旧电影的两倍。请问拍摄珍妮特的这部新电影总共需要多少资金？
dp["resp"] = (
"旧电影时长为2*60=120分钟
因此新电影比旧电影长120*0.6=72分钟
因此新电影总时长为192分钟
新电影每分钟拍摄成本为50*2=100美元
因此总拍摄成本为192*100=19200美元"
)
dp["ans"] = dp["gt_ans"] = "19200"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:474":
# 某俱乐部共有30名会员订购果汁。其中五分之二的会员订购了柠檬汁，剩余会员中的三分之一订购了芒果汁，其余会员订购了橙汁。请问有多少会员订购了橙汁？
dp["resp"] = (
"30 × 2/5 = 12名会员订购了柠檬汁
因此，30 - 12 = 18名会员未订购柠檬汁
由于剩余会员中的1/3订购了芒果汁，因此18 × 1/3 = 6名会员订购了芒果汁
因此，18 - 6 = 12名会员订购了橙汁。"
)
dp["ans"] = dp["gt_ans"] = "12"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:1081":
# 约翰有两箱蜜蜂。第一箱有1000只蜜蜂，可产出500升蜂蜜。第二箱的蜜蜂数量比第一箱少20%，但每只蜜蜂的蜂蜜产量比第一箱高40%。请问约翰总共能产出多少蜂蜜？
dp["resp"] = (
"第二箱的蜜蜂数量比第一箱少20/100*1000 = 200只
因此第二箱共有1000-200 = 800只蜜蜂
第一箱每只蜜蜂的蜂蜜产量为500/1000 = 0.5升
第二箱每只蜜蜂的蜂蜜产量为1.4*0.5 = 0.7升
第二箱总蜂蜜产量为0.7*800 = 560升
约翰的总蜂蜜产量为500+560 = 1060升"
)
dp["ans"] = dp["gt_ans"] = "1060"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:2338":
# 一个瓶子可容纳2杯水。请问装满10个整瓶和5个半容量瓶总共需要多少杯水？
dp[
"resp"
] = r"""装满10个整瓶需要10*2=20杯水。
装满5个半容量瓶需要5*1=5杯水。
总共需要20+5=25杯水。"""
dp["ans"] = dp["gt_ans"] = "25"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:2620":
# 瑞安在游戏厅初始有36枚代币。他在吃豆人上花费了三分之一的代币，在糖果粉碎上花费了四分之一的代币，另外在滑雪球上花费了7枚代币。之后他的父母给了他相当于滑雪球花费7倍的代币。请问瑞安最终拥有多少枚代币？
dp[
"resp"
] = r"""瑞安在吃豆人上花费了36/3 = 12枚代币。
瑞安在糖果粉碎上花费了36/4 = 9枚代币。
瑞安在三款游戏上总共花费了12+9+7 = 28枚代币。
瑞安剩余代币数为36-28 = 8枚。
瑞安的父母给了他7*7 = 49枚代币。
瑞安最终拥有的代币数为8+49 = 57枚。"""
dp["ans"] = dp["gt_ans"] = "57"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:2770":
# 罗伯特和泰迪打算为朋友们购买零食。罗伯特订购了5盒披萨，每盒10美元，以及10罐软饮料，每罐2美元。泰迪订购了6个汉堡，每个3美元，另外还订购了10罐软饮料。请问他们总共花费了多少钱？
dp["ans"] = dp["gt_ans"] = "108"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:3263":
# 安迪每吃4根拐杖糖就会长一颗蛀牙。他从父母那里得到了2根拐杖糖，从4位老师那里每人得到了3根拐杖糖。之后他用零花钱买了相当于他收到的拐杖糖总数1/7的拐杖糖。请问安迪总共会长多少颗蛀牙？
dp["ans"] = dp["gt_ans"] = "4"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:3529":
# 在一场比赛中，共有50名观众到场观看棒球队的比赛。其中40%的观众支持第一支球队，34%的观众支持第二支球队。请问有多少观众不支持这两支球队中的任何一支？
dp[
"resp"
] = r"""支持第一支球队的观众人数为50 × 40/100 = 20人。
支持第二支球队的观众人数为50 × 34/100 = 17人。
支持两支球队的观众总人数为20 + 17 = 37人。
因此，不支持任何一支球队的观众人数为50 - 37 = 13人。"""
dp["ans"] = dp["gt_ans"] = "13"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:3995":
# 木崎女士正在重新铺贴厨房地板。每块瓷砖的面积为6平方英寸。她的厨房尺寸为48英寸×72英寸，请问木崎女士需要购买多少块瓷砖？
dp[
"resp"
] = r"""要计算木崎女士需要的瓷砖数量，我们首先需要计算厨房地板的总面积（平方英寸）。
厨房地板的面积等于长度乘以宽度：
面积 = 长度 × 宽度
对于木崎女士的厨房：
面积 = 48英寸 × 72英寸
计算可得：
面积 = 48英寸 × 72英寸 = 3456平方英寸
接下来，我们需要计算总面积中可以容纳多少块6平方英寸的瓷砖。用总面积除以每块瓷砖的面积即可：
瓷砖数量 = 总面积 / 单块瓷砖面积
瓷砖数量 = 3456平方英寸 / 6平方英寸/块
瓷砖数量 ≈ 576
因此，木崎女士需要购买约576块瓷砖来铺贴厨房地板。"""
dp["ans"] = dp["gt_ans"] = "576"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:4099":
# Big Dig矿业公司开采三种矿石：铜、铁和镍。在所有矿石产量中，10%为镍，60%为铁，剩余部分为铜。该公司每日开采720吨镍。请问该公司每日开采多少吨铜？
dp[
"resp"
] = r"""设R为该公司的总矿石产量。
Big Dig的总产量中，100 - 10 - 60 = 30%为铜。
由于每日开采的镍产量为720吨，占总产能的10%，因此720 / R = 10 / 100。
由此可得该公司每日总矿石产量R = 100 × 720 / 10 = 7200吨。
因此，该公司每日开采的铜产量为7200 × 30 / 100 = 2160吨。"""
dp["ans"] = dp["gt_ans"] = "2160"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:4105":
# 安德鲁周一吃了14个甜甜圈，周二吃的数量是周一的一半。周三吃的数量是周一的4倍。请问安德鲁三天总共吃了多少个甜甜圈？
dp[
"resp"
] = r"""周一：14个
周二：14/2=7个
周三：4*14=56个
总数量：14+7+56=77个"""
dp["ans"] = dp["gt_ans"] = "77"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:4796":
# 某院子里的坦克数量是卡车的5倍。如果院子里有20辆卡车，请问院子里的坦克和卡车总共有多少辆？
dp["ans"] = dp["gt_ans"] = "120"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:5598":
# 玛丽刚到海滩。她的毛巾数量是弗朗西斯的4倍。两人的毛巾总重量为60磅。如果玛丽有24条毛巾，请问弗朗西斯的毛巾总重量为多少盎司？
dp[
"resp"
] = r"""弗朗西斯的毛巾数量为24/4 = 6条。
两人的毛巾总数量为24+6=30条。
每条毛巾的重量为60/30=2磅。
弗朗西斯的毛巾总重量为2*6 = 12磅。
弗朗西斯的毛巾总重量为12*16 = 192盎司"""
dp["ans"] = dp["gt_ans"] = "192"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:6768":
# 凯特的头发长度是艾米丽的一半。艾米丽的头发比洛根长6英寸。如果洛根的头发长20英寸，请问凯特的头发长多少英寸？
dp[
"resp"
] = r"""艾米丽的头发长度为20+6 = 26英寸。
凯特的头发长度为26/2= 13英寸。"""
dp["ans"] = dp["gt_ans"] = "13"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:7182":
# 如果将一个数翻倍后再加5，得到的结果比原数的一半多20。请问原数是多少？
dp["ans"] = dp["gt_ans"] = "10"
elif dp["id"] == "grade-school-math/grade_school_math/data/train.jsonl:7401":
# 芬尼根先生有三个水箱，容量分别为7000加仑、5000加仑和3000加仑。他将第一个水箱装至3/4满，第二个水箱装至4/5满，第三个水箱装至一半满。请问三个水箱中总共有多少加仑水？
dp[
"resp"
] = r"""第一个水箱容量为7000加仑，装至3/4满时，水量为3/4*7000 = 5250加仑。
第二个水箱装至4/5满时，水量为4/5*5000 = 4000加仑。
前两个水箱的总水量为5250+4000 = 9250加仑。
第三个水箱装至一半满时，水量为1/2*3000 = 1500加仑。
因此三个水箱的总水量为9250+1500 = 10750加仑。"""
dp["ans"] = dp["gt_ans"] = "10750"