deepcoder-gold-standard-solutions
收藏魔搭社区2025-11-21 更新2025-08-16 收录
下载链接：
https://modelscope.cn/datasets/RealmSky/deepcoder-gold-standard-solutions
下载链接
链接失效反馈官方服务：
资源简介：
# deepcoder-gold-standard-solutions
A dataset with gold standard solutions for DeepCoder examples. Can be used to debug DeepCoder rewards.
It is not guaranteed that all gold standard solutions pass all tests though!
Derived from `justus27/deepcoder-train` and `justus27/test-vfc`.
The following script was used to generate the dataset.
```python
from datasets import load_dataset
NUM_PROC = 8
ds = load_dataset("justus27/deepcoder-train", split="train")
ds_gold = load_dataset("justus27/test-vfc", split="train")
ds_gold = ds_gold.filter(lambda x: x["gold_standard_solution"] is not None)
ds_prompts = set(ds["prompt"])
ds_gold_prompts = set(ds_gold["prompt"])
common_prompts = ds_prompts.intersection(ds_gold_prompts)
ds = ds.filter(lambda x: x["prompt"] in common_prompts, num_proc=NUM_PROC)
ds_gold = ds_gold.filter(lambda x: x["prompt"] in common_prompts, num_proc=NUM_PROC)
map_gold = {x["prompt"]: x["gold_standard_solution"] for x in ds_gold}
ds = ds.map(lambda x: {"gold_standard_solution": map_gold[x["prompt"]]}, num_proc=NUM_PROC)
ds.push_to_hub("PrimeIntellect/deepcoder-gold-standard-solutions", split="train")
```
# deepcoder-gold-standard-solutions
本数据集为DeepCoder示例提供金标准解决方案，可用于调试DeepCoder的奖励信号。不过，本数据集不保证所有金标准解决方案均可通过全部测试用例！
本数据集衍生自`justus27/deepcoder-train`与`justus27/test-vfc`数据集。
下述Python脚本用于生成本数据集：
python
from datasets import load_dataset
NUM_PROC = 8
ds = load_dataset("justus27/deepcoder-train", split="train")
ds_gold = load_dataset("justus27/test-vfc", split="train")
ds_gold = ds_gold.filter(lambda x: x["gold_standard_solution"] is not None)
ds_prompts = set(ds["prompt"])
ds_gold_prompts = set(ds_gold["prompt"])
common_prompts = ds_prompts.intersection(ds_gold_prompts)
ds = ds.filter(lambda x: x["prompt"] in common_prompts, num_proc=NUM_PROC)
ds_gold = ds_gold.filter(lambda x: x["prompt"] in common_prompts, num_proc=NUM_PROC)
map_gold = {x["prompt"]: x["gold_standard_solution"] for x in ds_gold}
ds = ds.map(lambda x: {"gold_standard_solution": map_gold[x["prompt"]]}, num_proc=NUM_PROC)
ds.push_to_hub("PrimeIntellect/deepcoder-gold-standard-solutions", split="train")
提供机构：
maas
创建时间：
2025-08-14
