ColossalAI
ColossalAI copied to clipboard
hpcaitech
能否提供一个可以直接运行的grpo数据集
想要运行grpo，根据readme开始执行，下载qwedsacf/competition_math dataset数据集
prepare_prompt_dataset.sh开始就出现错误，然后就运行不下去了。
希望作者可以对readme的可复现方式详细说明一下，谢谢
Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑🤝🧑👫🧑🏿🤝🧑🏻👩🏾🤝👨🏿👬🏿
Title: Can you provide a grpo dataset that can be run directly
If you want to run grpo, start executing according to readme
The error occurred at the beginning of prepare_prompt_dataset.sh and then it could not be run anymore.
I hope the author can explain in detail the reproducible method of readme. Thank you
Hi, we are under intensive development and will release a new version soon including more concise documentation and speed optimization.
想要运行grpo，根据readme开始执行，下载qwedsacf/competition_math dataset数据集
prepare_prompt_dataset.sh开始就出现错误，然后就运行不下去了。
希望作者可以对readme的可复现方式详细说明一下，谢谢
需要将数据集转换成jsonl格式再运行prepare_prompt_dataset.sh
想要运行grpo，根据readme开始执行，下载qwedsacf/competition_math dataset数据集 prepare_prompt_dataset.sh开始就出现错误，然后就运行不下去了。 希望作者可以对readme的可复现方式详细说明一下，谢谢
需要将数据集转换成jsonl格式再运行prepare_prompt_dataset.sh
实验过 ds = load_from_disk('./deepseek/data/NuminaMath-TIR') ds['train'].to_json("./deepseek/data/NuminaMath-TIR-jsonl/train.jsonl") ds['test'].to_json("./deepseek/data/NuminaMath-TIR-jsonl/test.jsonl")
然后跑完还是报错，所以我怀疑这个代码跑通的真实性。
Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑🤝🧑👫🧑🏿🤝🧑🏻👩🏾🤝👨🏿👬🏿
If you want to run grpo, start execution according to readme, download the qwedsacf/competition_math dataset dataset prepare_prompt_dataset.sh error occurs at the beginning and then it cannot be run. I hope the author can explain in detail the reproducible method of readme, thank you
You need to convert the dataset into jsonl format and then run prepare_prompt_dataset.sh
Experimented ds = load_from_disk('./deepseek/data/NuminaMath-TIR') ds['train'].to_json("./deepseek/data/NuminaMath-TIR-jsonl/train.jsonl") ds['test'].to_json("./deepseek/data/NuminaMath-TIR-jsonl/test.jsonl")
Then after running, it still reported an error, so I doubt the authenticity of this code running.