Datasets:
Emilia-YODAS
Hi,
Congratulations on the release of Emilia-YODAS—it looks awesome!
I had two questions:
- I noticed that the gated repo prompt still requires the user to agree to only use it for NC, but if I understand correctly the YODAS subset should be commercially viable?
- Would it be possible to upload the YODAS subset into a separate HF repo to make it easier to load?
Thanks!
Hi, thanks for the reminder! I have just updated the gated repo prompt; you are free to use the Yodas subset under cc-by.
For the second inquiry, since the Emilia-Yodas and the original Emilia datasets are similar in nature, we have decided to maintain a single repository. If higher-quality or more finely annotated data becomes available in the future, we will consider creating a new repo.
Please note that Emilia-Yodas can also be loaded separately within this repo. For example:
from datasets import load_dataset
path = "Emilia-YODAS/DE/*.tar"
dataset = load_dataset("amphion/Emilia-Dataset", data_files={"de": path}, split="de", streaming=True)
print(dataset) # This should only show 90 n_shards
print(next(iter(dataset['train'])))
Thanks for your questions, the README page has updated accordingly about how to load Emilia-YODAS only.
A detailed usage guide (install required packages/load partial data) can be found on the README page.
The code you might want to download Emilia-YODAS only:
from datasets import load_dataset
path = "Emilia-YODAS/**/*.tar" # Same for Emilia; just replace "Emilia-YODAS/" with "Emilia/"
dataset = load_dataset("amphion/Emilia-Dataset", data_files={"train": path}, split="train", streaming=True)
print(dataset) # here should only shows 1983 n_shards
print(next(iter(dataset)))
Thank you for the additional information! Super excited to try out the dataset!
@HarryHe @yuantuo666 Were most of the Chinese data deleted from the Emilia dataset? Should I combine the Emilia and Emilia-Yodas data? Or is it an expanded version of the same dataset?
Q: Were most of the Chinese data deleted from the Emilia dataset?
A: No. YODAS2 did not contain much Chinese data, so the total hour is fewer.
Q: Should I combine the Emilia and Emilia-YODAS data?
A: Yes and No.
- Yes: if you are doing research stuff, you can use all of them.
- No: if you are doing a commercial jobs, since Emilia is licensed with CC-BY-NC, which not allow commercial usage.
Q: Or is it an expanded version of the same dataset?
A: See above answer. It's two sets without intersection.