🎓 For labs, students & ML engineers

The weights are on pan.baidu.com.
You can't register. We can.

Papers, GitHub READMEs and competition pages love Baidu Pan links. We pull them with a premium account at full speed and give you a normal download — no Chinese phone number, no client app, no 50 KB/s.

Paste the dataset link →

19 MB/s
measured fetch speed — 50GB in under an hour
Any size
up to 5GB ready in minutes; larger delivered within ~24h
Byte-verified
folder structure preserved, total size checked against Baidu

How it works for a dataset

Three steps, and you see the price before anything downloads.

1

Paste the link + code

The pan.baidu.com/s/… link and the 4-character 提取码 from the README. We open it and measure the real size — folders included.

2

See size & price, confirm

Priced per GB from the size on Baidu: $0.79 for the first GB, $0.35 each after. A 20GB checkpoint is $7.44. Nothing is charged until you confirm.

3

Get a direct link

Single files come as-is; folders as one ZIP with the exact directory tree. wget-friendly, valid for 1 days.

Built for the ugly cases

Nested folders

Datasets split across train/, val/, annotations/ keep their structure. Duplicate file names in different subfolders are handled correctly — many tools silently drop them.

Thousands of files

We page through large folders instead of stopping at Baidu's first 1,000 entries.

Big checkpoints

Multi-GB safetensors and split archives are the normal case here, not the exception. Anything over the automatic limit goes to a human and you get a Telegram ping when it's ready.

Researcher FAQ

Can you handle a 40GB dataset folder?

Yes. Most links are ready within minutes; very large ones take longer and are delivered within about 24 hours — you still see the exact price before confirming, and the folder tree is preserved.

Can I pay with a university card or PayPal?

PayPal (including cards through PayPal), per link. We can issue a receipt on request.

Is this legal?

We fetch files that the owner has deliberately shared publicly via a share link, on your request, and delete them after the retention window. We do not host content ourselves and we honour takedown requests.

Can you push it straight to my Google Drive or Hugging Face?

Google Drive delivery is coming next; Hugging Face upload is on the roadmap. For now you get a direct link that works with wget/curl on your cluster.

The link in the paper needs a code I don't have.

Baidu shares can require a 4-character extraction code (提取码) — it is usually right next to the link in the README or paper. Without it nobody can open the share.

Is the link from that 2021 paper still alive? Check it free first.