Learn AI Create the Future — with Pranay
Main site

Error database

batch response: This repository is over its data quota (git-lfs)

The repo owner's Git LFS bandwidth allowance on GitHub is exhausted, so large files will not download for anyone. As a downloader, get the weights from another host; as an owner, stop storing model weights in GitHub LFS.

The message you saw
batch response: This repository is over its data quota (git-lfs)

By Updated

The error

Output
batch response: This repository is over its data quota. Account responsible for LFS bandwidth should purchase more data packs to restore access.
error: failed to fetch some objects from 'https://github.com/user/repo.git/info/lfs'

The clone or pull completes for normal files, and the large files arrive as tiny pointer stubs instead.

What it means

Git LFS (Large File Storage) keeps big files outside normal Git history; the repo stores small pointer files, and clients fetch the real content from an LFS server. GitHub meters that fetching: each account gets a limited free storage and bandwidth allowance, and popular repos with model weights burn through it fast. Once exhausted, the LFS server refuses everyone — including the owner — until the quota resets or is raised. Note who the message blames: the repository owner's quota, not yours.

Why it happens

Model checkpoints are gigabytes, and every clone re-downloads them. A repo with 5 GB of weights and a few hundred clones a month exceeds free allowances comfortably. Nothing anyone did was wrong; GitHub LFS is priced for source-adjacent assets, not for ML weight distribution.

How to fix it

1. As a downloader: get the code without the LFS files, then find the weights elsewhere.

bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://github.com/user/repo.git

The clone succeeds with pointer stubs in place of weights. Then check the repo's README and releases page — maintainers of quota-hit repos typically host weights on Hugging Face Hub, in GitHub Releases, or on a mirror. A Hugging Face copy downloads cleanly with:

bash
huggingface-cli download user/model-name

2. As a downloader in a hurry, try again after the monthly reset — quotas are per-month, and early-month clones succeed where late-month ones fail. Unreliable, but occasionally sufficient.

3. As the repo owner: move weight distribution off GitHub LFS. The sustainable options:

  • Hugging Face Hub — free hosting for public model weights, resumable downloads, no bandwidth anxiety; the standard home for ML artifacts.
  • GitHub Releases — release assets are free to download and hold files up to 2 GiB each; fine for modest checkpoints.

Keep the Git repo for code, and point to the weights in the README. Buying LFS data packs postpones the next overage without preventing it.

4. As the owner, purge weights from LFS after migrating so fresh clones stop attempting LFS fetches — remove the files, update .gitattributes, and note the new location prominently.

How to prevent it

The dividing rule: Git (and LFS) for code and small assets; artifact hosts for model weights and datasets. Adopt it before publishing anything popular, because migrating after a repo takes off is exactly when the quota is already gone.