Last updated:

Upload Models

Upload Methods Overview

The platform supports four methods for uploading model files:

Method	Best For
Web Upload	Small files under 5MB per file
Git Upload	Version-controlled code and config files, with Git LFS for large files
CLI (csghub-cli)	Very large files (over 5GB) with resumable uploads
Python SDK (pycsghub)	Programmatic batch uploads

Web Upload

Navigate to the Files tab of the model repository.
Click the Add File button.
Select local files to upload (individual file size limited to 5MB).
Enter a commit message and confirm the submission.

Note

The Web interface is suitable for uploading configuration files, READMEs, and other small files. For large model weight files, use Git LFS, csghub-cli, or the SDK.

Git Upload

Prerequisites

Before uploading model repositories containing large files (e.g., .safetensors, .bin), install Git LFS:

# Install Git LFS (macOS)
brew install git-lfs

# Install Git LFS (Ubuntu/Debian)
sudo apt-get install git-lfs

# Initialize Git LFS
git lfs install

Clone and Upload

# Clone the model repository
git clone https://<platform-host>/<namespace>/<model-name>
cd <model-name>

# Copy model files into the repository directory
cp /path/to/your/model-files/* .

# Add, commit, and push
git add .
git commit -m "Upload model files"
git push

For authentication, use an access token:

git clone https://<username>:<access-token>@<platform-host>/<namespace>/<model-name>

Automatic Git LFS Tracking

The platform automatically configures Git LFS tracking rules for common model file extensions, so you do not need to manually run git lfs track:

File Extension	Description
`.safetensors`	SafeTensors format weight files
`.bin`	PyTorch binary weight files
`.pt`	PyTorch model files
`.pth`	PyTorch checkpoint files
`.onnx`	ONNX format model files
`.gguf`	GGUF format model files (llama.cpp)
`.ggml`	GGML format model files
`.h5`	HDF5 format weight files (Keras/TensorFlow)
`.msgpack`	MessagePack format model files
`.zip`	Compressed archive files
`.tar.gz`	Compressed archive files

To track additional file types, add them manually:

git lfs track "*.custom_ext"
git add .gitattributes

CLI Upload (csghub-cli)

csghub-cli is ideal for uploading very large files with resumable upload support.

Installation

pip install csghub-sdk

Upload a Large Folder

csghub-cli upload <namespace>/<model-name> /path/to/local/folder \
  -e https://<platform-host> \
  -t <access-token>

Upload a Single File

csghub-cli upload <namespace>/<model-name> /path/to/model.safetensors \
  -e https://<platform-host> \
  -t <access-token>

Python SDK Upload (pycsghub)

Use the Repository class from pycsghub to upload model files programmatically.

Installation

pip install pycsghub

Upload Example

from pycsghub.repo_reader import AutoModelForCausalLM

token = "<access-token>"
endpoint = "https://<platform-host>"
repo_id = "<namespace>/<model-name>"

from pycsghub.repository import Repository

repo = Repository(
    repo_id=repo_id,
    repo_type="model",
    token=token,
    endpoint=endpoint
)

# Upload a local directory to the repository
repo.upload_folder(
    folder_path="/path/to/local/model",
    commit_message="Upload model files via SDK"
)

View Commit History

After uploading, you can view the file commit history in the Commits tab of the model repository, including the timestamp, author, and list of changed files for each commit.

Note

Access tokens can be generated in User Settings → Access Tokens. It is recommended to create separate tokens for different purposes and keep them secure.