hf-s3-mirror

CLI tool for mirroring HuggingFace models to S3-compatible storage.

Installation

pip install hf-s3-mirror

Configuration

Create .env file in your working directory:

S3_ACCESS_KEY=your_access_key
S3_SECRET_KEY=your_secret_key
S3_ENDPOINT_URL=http://your-s3-endpoint
S3_REGION=ru-1
S3_BUCKET=model-bucket
S3_USE_SSL=false

Or export environment variables:

export S3_ACCESS_KEY=your_access_key
export S3_SECRET_KEY=your_secret_key
export S3_ENDPOINT_URL=http://your-s3-endpoint
export S3_REGION=ru-1
export S3_BUCKET=model-bucket

Usage

Upload model from HuggingFace to S3

# Basic usage
hf-s3 upload openai/whisper-large-v3

# With custom bucket and prefix
hf-s3 upload lab-ii/whisper-large-v3 -b my-bucket -p models/

Upload local directory to S3

hf-s3 upload-local ./my-model -b model-bucket -p models/

Download model from S3 mirror

hf-s3 download lab-ii/whisper-large-v3 -o ./models

List bucket contents

# List root
hf-s3 list-bucket

# List specific path
hf-s3 list-bucket -p models--lab-ii--whisper-large-v3

Delete model from S3

# By repo_id (will ask for confirmation)
hf-s3 delete lab-ii/whisper-large-v3

# By direct path
hf-s3 delete models--lab-ii--whisper-large-v3

# Skip confirmation (for scripts)
hf-s3 delete lab-ii/bad-model --force

Commands Reference

Common Options

S3 Storage Structure

Models are stored with the following naming convention:

bucket/
  models--organization--model-name/
    config.json
    model.bin
    tokenizer.json
    ...

License

Apache 2.0