convert

Automated pipeline for converting models to GGUF format and uploading them to HF.

Supported models: models/

Sample models converted with this pipeline:

• https://huggingface.co/ggml-org/gemma-4-31B-it-GGUF
• https://huggingface.co/ggml-org/Qwen3.6-35B-A3B-GGUF
• https://huggingface.co/ggml-org/gpt-oss-20b-GGUF

Usage

Requires HF_TOKEN with write permission to the target org.

# Convert all models
bash convert.sh --owner <org>

# Convert a single model
bash convert.sh --owner <org> --one gemma-4-12b

# Convert models matching a filter
bash convert.sh --owner <org> --filter '^gemma'

# Force re-convert (ignore SHA checks)
bash convert.sh --owner <org> --force

# Keep downloaded models and generated upload directories
bash convert.sh --owner <org> --keep

# Keep downloaded models and generated upload directories, but don't upload
bash convert.sh --owner <org> --keep --no-upload

# Use a specific llama.cpp commit
bash convert.sh --owner <org> --llama-commit <commit>

# Run via HF Jobs (cloud infrastructure)
bash hf-job.sh --owner <org>

# Run via HF Jobs with a custom hardware flavor
bash hf-job.sh --owner <org> --hardware cpu-basic

Notes

• Models are converted only if at least one of the source models has been updated

• The README.md files are always updated, regardless if the source models have been updated

• All models in ggml-org are auto-converted by a GitHub Actions workflow once per week

  HF_TOKEN=xxx bash hf-job.sh --owner ggml-org
  

• A maintainer from the ggml-org/hf team can start the workflow manually from the Actions pane

• The conversion uses the latest master of llama.cpp unless --llama-commit is set