markutils

Small grab-bag of data and model utilities. Two CLI subcommands so far:

# Convert between csv / tsv / json / jsonl / parquet (format inferred from extension)
markutils convert input.csv output.parquet

# Print tensor shapes from a safetensors checkpoint
markutils inspect model.safetensors

Library-side: markutils.utils.data.load_dataframe / save_dataframe for the same format-roundtrip helpers, and markutils.checkpoints.get_state_dict / print_state_dict_shapes for safetensors inspection.

Install

pip install git+https://github.com/marksverdhei/markutils.git

Develop

This project uses uv:

curl -LsSf https://astral.sh/uv/install.sh | sh
git clone https://github.com/marksverdhei/markutils.git
cd markutils
uv sync --dev
source .venv/bin/activate
uv run pytest