Skip to main content

​
Core install

Install the latest version straight from GitHub — no clone required: 
pip install "mill-eval @ git+https://github.com/haideraltahan/Mill.git"
For development (editable install with test/lint tooling), clone first: 
git clone https://github.com/haideraltahan/Mill
cd Mill
pip install -e ".[dev]"
The core package pulls in:
DependencyPurpose
torch >= 2.2Tensor computation
transformers >= 4.40HF model loading
datasets >= 2.18HF dataset streaming
pandas + pyarrowFeather output caching
pydantic >= 2.0ChatMessages protocol
pyyamlSLURM cluster config
accelerate >= 0.26Multi-GPU device maps
fireCLI routing
rich + prompt_toolkitTerminal UI

​
Optional extras

Add the extras for the backends you need in the brackets (no clone required): 
# vLLM backend — fastest local inference
pip install "mill-eval[vllm] @ git+https://github.com/haideraltahan/Mill.git"

# LiteLLM backend — OpenAI, Anthropic, and 100+ API providers
pip install "mill-eval[litellm] @ git+https://github.com/haideraltahan/Mill.git"

# Vision backends — CLIP (open_clip) + timm   (or [clip] / [timm] individually)
pip install "mill-eval[vision] @ git+https://github.com/haideraltahan/Mill.git"

# Video decoding (decord)
pip install "mill-eval[video] @ git+https://github.com/haideraltahan/Mill.git"

# Audio tasks (librosa)
pip install "mill-eval[audio] @ git+https://github.com/haideraltahan/Mill.git"

# Code execution tasks (evaluate library)
pip install "mill-eval[code] @ git+https://github.com/haideraltahan/Mill.git"
You can combine extras: 
pip install "mill-eval[vllm,litellm,video] @ git+https://github.com/haideraltahan/Mill.git"
If you cloned the repo for development, the editable equivalent is pip install -e ".[vllm]".

​
Verify

mill ls
If the TUI opens, your install is working.

​
Environment variables

VariableDefaultDescription
MILL_OUTPUT_DIR./mill_resultsDefault output directory
MILL_CACHE_DIR~/.cache/millConfig, SLURM jobs, logs
These are overridden by the --output_dir and --cache_dir CLI flags.