LOCAL.md

Local install guide (Phase 4)

Install from PyPI

pip install tuneprompt

• PyPI: https://pypi.org/project/tuneprompt/
• Import: from promptforge import PromptForge
• CLI: tuneprompt or promptforge (also python -m promptforge)

Then download models:

python -m promptforge download \
  --quality-repo ArjunShukla/PromptForge-Quality \
  --optimizer-repo ArjunShukla/PromptForge-Optimizer

Install from source

# from repo root
python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
# source .venv/bin/activate

pip install -U pip
pip install -e ".[demo,dev]"

GPU (required for local training)

pip install torch often installs the CPU build. On NVIDIA GPUs (including RTX 50-series / Blackwell), install CUDA wheels explicitly:

pip uninstall -y torch torchvision torchaudio
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

Then verify:

python -c "import torch; print(torch.__version__, torch.cuda.is_available(), torch.cuda.get_device_name(0))"
promptforge doctor

You want cuda_available: True and a CUDA torch build like 2.x.x+cu128 (not +cpu).

First-time setup

promptforge init
promptforge doctor

# After you publish models to the Hub:
promptforge download --quality-repo ArjunShukla/PromptForge-Quality \
  --optimizer-repo ArjunShukla/PromptForge-Optimizer

# Or point at Colab exports:
promptforge init \
  --quality-model outputs/promptforge-quality-model \
  --optimizer-model outputs/promptforge-optimizer-model

CLI

promptforge analyze "Build me a website"
promptforge optimize "Build me a website"
promptforge run "Build me a website"
promptforge run --file prompt.txt --json
echo "Make an app." | promptforge analyze -
promptforge eval
promptforge space

Python API

from promptforge import PromptForge

pf = PromptForge.from_config()

print(pf.analyze("Build me a website"))
print(pf.optimize("Build me a website"))
print(pf.run("Build me a website"))

Notebooks

• Self-contained Colab: notebooks/colab/
• Package-driven: notebooks/package/
• Index: notebooks/README.md

Phase 4 smoke notebook: notebooks/colab/04_local_package.ipynb

Environment variables

Variable	Meaning
PROMPTFORGE_QUALITY_MODEL	Quality model path or Hub id
PROMPTFORGE_OPTIMIZER_MODEL	Optimizer model path or Hub id
PROMPTFORGE_HOME	Override ~/.promptforge
PROMPTFORGE_CONFIG	Override config yaml path
PROMPTFORGE_PREFER_GPU	1 / 0
PROMPTFORGE_REQUIRE_GPU	Fail training if no CUDA

Tests

pytest -q