Python 3.8 or higher
pip or conda
For development or CPU-only inference:
1
pip install slm-builder
This installs the base dependencies without heavy ML libraries.
For training and GPU support:
1
pip install slm-builder[full]
This includes:
PyTorch
Transformers
PEFT (for LoRA)
Accelerate
ONNX Runtime
All training dependencies
For contributing to the project:
1
2
3
git clone https://github.com/isathish/slm.git
cd slm
pip install -e ".[full,dev]"
This installs in editable mode with development tools (pytest, black, flake8, etc.)
1
python - c "import slm_builder; print(slm_builder.__version__)"
Or test the CLI:
If you have NVIDIA GPU:
1
2
3
4
5
# Install PyTorch with CUDA support first
pip install torch --index-url https://download.pytorch.org/whl/cu118
# Then install SLM Builder
pip install slm-builder[full]
1
pip install beautifulsoup4
1
2
3
4
5
# PostgreSQL
pip install psycopg2-binary
# MongoDB
pip install pymongo
Solution: Install full version:
1
pip install slm-builder[full]
Solution: Use CPU or reduce batch size:
1
slm build --source data.csv --device cpu
Or in config:
1
2
training :
batch_size : 4 # Reduce from default 8
peft not available
Solution: Install full version with PEFT:
1
pip install slm-builder[full]
1
2
# For M1/M2 Macs, use MPS acceleration
pip install slm-builder[full]
SLM Builder will automatically detect and use MPS when available.
1
pip install slm-builder[full]
For best performance on Windows, use WSL2 with Ubuntu.
1
2
# Standard installation works well
pip install slm-builder[full]
1
2
3
4
5
# CPU version
docker pull slmbuilder/slm-builder:latest
# GPU version (requires nvidia-docker)
docker pull slmbuilder/slm-builder:gpu
After installation, check out: