llama.cpp Guidance

The llama-cpp-guidance package provides an LLM client compatibility layer between llama-cpp-python and guidance.

Installation

The llama-cpp-guidance package can be installed using pip.

pip install llama-cpp-guidance

⚠️ It is highly recommended that you follow the installation instructions for llama-cpp-python after installing llama-cpp-guidance to ensure that you have hardware acceleration setup appropriately.

Basic Usage

Once installed, you can use the LlamaCpp class like any other guidance-compatible LLM class.

from pathlib import Path
from llama_cpp_guidance.llm import LlamaCpp
import guidance

guidance.llm = LlamaCpp(
    model_path=Path("../path/to/llamacpp/model.gguf"),
    n_gpu_layers=1,
    n_threads=8
)

program = guidance(
    "The best thing about the beach is {{~gen 'best' temperature=0.7 max_tokens=10}}"
)
output = program()
print(output)

The best thing about the beach is that there’s always something to do.