Just for you, I spun this up on my mac π
Here's the steps
In a fresh terminal
python -m venv venv
source venv/bin/activate
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python --no-cache-dir --force-reinstall
pip install llama-index llama-index-llms-llama-cpp
Then, I ran this code
from llama_index.core import SimpleDirectoryReader, VectorStoreIndex
from llama_index.llms.llama_cpp import LlamaCPP
from llama_index.llms.llama_cpp.llama_utils import (
messages_to_prompt,
completion_to_prompt,
)
llm = LlamaCPP(
# You can pass in the URL to a GGML model to download it automatically
model_url=model_url,
# optionally, you can set the path to a pre-downloaded model instead of model_url
model_path=None,
temperature=0.1,
max_new_tokens=256,
# llama2 has a context window of 4096 tokens, but we set it lower to allow for some wiggle room
context_window=3900,
# kwargs to pass to __call__()
generate_kwargs={},
# kwargs to pass to __init__()
# set to at least 1 to use GPU
model_kwargs={"n_gpu_layers": -1},
# transform inputs into Llama2 format
messages_to_prompt=messages_to_prompt,
completion_to_prompt=completion_to_prompt,
verbose=True,
)
And in the terminal, I see
llm_load_tensors: offloading 40 repeating layers to GPU
llm_load_tensors: offloading non-repeating layers to GPU
llm_load_tensors: offloaded 41/41 layers to GPU
I get about 20 Tokens/sec after testing with a few prompts