Skip to content

RAPIDS Integration¶

Use RAPIDS cuDF/cuGraph on GPU 1 alongside LLM on GPU 0.

Overview¶

RAPIDS provides GPU-accelerated dataframes and graph analytics.

Installation¶

pip install cudf-cu12 cugraph-cu12 graphistry

Basic Usage¶

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "1"  # Use GPU 1

import cudf
import cugraph

# Create graph
df = cudf.DataFrame({"src": [0, 1, 2], "dst": [1, 2, 0]})
G = cugraph.Graph()
G.from_cudf_edgelist(df, source="src", destination="dst")

With LLM¶

# GPU 0: LLM
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
from llcuda.server import ServerManager
server = ServerManager()

# GPU 1: RAPIDS
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
import cudf
# Process data on GPU 1

See Also¶