You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I downloaded rapids-24.12 via conda and start a conda environment to running Cugraph.
I was able to run single GPU algorithm, build graph and generate graph. However, when I tried to explore multi-gpu usage in cugraph, I receive keyError when I tried to build multi-gpu graph by using from_dask_cudf_edgelist and cugraph.generators.rmat by setting mg = True and set up multi gpu cluster and client
cluster = LocalCUDACluster()
client = Client(cluster)
Comms.initialize(p2p=True) # multi gpu communication .
I guess maybe some GPU setup was incorrect so I ran ./build.sh and shows error as well.
Could you please tell me how to fix the multi-gpu error and if that is relates to my build fail?
Thanks
Minimum reproducible example
import dask_cudf
from dask.distributed import Client
from dask_cuda import LocalCUDACluster
import cudf
import cugraph
import cugraph.dask.comms.comms as Comms
from cugraph.generators.rmat import rmat
# Set up a Dask CUDA cluster for multi-GPU processing
cluster = LocalCUDACluster()
client = Client(cluster)
Comms.initialize(p2p=True) # multi gpu communication
import pandas as pd
import dask.dataframe as dd
e_list = pd.read_csv(
'rmat.csv'
)
print(e_list.head())
e_list_cudf = dd.from_pandas(e_list)
e_list_cudf = e_list_cudf.to_backend("cudf")
G = cugraph.Graph(directed=True)
G.from_dask_cudf_edgelist(e_list_cudf, source='src', destination='dst')
graph = cugraph.generators.rmat(
scale=scale,
num_edges=num_edges,
a=0.57,
b=0.19,
c=0.19,
seed=42,
clip_and_flip=False,
scramble_vertex_ids=False,
include_edge_weights=False,
dtype="int32",
create_using=cugraph.Graph,
mg=True,
)
These two calls will occur KeyError('handle').
Relevant log output
For multi-gpu the erros shows below:
`G.from_dask_cudf_edgelist(e_list_cudf, source='src', destination='dst')` or
`graph = cugraph.generators.rmat( scale=scale, num_edges=num_edges, a=0.57, b=0.19, c=0.19, seed=42, clip_and_flip=False, scramble_vertex_ids=False, include_edge_weights=False, dtype="int32", create_using=cugraph.Graph, mg=True, )`
Exception: "KeyError('handle')"
Traceback: ' File "/home/user/.conda/envs/rapids-24.12/lib/python3.10/site-packages/cugraph/structure/graph_implementation/simpleDistributedGraph.py", line 139, in _make_plc_graph\n resource_handle=ResourceHandle(Comms.get_handle(sID).getHandle()),\n File "/home/user/.conda/envs/rapids-24.12/lib/python3.10/site-packages/cugraph/dask/comms/comms.py", line 238, in get_handle\n./build.sh libcugraph pylibcugraph cugraph --skip_cpp_tests will output below error: Building for the architecture of the GPU in the system...CMake Error at CMakeLists.txt:19 (include): include could not find requested file: rapids-cmakeCMake Error at CMakeLists.txt:20 (include): include could not find requested file: rapids-cpmCMake Error at CMakeLists.txt:21 (include): include could not find requested file: rapids-cudaCMake Error at CMakeLists.txt:22 (include): include could not find requested file: rapids-exportCMake Error at CMakeLists.txt:23 (include): include could not find requested file: rapids-findCMake Error at CMakeLists.txt:25 (rapids_cuda_init_architectures): Unknown CMake command "rapids_cuda_init_architectures".-- Configuring incomplete, errors occurred!after I manually download rapids-cmake from git and update the include path like:`include(~/cugraph/rapids_config.cmake)include(~/cugraph/rapids-cmake/rapids-cmake/rapids-cmake.cmake)include(~/cugraph/rapids-cmake/rapids-cmake/rapids-cpm.cmake)include(~/cugraph/rapids-cmake/rapids-cmake/rapids-cuda.cmake)include(~/cugraph/rapids-cmake/rapids-cmake/rapids-export.cmake)include(~/cugraph/rapids-cmake/rapids-cmake/rapids-find.cmake)`, the error becomes:Building for the architecture of the GPU in the system...CMake Error: CMake was unable to find a build program corresponding to "Ninja". CMAKE_MAKE_PROGRAM is not set. You probably need to select a different build tool.CMake Error: CMAKE_C_COMPILER not set, after EnableLanguageCMake Error: CMAKE_CXX_COMPILER not set, after EnableLanguageCMake Error: CMAKE_CUDA_COMPILER not set, after EnableLanguage-- Configuring incomplete, errors occurred!
Version
24.12
Which installation method(s) does this occur on?
Conda
Describe the bug.
I downloaded rapids-24.12 via conda and start a conda environment to running Cugraph.
I was able to run single GPU algorithm, build graph and generate graph. However, when I tried to explore multi-gpu usage in cugraph, I receive keyError when I tried to build multi-gpu graph by using
from_dask_cudf_edgelist
andcugraph.generators.rmat
by setting mg = True and set up multi gpu cluster and clientBesides multi-gpu in cugraph, I was not able to use
cudf.read_csv()
with internal error described in https://community.deeplearning.ai/t/has-anyone-gotten-cudf-read-csv-to-work/558425 so I use pandas to convert to cudf for input offrom_dask_cudf_edgelist
.I guess maybe some GPU setup was incorrect so I ran
./build.sh
and shows error as well.Could you please tell me how to fix the multi-gpu error and if that is relates to my build fail?
Thanks
Minimum reproducible example
Relevant log output
Environment details
Other/Misc.
No response
Code of Conduct
The text was updated successfully, but these errors were encountered: