Closed bibekyess closed 7 months ago
I am also facing similar issue
Traceback (most recent call last):
File "/home/ubuntu/miniconda3/envs/muti-tenant-test-1/lib/python3.8/runpy.py", line 185, in _run_module_as_main
mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
File "/home/ubuntu/miniconda3/envs/muti-tenant-test-1/lib/python3.8/runpy.py", line 111, in _get_module_details
import(pkg_name)
File "/home/ubuntu/multi-tenant-test/punica/punica/init.py", line 1, in
We follow PyTorch's default environment, which is CUDA 12.1 now. Other CUDA versions should also work, but I haven't tested. I'll add CI tests once I got time.
But I think @bibekyess you have successfully built the package. Not sure why this happens... Can you tell me which container image you are using?
Ahh Ok. I tried with two images: nvidia/cuda:11.8.0-devel-ubuntu20.04
(above log was with this) and nvidia/cuda:12.0.0-devel-ubuntu20.04
. In both, the build was successfull but I got the error.
@abcdabcd987 I get import punica.ops._kernels as _kernels ModuleNotFoundError: No module named 'punica.ops._kernels'
when running python -m punica.utils.convert_lora_weight model/gsm8k-r16/adapter_model.bin model/gsm8k-r16.punica.pt
@abcdabcd987 Additional information: I tried with nvidia/cuda:12.1.1-devel-ubuntu20.04
image and also used switch-cuda
to try on my local machine but still I am getting the same error. I am using A100-SXM4-40GB
GPU. Thank you :)
Ha I see what is going on here. You are running in the project root directory, which has a punica
folder. In this case, import punica
will import from this folder. I'll improve the project directory structure to avoid this issue.
For now, @bibekyess @luciferlinx101 can you try with editable installation (-e
)?
env TORCH_CUDA_ARCH_LIST="8.0" pip install -v -e .
@abcdabcd987 Thank you! It gets solved. :)
Hello, I am trying to run
punica
in cuda-toolkit-11.8 but I get this errorModuleNotFoundError: No module named 'punica.ops._kernels'
, when running:python -m benchmarks.bench_textgen_lora --system punica --batch-size 32
.The build seems successful except one warning:
The detailed log is this when running
env TORCH_CUDA_ARCH_LIST="8.0" pip install -v --no-build-isolation
: (I tried running inside the docker container and also outside. In both cases, I get the ModuleNotFoundError.)Can you inform the suggested cudatookit-version for building? Thank you!