Afterwards, I copied the resulting bindings*.so into tensorrt_llm's directory inside the dist-packages dir -- according to the dockerfile. Then I followed it to install nvidia-ammo 0.3, then added the optimum-nvidia dir to python path.
I also went into optimum-nvidia directory, and ran pip install -e ., so that in my environment, when using pip list | grep optimum I could get:
However, I still could not import optimum.nvidia properly, while it's okay to import tensorrt_llm and tensorrt_llm.bindings.
>>> from optimum.nvidia.pipelines import pipeline
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'optimum.nvidia'
>>>
Could someone please help me on how to install optimum nvidia properly without building a new image or pulling from dockerhub?
It's quite hard for me to build a docker image, so I started from a docker environment with TensorRT LLM 0.6.1 inside.
I checked your dockerfile, followed the process, and built TensorRT LLM using (I am using 4090 so that cuda arch is 89):
Afterwards, I copied the resulting bindings*.so into tensorrt_llm's directory inside the dist-packages dir -- according to the dockerfile. Then I followed it to install nvidia-ammo 0.3, then added the optimum-nvidia dir to python path.
I also went into optimum-nvidia directory, and ran
pip install -e .
, so that in my environment, when usingpip list | grep optimum
I could get:However, I still could not import optimum.nvidia properly, while it's okay to
import tensorrt_llm
andtensorrt_llm.bindings
.Could someone please help me on how to install optimum nvidia properly without building a new image or pulling from dockerhub?
Thank you!