TensorRT LLM Docker Issues

[ashish007git]

I am facing multiple issues while trying to build the docker container by following the steps in here

No internet in the docker container.

Running the following commands

make -C docker build

or

make -C docker run

gave errors relating to internet connection failures. The docker container did not have an active internet connection. I was able to fix them using the following fix below:

@@ -22,7 +23,7 @@ STAGE              ?=
 # Set this to define a custom image name and tag
 IMAGE_WITH_TAG     ?= $(IMAGE_NAME)$(if $(STAGE),/$(STAGE)):$(IMAGE_TAG)
 DOCKER_BUILD_OPTS  ?= --pull
-DOCKER_BUILD_ARGS  ?=
+DOCKER_BUILD_ARGS  ?= --network host
 DOCKER_PROGRESS    ?= auto
 CUDA_ARCHS         ?=
 BUILD_WHEEL_ARGS   ?= $(shell grep 'ARG BUILD_WHEEL_ARGS=' Dockerfile.multi | grep -o '=.*' | tr -d '="')$(if $(CUDA_ARCHS), --cuda_architectures $(CUDA_ARCHS))
@@ -63,7 +64,7 @@ endef
        docker push $(IMAGE_WITH_TAG)$(IMAGE_TAG_SUFFIX)
 
 DOCKER_RUN_OPTS   ?= --rm -it --ipc=host --ulimit memlock=-1 --ulimit stack=67108864
-DOCKER_RUN_ARGS   ?=
+DOCKER_RUN_ARGS   ?= --net host
 GPU_OPTS          ?= --gpus=all
 SOURCE_DIR        ?= $(shell readlink -f ..)
 CODE_DIR          ?= /code/tensorrt_llm

I suspected that this could be an error pertaining to my VM instance, or issue with docker installation. However, I was able to install and build other docker images without any problems.

Building docker image fails

Creating the docker image with the previous internet fix gave me the error shown in the image below:

The build command fails to build mpi4py.

Any leads on how to resolve these issues are appreciated!

[robosina]

@ashish007git I believe your problem has been fixed in #447

you just need to use

pip install git+https://github.com/Shixiaowei02/mpi4py.git@fix-setuptools-version

instead of pip install mpi4py in docker/common/install_base.sh