Docker image fails to build: "List directory /var/lib/apt/lists/partial is missing."

[CaseyLabs]

Primary error message

E: List directory /var/lib/apt/lists/partial is missing. - Acquire (13: Permission denied)

Commands executed

git clone https://github.com/slashml/amd-gpu-inference.git
cd amd-gpu-inference
./run-docker-amd.sh "meta-llama/Llama-2-7b-chat-hf" "Translate the following English text to French: 'Hello, how are you?'"

Full Output

Building Docker image...
[+] Building 25.3s (4/12)                                                                                                                                                           docker:default
[+] Building 91.4s (6/12)                                                                                                                                                           docker:default
 => [internal] load build definition from Dockerfile                                                                                                                                          0.0s
 => => transferring dockerfile: 886B                                                                                                                                                          0.0s
 => [internal] load metadata for docker.io/rocm/rocm-terminal:5.4.2                                                                                                                           1.0s
 => [internal] load .dockerignore                                                                                                                                                             0.0s
 => => transferring context: 2B                                                                                                                                                               0.0s
 => [1/8] FROM docker.io/rocm/rocm-terminal:5.4.2@sha256:2fa4d31e18bfd8e91a353c114cfcbaea2984d0fb1727e64a5b0afd5e7b93eb63                                                                    90.2s
 => => resolve docker.io/rocm/rocm-terminal:5.4.2@sha256:2fa4d31e18bfd8e91a353c114cfcbaea2984d0fb1727e64a5b0afd5e7b93eb63                                                                     0.0s
 => => sha256:71b27aa37b922c5d8af1f1da6e55b5783d3530026bb40107bc146d6cc3caa917 4.54kB / 4.54kB                                                                                                0.0s
 => => sha256:846c0b181fff0c667d9444f8378e8fcfa13116da8d308bf21673f7e4bea8d580 28.58MB / 28.58MB                                                                                              1.1s
 => => sha256:1c1ec7303bc635626bae18ee15f655e3d29e44f7f9d11dbba3d4530c47f16fd8 1.73GB / 1.73GB                                                                                               57.6s
 => => sha256:da1fd9caaff1498af47a868c962080bf868e48876b80a621d185fc826a1d6cc0 265B / 265B                                                                                                    0.4s
 => => sha256:2fa4d31e18bfd8e91a353c114cfcbaea2984d0fb1727e64a5b0afd5e7b93eb63 1.16kB / 1.16kB                                                                                                0.0s
 => => sha256:a34f2e2b5e960d56c5c139af462b9c5a8c00c2a3c09a1734bdb6643e59068d6e 4.36kB / 4.36kB                                                                                                0.6s
 => => extracting sha256:846c0b181fff0c667d9444f8378e8fcfa13116da8d308bf21673f7e4bea8d580                                                                                                     0.6s
 => => extracting sha256:1c1ec7303bc635626bae18ee15f655e3d29e44f7f9d11dbba3d4530c47f16fd8                                                                                                    31.4s
 => => extracting sha256:da1fd9caaff1498af47a868c962080bf868e48876b80a621d185fc826a1d6cc0                                                                                                     0.0s
 => => extracting sha256:a34f2e2b5e960d56c5c139af462b9c5a8c00c2a3c09a1734bdb6643e59068d6e                                                                                                     0.0s
 => [internal] load build context                                                                                                                                                             0.0s
 => => transferring context: 7.91kB                                                                                                                                                           0.0s
 => ERROR [2/8] RUN apt-get update && apt-get install -y     python3-pip     python3-dev     git     && rm -rf /var/lib/apt/lists/*                                                           0.2s
------
 > [2/8] RUN apt-get update && apt-get install -y     python3-pip     python3-dev     git     && rm -rf /var/lib/apt/lists/*:
0.111 Reading package lists...
0.119 E: List directory /var/lib/apt/lists/partial is missing. - Acquire (13: Permission denied)
------
Dockerfile:9
--------------------
   8 |     # Install system dependencies
   9 | >>> RUN apt-get update && apt-get install -y \
  10 | >>>     python3-pip \
  11 | >>>     python3-dev \
  12 | >>>     git \
  13 | >>>     && rm -rf /var/lib/apt/lists/*
  14 |     
--------------------
ERROR: failed to solve: process "/bin/sh -c apt-get update && apt-get install -y     python3-pip     python3-dev     git     && rm -rf /var/lib/apt/lists/*" did not complete successfully: exit code: 100

[CaseyLabs]

Has this project been tested thoroughly? The issue here is that:

• rocm-terminal is being used as a base image
• That base image runs as a non-root user (rocm-user)
• And therefore that user can't install any system packages.

The quick workaround is to add a USER 0 before the system package installs, and then to switch back to rocm-user afterwards.

Was this project generated by AI?

[eff-kay]

Has this project been tested thoroughly? The issue here is that:

• rocm-terminal is being used as a base image
• That base image runs as a non-root user (rocm-user)
• And therefore that user can't install any system packages.

The quick workaround is to add a USER 0 before the system package installs, and then to switch back to rocm-user afterwards.

Was this project generated by AI?

Yes, sorry about that, it was a small fix prevent security issues, forgot to update the rest.

[CaseyLabs]

In the PR, instead of using:

FROM rocm/dev-ubuntu-22.04:6.0-complete

Why not use Ubuntu 24.04 + ROCm 6.2? ROCm 6.2 offers significant AMD GPU support improvements.

Also, the PR still doesn't build:

=> [2/8] RUN apt-get update && apt-get install -y     python3-pip      8.4s
 => [3/8] RUN useradd -m -s /bin/bash user                              0.2s
 => [4/8] WORKDIR /home/user/app                                        0.0s
 => [5/8] RUN python3 -m pip install --user --upgrade pip               3.2s
 => [6/8] COPY --chown=user:user requirements.txt .                     0.0s
 => ERROR [7/8] RUN pip install --user --no-cache-dir -r requirements.  1.5s
------
 > [7/8] RUN pip install --user --no-cache-dir -r requirements.txt:
0.346 Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/rocm6.2
1.135 ERROR: Could not find a version that satisfies the requirement torch==2.2.0+rocm6.2 (from versions: 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 2.0.0, 2.0.1, 2.1.0, 2.1.1, 2.1.2, 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.4.0, 2.4.1)
1.420 ERROR: No matching distribution found for torch==2.2.0+rocm6.2
------
Dockerfile:26
--------------------
  24 |     COPY --chown=user:user requirements.txt .
  25 |     
  26 | >>> RUN pip install --user --no-cache-dir -r requirements.txt
  27 |     
  28 |     # Copy the application code
--------------------
ERROR: failed to solve: process "/bin/sh -c pip install --user --no-cache-dir -r requirements.txt" did not complete successfully: exit code: 1

Make sure you do an uncached Docker build when testing:

docker build --no-cache -t amd-gpu-inference .

[eff-kay]

In the PR, instead of using:

FROM rocm/dev-ubuntu-22.04:6.0-complete

Why not use Ubuntu 24.04 + ROCm 6.2? ROCm 6.2 offers significant AMD GPU support improvements.

Also, the PR still doesn't build:

=> [2/8] RUN apt-get update && apt-get install -y     python3-pip      8.4s
 => [3/8] RUN useradd -m -s /bin/bash user                              0.2s
 => [4/8] WORKDIR /home/user/app                                        0.0s
 => [5/8] RUN python3 -m pip install --user --upgrade pip               3.2s
 => [6/8] COPY --chown=user:user requirements.txt .                     0.0s
 => ERROR [7/8] RUN pip install --user --no-cache-dir -r requirements.  1.5s
------
 > [7/8] RUN pip install --user --no-cache-dir -r requirements.txt:
0.346 Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/rocm6.2
1.135 ERROR: Could not find a version that satisfies the requirement torch==2.2.0+rocm6.2 (from versions: 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 2.0.0, 2.0.1, 2.1.0, 2.1.1, 2.1.2, 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.4.0, 2.4.1)
1.420 ERROR: No matching distribution found for torch==2.2.0+rocm6.2
------
Dockerfile:26
--------------------
  24 |     COPY --chown=user:user requirements.txt .
  25 |     
  26 | >>> RUN pip install --user --no-cache-dir -r requirements.txt
  27 |     
  28 |     # Copy the application code
--------------------
ERROR: failed to solve: process "/bin/sh -c pip install --user --no-cache-dir -r requirements.txt" did not complete successfully: exit code: 1

Make sure you do an uncached Docker build when testing:

docker build --no-cache -t amd-gpu-inference .

I think Ubuntu 24, doesn't support ROCm.
https://askubuntu.com/questions/1517236/rocm-not-working-on-ubuntu-24-04-desktop

Will test with 24 though.

[eff-kay]

In the PR, instead of using:
FROM rocm/dev-ubuntu-22.04:6.0-complete
Why not use Ubuntu 24.04 + ROCm 6.2? ROCm 6.2 offers significant AMD GPU support improvements.
Also, the PR still doesn't build:

=> [2/8] RUN apt-get update && apt-get install -y     python3-pip      8.4s
 => [3/8] RUN useradd -m -s /bin/bash user                              0.2s
 => [4/8] WORKDIR /home/user/app                                        0.0s
 => [5/8] RUN python3 -m pip install --user --upgrade pip               3.2s
 => [6/8] COPY --chown=user:user requirements.txt .                     0.0s
 => ERROR [7/8] RUN pip install --user --no-cache-dir -r requirements.  1.5s
------
 > [7/8] RUN pip install --user --no-cache-dir -r requirements.txt:
0.346 Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/rocm6.2
1.135 ERROR: Could not find a version that satisfies the requirement torch==2.2.0+rocm6.2 (from versions: 1.11.0, 1.12.0, 1.12.1, 1.13.0, 1.13.1, 2.0.0, 2.0.1, 2.1.0, 2.1.1, 2.1.2, 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.4.0, 2.4.1)
1.420 ERROR: No matching distribution found for torch==2.2.0+rocm6.2
------
Dockerfile:26
--------------------
  24 |     COPY --chown=user:user requirements.txt .
  25 |     
  26 | >>> RUN pip install --user --no-cache-dir -r requirements.txt
  27 |     
  28 |     # Copy the application code
--------------------
ERROR: failed to solve: process "/bin/sh -c pip install --user --no-cache-dir -r requirements.txt" did not complete successfully: exit code: 1

Make sure you do an uncached Docker build when testing:
docker build --no-cache -t amd-gpu-inference .

I think Ubuntu 24, doesn't support ROCm. https://askubuntu.com/questions/1517236/rocm-not-working-on-ubuntu-24-04-desktop

Will test with 24 though.

Nvm, it doesnt for sure

https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/docker-image-support-matrix.html

[CaseyLabs]

To clarify:

• Ubuntu 24.04 does support ROCm 6.2:
  https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/native-install/ubuntu.html

• However, Pytorch at the moment only distributes Docker images for Ubuntu 20.04 and 22.04:
  https://hub.docker.com/r/rocm/pytorch/tags?name=6.2

Have a look at this repo for ideas on how to fix+package your project:

• https://github.com/nktice/AMD-AI

[eff-kay]

To clarify:

• Ubuntu 24.04 does support ROCm 6.2:
  https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/native-install/ubuntu.html
• However, Pytorch at the moment only distributes Docker images for Ubuntu 20.04 and 22.04:
  https://hub.docker.com/r/rocm/pytorch/tags?name=6.2

Have a look at this repo for ideas on how to fix+package your project:

• https://github.com/nktice/AMD-AI

Ok, this helps thanks so much 🙏

[CaseyLabs]

Sounds good. I would recommend making this repo private in the meantime, it's not ready for publication yet.