No module named 'llama_index'

[avianinc]

When I try and Install a pipline in open-webui from a Github URL I get the error No module named 'llama_index'.

Its interesting to note that when I try and add the pipeline docker image to a docker compose file I recieve the same error during the build and the container will not start. Here is the log output.

INFO:     Started server process [7]
pipelines-1    | INFO:     Waiting for application startup.
pipelines-1    | ERROR:    Traceback (most recent call last):
pipelines-1    |   File "/usr/local/lib/python3.11/site-packages/starlette/routing.py", line 732, in lifespan
pipelines-1    |     async with self.lifespan_context(app) as maybe_state:
pipelines-1    |   File "/usr/local/lib/python3.11/contextlib.py", line 210, in __aenter__
pipelines-1    |     return await anext(self.gen)
pipelines-1    |            ^^^^^^^^^^^^^^^^^^^^^
pipelines-1    |   File "/app/main.py", line 210, in lifespan
pipelines-1    |     await on_startup()
pipelines-1    |   File "/app/main.py", line 189, in on_startup
pipelines-1    |     await module.on_startup()
pipelines-1    |   File "/app/./pipelines/llamaindex_ollama_github_pipeline.py", line 23, in on_startup
pipelines-1    |     from llama_index.embeddings.ollama import OllamaEmbedding
pipelines-1    | ModuleNotFoundError: No module named 'llama_index'

If I run the docker compose with just ollama and open-webui images the compose starts fine. I can then run the pipelines as an individial container and runs. I can add the pipelines containter address to the OpenAPI URL and it sees it..

Any thoughts?

Here is the compose file that does not work

services:
  ollama:
    image: ollama/ollama:latest
    deploy:
      resources:
        reservations:
          devices:
            - capabilities: [gpu]
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    volumes:
      - ollama:/root/.ollama
    ports:
      - "11434:11434"
    runtime: nvidia

  open_web_ui:
    image: dyrnq/open-webui
    deploy:
      resources:
        reservations:
          devices:
            - capabilities: [gpu]
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    volumes:
      - open-webui:/app/backend/data
    ports:
      - "3000:8080"
    extra_hosts:
      - "host.docker.internal:host-gateway"
    restart: always
    runtime: nvidia

  pipelines:
    image: ghcr.io/open-webui/pipelines:main
    deploy:
      resources:
        limits:
          memory: 4g
          cpus: '2.0'
        reservations:
          devices:
            - capabilities: [gpu]
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    volumes:
      - pipelines:/app/pipelines
    ports:
      - "9099:9099"
    extra_hosts:
      - "host.docker.internal:host-gateway"
    restart: always
    runtime: nvidia

volumes:
  ollama:
  open-webui:
  pipelines:

[justinh-rahb]

For reasons the UI pipeline installer from URL does not run the installation of dependencies. You should use the envvar PIPELINES_URLS in your compose/docker run to make sure that pipelines with additional depends will work:

  pipelines:
    image: ghcr.io/open-webui/pipelines:main
    deploy:
      resources:
        limits:
          memory: 4g
          cpus: '2.0'
        reservations:
          devices:
            - capabilities: [gpu]
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
      - PIPELINES_URLS="https://raw.githubusercontent.com/open-webui/pipelines/main/examples/pipelines/rag/llamaindex_ollama_github_pipeline.py"
    volumes:
      - pipelines:/app/pipelines
    ports:
      - "9099:9099"
    extra_hosts:
      - "host.docker.internal:host-gateway"
    restart: always
    runtime: nvidia

[thiswillbeyourgithub]

Related to #239
Curious about your thoughts on this

[justinh-rahb]

It might also be necessary to delete your pipelines volume and start fresh.

[avianinc]

Thanks for the feedback... it's interesting I can't seem to get this to load up via docker compose. If I make a separate docker container its runs and communicates fine. I'll be working on this over the weekend, and I'll leave my notes here.

[musicheardworldwide]

I changed my dockerfile to download the dependencies that weren't loading right to the end of the file and it fixed all that little stuff, as well as the ./data file, I also added that to the end of my dockerfile

# Install Python dependencies
COPY ./requirements.txt .
RUN pip3 install uv && \
    if [ "$USE_CUDA" = "true" ]; then \
    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/$USE_CUDA_DOCKER_VER --no-cache-dir && \
    pip3 install -r requirements.txt --no-cache-dir; \
    else \
    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu --no-cache-dir && \
    pip3 install -r requirements.txt --no-cache-dir; \
    fi

# Copy the application code
COPY . .

# Ensure the data directory is created
RUN mkdir -p /app/data
RUN pip3 install aiohttp
RUN pip3 install llama-index

# Expose the port
ENV HOST="0.0.0.0"
ENV PORT="9099"

ENTRYPOINT [ "bash", "start.sh" ]

[atgehrhardt]

Wait so how does it work be default?

UI installation: No external dependencies installed
Docker envvar: Installs dependencies

If this is the case, is it automatically installing dependencies into a python venv or conda env? Or does it just install all dependencies globally?

I'd assume it's not installing globally because that could cause major issues.

Sorry if this is obvious. I was reviewing the code, but couldn't determine how this works.