Add LLaVA component to components.yml

[ZachCafego]

Issues:

• Create a component that uses LLAVA for extracting text features from imagery openmpf#1836

Related PRs:

• Create LlavaDetection component openmpf-components#392

Please review our Contributor Guide.

---

This change is 

[jrobble]

Reviewable status: 0 of 1 files reviewed, 2 unresolved discussions

---

docker-compose.components.yml line 159 at r1 (raw file):

  
  llava-detection-server:
    <<: *detection-component-base

This is not a component and should not extend from the component base.

---

docker-compose.components.yml line 167 at r1 (raw file):

      memlock: -1
      stack: 67108864
    ports:

This is for exposing ports so that the server can be accessed externally to Docker. Since only your LLAVA component service needs to access it, I think this is unnecessary and should be removed.

[ZachCafego]

Reviewable status: 0 of 1 files reviewed, 2 unresolved discussions (waiting on @jrobble)

---

docker-compose.components.yml line 159 at r1 (raw file):

Previously, jrobble (Jeff Robble) wrote…

This is not a component and should not extend from the component base.

Done.

---

docker-compose.components.yml line 167 at r1 (raw file):

Previously, jrobble (Jeff Robble) wrote…

This is for exposing ports so that the server can be accessed externally to Docker. Since only your LLAVA component service needs to access it, I think this is unnecessary and should be removed.

Done.

[jrobble]

docker-compose.components.yml line 167 at r2 (raw file):

      stack: 67108864
    environment:
      NVIDIA_VISIBLE_DEVICES: all

Making a note to consider adding:

OLLAMA_KEEP_ALIVE=-1

to address losing GPU connectivity when ollama unloads the model after inactivity.

Issue mentioned here: ollama/ollama#6928 (comment)

[jrobble]

docker-compose.components.yml line 159 at r2 (raw file):

  
  llava-detection-server:
    image: ${REGISTRY}ollama_server:${TAG}

Change to openmpf_llava_detection_server