remzi@sparkai:~/dgx-spark-playbooks/dgx-spark-playbooks/nvidia/txt2kg/assets$ docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
d513d9ffa0ea compose-app “docker-entrypoint.s…” 6 hours ago Up 3 hours 0.0.0.0:3001->3000/tcp, [::]:3001->3000/tcp compose-app-1
e398d0a7150b compose-vllm “/opt/nvidia/nvidia_…” 6 hours ago Restarting (1) 1 second ago vllm-service
37567d7e61a1 compose-sentence-transformers “gunicorn --bind 0.0…” 6 hours ago Up 3 hours 0.0.0.0:8000->80/tcp, [::]:8000->80/tcp compose-sentence-transformers-1
209d2704645c Package pinecone-index · GitHub “/engine” 6 hours ago Restarting (255) 8 seconds ago entity-embeddings
47af05ab9967 ollama-custom:latest “/entrypoint.sh” 5 days ago Up 3 hours (unhealthy) 0.0.0.0:11434->11434/tcp, [::]:11434->11434/tcp ollama-compose
8b17c523f2c4 arangodb:latest “/entrypoint.sh aran…” 5 days ago Up 3 hours 0.0.0.0:8529->8529/tcp, [::]:8529->8529/tcp compose-arangodb-1
Hi @cirit , did you get pinecone to play nice? I just ran the startup, most comes up, pinecone keeps rebooting. Most what I found so far is “exec /engine: exec format error” so it appears the pinecone container is not for ARM, would be weird to have it in the DGX/Arm playbook right?
OK Thanks. Never mind, there just is no pinecone Docker for ARM and no progress.
The TXT2KG playbook is not compatible with DGX, shame it is in the DGX playbooks repo. Maybe it can work with another vector database, will look into it.
Hi! Creator of txt2kg here. I was waiting to get the latest code pushed and it went out last week. Please try it again.
@nvidia3869 is correct that Pinecone is not supported on ARM so I switched to Qdrant for the vector database. Please feel free to create a github issue and I’ll follow up on it.
I can create draft PRs for the fixes and then we will periodically update the official repo.
It does not make sense to only modify docker-compose.yml and not modify descriptions in the accompanying REAMD.md files
Reusing existing playbooks is fine with me, I just wish you would test them against DGX-spark before publishing it in the DGX-spark playbook.
Please do remember that the DGX-spark playbook is a selling point for DGX-spark and referred to as such by Nvidia marketing. Those playbooks should therefor be accurate and complete unless you want to discourage prospective and actual owners or DGX-spark.
Alas… Docker’s ollama-custom:latest image is not available (updates 18hrs ago) but should work with previously built version.
In case anyone would like to try bleeding edge:
modify the compose file to say ollama/ollama:latest instead of docker-custom:latest
Sorry to say but the txt2kg playbook has become a bit of a mess:
pinecone / qdrant need not be in docker-compose.yml
qdrant is pulled nor built using start.sh
qdrant is not in docker-compose.complete.yml
qdrant is pulled nor built using start.sh –complete
start.sh does not start qdrant as it is optional
start.sh –complete does not start qdrant as it was not built
Dockerfile does include pinecone, not qdrant
So setup fails. This needs a bit more work, I hope someone feels responsible to put in the work.
I’d suggest taking the playbook offline, correcting errors and test functionality before putting it online again. I’ve been through one too many “fixed it please try and let me know” loops in my life thanks.
