Hey there! We've found the following might help you get unblocked faster:

• 🧵 How can I increase my disk space on Hobby plan?

• 🧵 Increase disk size beyond 250GB

• 🧵 URGENT: Database crashed - No space left on device - Production down

• 🧵 Postgres Chewing Up Disk Space

If you find the answer from one of these, please let us know by solving the thread!

The default disk size for hobby is 5gb, and it's 50gb for pro. Luckily, since you're on pro plan you can increase your disk space by clicking on the volume within the service and growing it to the size you need, max size on pro plan is 250gb.

Read More:

• https://docs.railway.com/reference/volumes

• https://docs.railway.com/overview/the-basics#volume-settings

Thank you for the information!

My ollama service takes a lot of time to respond. The instance I have locally (my laptop) goes a lot faster. Could you help me out? What is the issue that it takes so much time?

Now it turns out it does not even accept connections from the web service :S

Railway doesn't have GPUs which is important for making AI models run fast, your laptop has a GPU (either integrated or dedicated) which makes it much faster to run AI models on ollama

I have a template for ollama (https://railway.com/deploy/ollama-api) and more specifically nomic text embed (I'm assuming that's what you meant with nomadic-text) (https://railway.com/deploy/nomic-embed) if you need them. They should be set up to expose authenticated endpoints for your ollama service automatically, this is assuming you're using it as an API anyway

If you prefer to use your existing service then I can try to help fix the issue, what exact error are you getting?

How can I solve my problem then? Do you have a solution for me? I thought that when we install your ollama service that it would run on GPUs.

Thank you for your reply. Yes nomic-embed-text. I'm sorry about the mistake.

I need an ollama service with both these models installed. I was able to install them in my current ollama version via ssh (after increasing disk size)

models needed:
nomic-embed-text:latest

llama2:latest

I have a few questions more:

will your template run in GPUs?
what should I do so that my web instance is able to connect to the ollama service?

I thought that when we install your ollama service that it would run on GPUs

Unfortunately not, Railway has no GPUs

If GPUs are you important to you then there isn't much we can do

I need an ollama service with both these models installed. I was able to install them in my current ollama version via ssh (after increasing disk size)

Oh alright, that's fair, quick tip - you can download models via API request instead of SSHing into the container

curl https://YOUR_DOMAIN.up.railway.app/api/pull -d '{
  "model": "llama3.2"
}'

will your template run in GPUs

I wish, but no - Railway doesn't have GPUs

what should I do so that my web instance is able to connect to the ollama service

You just need to expose the ollama service publicly by giving it a domain if it doesnt have one already, you can then use that API to communicate with your ollama service from your website. Heads up, though, ollama has no authentication by default (that's actually why i built that template of mine, it packages auth with ollama)- you might need to build an authentication layer that works for your use-case in order to ensure someone doesnt hog your ollama deployment 100% of the time. Something like a backend that authenticates oauth so people can login and then based on that provide access, or whatever you actually need.

Going open ai api. I would have preferred to have my own instance. But I can move things around and then try to deploy something for me and train it for my data.

I am not sure if I was able to explain myself correctly. The mobile app uses web service (api) which uses the ollama instance. So between web and ollama services an internal private work suffice and should already be working by default. Is it? No need for public access to ollama.

Oh yea that works, in this case the api is the layer I was refering to. The private network is indeed fine for that, just make sure to use http and specify the port within the url like: http://ollama.railway.internal:8080 for example

that's fair

so all clear to mark this as solved then or do you have more questions?

all good. thank you for your help.