Troubleshooting and FAQ

Common obstacles

502 Bad Gateway

We have just updated one of your resources, which is about to restart. It should be available after some seconds; hold tight and regularly refresh the page.

Cannot connect via ssh

Check if you are connected to a BFH network locally or via VPN.

Switch Login failed

Check if you are using the right URL: https://inference.mlmp.ti.bfh.ch/ and not other one (e.g. https://beechat.ti.bfh.ch/)

Your session has expired or the token is invalid. Please sign in again.

Tokens are configured to expire after four weeks. Please generate a new token and update your code accordingly.

HTTP 429: Too many requests

You exceeded the currently defined rate limits. Decrease context size or send fewer requests; the limit resets every minute.

Inference

Why is inference so slow?

There might be a lot of concurrent requests coming in at the moment; feel free to give us a heads-up so we can optimize resource allocation.

Are there any token or request limits?

Not at the moment. We plan to configure rate limits soon but have not yet settled on sensible values (e.g., requests per minute or hour). When limits are introduced, exceeding them will return an error response — design your client with retry logic in mind.

Do I have to pay for using the service?

Currently, using the inference service is free of charge.

Can I use the inference service if I am not a BFH member (student or staff)?

For generating an access token, we require you to be associated with BFH (employee or student). Authentication is based on Switch Edu-ID.

Can I use the inference service from outside the BFH network?

Yes, just replace the endpoint with https://inference.proxy.ti.bfh.ch/api/v1/

Are there rate limits, e.g. requests per minute or hour?

Not currently. We plan to introduce rate limits as soon as possible but have not yet determined appropriate values. We will announce changes when they take effect.

What happens when rate limits are exceeded?

Requests that exceed the limit receive an error response from the API.

How is availability guaranteed?

The inference service is provided on a best-effort basis. There is no formal SLA. Check the system status page for current outages.

Are there scheduled maintenance windows when the API may be unavailable?

We may perform maintenance on Friday mornings. The service may be briefly unavailable during these windows. Downtime and breaking changes are announced in the MS Teams channel.

How is access managed — personal tokens or service accounts?

Access is via personal API tokens. Sign in with your Switch Edu-ID at inference.mlmp.ti.bfh.ch, then create a token under Settings → Account → API keys. Tokens expire after four weeks; generate a new one and update your application before they lapse.

Can you see my data?

No, we generally do not view, store, or log your requests and associated responses. However, to fulfill our obligation to ensure compliant use of our service, we might manually analyze requests and responses. You will be informed.