Not working after the update

[mac2net]

Platform

Linux/Fedora

Lemonade Version

Latest

GPU / APU Model

Minisforum um 780

Component

lemond (server)

Bug Description

Nothing is working now and I just sorted out everything with what I am trying to do with Lemonade the day before.

Look IMO, you should do things the normal way - config in /etc/lemonade/, models and other stuff in /var/lib/lemonade/ and get rid of this weird ".cache" stuff which makes no sense.

For me the models disappeared in the left sidebar and now they are there and in the chat box but now I get a "Error preparing model: Origin not allowed" error.

Remember, I didn't change anything. I am only prototyping now with litellm as the gateway, but moving on needs a stable platform.

Steps to Reproduce

Update from <11.8

Expected vs Actual Behavior

It works like it did yesterday

Log Output

Additional Context

No response

[jeremyfowers]

Same problem as #3394 ?

Please put the exact version number from lemonade --version in the issue template.

[mac2net]

I run this from a local hosted Fedora 44 server which I access through an alias in etc/hosts, through the local IP address - both http and port 13305 and via openlitespeed proxy and a cloudflare proxy with a real subdomain and https.

Before the last update everything worked.
Now nothing works

1. With local IP address or alias address, models are listed in chat box and in the sidebar, but don't load with:
   "Error preparing model: Origin not allowed" error or
   "Failed to load model. Origin not allowed."

2. With the real subhuman, models are listed in the chat box but not in the sidebar, so just get "Error preparing model: Origin not allowed"

lemonade --version lemonade version 11.8.1

FYI Ollama is in the Fedora repository. This should be the aspirational goal as it would indicate a certain amount of reliability. Even getting it in a third party repository or rawhide would help.

$dnf info ollama
Updating and loading repositories:
Repositories loaded.
Available packages
Name : ollama
Epoch : 0
Version : 0.12.11
Release : 4.fc44
Architecture : x86_64
Download size : 52.1 MiB
Installed size : 940.9 MiB
Source : ollama-0.12.11-4.fc44.src.rpm
Repository : fedora
Summary : Get up and running with OpenAI gpt-oss, DeepSeek-R1, Gemma 3 and other models
URL : https://github.com/ollama/ollama
License : Apache-2.0 AND BSD-2-Clause AND BSD-3-Clause AND BSD-3-Clause-HP AND BSL-1.0 AND CC-BY-3.0 AND CC-BY-4.0 AN
: D CC0-1.0 AND ISC AND LicenseRef-Fedora-Public-Domain AND LicenseRef-scancode-protobuf AND MIT AND NCSA AND
: NTP AND OpenSSL AND ZPL-2.1 AND Zlib
Description : Get up and running with OpenAI gpt-oss, DeepSeek-R1, Gemma 3 and other models.
Vendor : Fedora Project

[mac2net]

Update: I understand what happened re cors, moving the config - I figured out what to do and it appears all the functionality has returned including dynamically loading a model when in a chat in the Litellm GUI.

I fixed the cors with a gui text editor. The command line failed,

[jeremyfowers]

Got it, you are hitting #3427

@abn please support

[abn]

FYI Ollama is in the Fedora repository. This should be the aspirational goal as it would indicate a certain amount of reliability. Even getting it in a third party repository or rawhide would help.

This might be useful https://copr.fedorainfracloud.org/coprs/abn/lemonade/ (personal, not affiliated with the project officially). If you do migrate to that do check the "Upstream Differences & Migrations" section.

I fixed the cors with a gui text editor. The command line failed,

The fix right now for the released versions (v11.8.1) is documented at here. This is an env var configuration for now, and is planned to be added to config.json and the lemonade cli in the next release.

moving the config

I suspect you got hit with the cache -> config migration somehow. But it sounds like you have sorted it out, ping if you need support on anything in particular at this point.

[mac2net]

Thanks for responding.

I tried the Copr a few weeks ago and ran into a few problems. I prefer to stick to official releases.
If anything, I will move the install to a Quadlet.

The problem is the fix is wrong as far as I understand reading that page.

It says right here in lemond.service

Server settings (port, host, models_dir, …) live in config.json, not env vars.
The few env vars lemond still reads (HF_TOKEN, LEMONADE_API_KEY,
LEMONADE_ADMIN_API_KEY) can be set via systemctl edit lemond drop-ins or the
optional environment file below.

EnvironmentFile=-/etc/default/lemond
While from what I saw, the command created a file in

/etc/lemonade/conf.d/cors.conf

which is not used by the actual service.

I also had trouble setting the API key probably for the same reason. I haven't tried it again since discovering this information.

[abn]

EnvironmentFile=-/etc/default/lemond
While from what I saw, the command created a file in

/etc/lemonade/conf.d/cors.conf

which is not used by the actual service.

Can you please clarify what command you are talking about here?

And what do you mean not used - are you referring to the env var itself or the conf file you created?

I also had trouble setting the API key probably for the same reason. I haven't tried it again since discovering this information.

For systemd, the service dropins are usually at /etc/systemd/system/<service_name>.service.d/. And for your case if you want to use conf.d the unit file or a drop in must include this:

EnvironmentFile=-/etc/lemonade/conf.d/*.conf

[mac2net]

Not sure what's not clear here…
The correct instructions are in the lemond.service.
Everything else is irrelevant including the instructions linked in fix
export LEMONADE_ALLOWED_ORIGINS= does not work.