Weblate
failed automatic translation: Unsupported model: auto #21350
Replies: 2 comments 7 replies
|
It isn't the chat endpoint failing, it's the model listing. The configured model is now checked against what the API actually reports. def select_model(self) -> str:
models = self._models if self._models is not None else set()
if self.settings["model"] in models:
return self.settings["model"]
if self.settings["model"] == "auto":
for model, _name in self.settings_form.MODEL_CHOICES:
...
if model in models:
return model
if self.settings["model"] == "custom":
return self.settings["custom_model"]
msg = f"Unsupported model: {self.settings['model']}"
raise MachineTranslationError(msg)
The thing to test is therefore the listing, not translation: with whatever base URL you have configured. Usual causes are a base URL pointing at a gateway that only implements chat completions, or a restricted key — project keys can be issued without model read permission, and then listing fails while completions work fine. Two things while you test. The list is cached for an hour, so a corrected key won't take effect straight away. And if your endpoint genuinely has no listing, set the model to Custom model and put the id in Custom model name; that branch returns it without consulting the fetched list at all. |
|
That rules out the gateway, and it narrows things more than it looks. With no base URL set you're going straight to api.openai.com, and a rejected key wouldn't reach this error at all. What still fits is a project with restricted model access. If none of those ten come back, @nijel that's the output worth having: Two practical notes while you're testing. Whatever came back is cached for an hour and an empty set caches the same as a full one, so a change on the OpenAI side won't show up straight away. And to get it translating today, set the model to Custom model and put the exact id in Custom model name, since that branch returns before the list is consulted at all. One other thing. If the error literally printed |
|
I had to generate a new key, the api key I had supported a limited number of models which apparently the newer versions of weblate don't support? With a new key: The weblate configuration only accepts gpt-5 models? I left the selection on "automatic", but without a key that supports gpt-5 models the translation just stopped working. |
|
But now automatic suggestions are no longer working? I used to have suggestions from OpenAI for empty translations, the suggestions would load on demand when you navigate to the empty translation. Now I have zero suggestions, either from weblate or from openai. |
|
Upon trying again, I see that the automatic suggestion just takes much longer than it used to, I'm guessing because it's using gpt-5 instead of gpt-4 |
|
I've added back legacy models in #21390. It also tries to reorder models to give better quality and performance while not sacrificing costs. |
|
https://github.com/orgs/WeblateOrg/discussions/21350 should improve the feedback to the user on such an error. Still having the |
|
Eh, the actual link was supposed to be: |
Uh oh!
There was an error while loading. Please reload this page.
Yesterday I pulled the more recent docker images, upgrading my weblate instance from 5.15 to 2026.8.1, and now AI translation has stopped working, I get an error message:
I tried switching the OpenAI model from "Automatic selection" to "GPT-5.4 mini" but this results in the same error message "failed automatic translation: Unsupported model: GPT-5.4 mini".
Looks like OpenAI translation is broken?
All reactions