-
Type:
Improvement
-
Resolution: Unresolved
-
None
-
Affects Version/s: None
-
Component/s: translate5 AI
-
High
-
None
-
Enhancement: Enable setting an LLM Models capabilities when configuring Models unknown to translate5
-
None
-
Emptyshow more show less
PROBLEM
We have several Problems / Bugs when Models are configured, that are unknown for our static configuration. In these case usually inappropriate defaults are used, that e.g. lead to token-errors
SOLUTION
When a user configures a model, that is unknown to translate5, he must provide the needed configurational data. There are open questions:
- where to store that data: Most appropriate seems "specific data" where the other model-properties are stored
- how to provide it: Must be part of the wizard that creates the LR
- Data format: Currently we use a JSON-stucture to store the capabilities:
{
'maxOutput' => 100000,
'maxContext' => 200000,
'reasoning' => true,
'reasoningEffort' => ['low', 'medium', 'high'],
'unsupportedParams' => ['temperature', 'top_p', 'presence_penalty', 'frequency_penalty'],
}
We will need some more capabilities: "chatApiSupported" & "responsesApiSupported"