Standard model parameters include temperature, top_k, top_p, typical_p, repetition_penalty,...
Those parameters seems to be part of the caikit-nlp stack, or even the base TGIS.
However those parameters are not exposed in the client (or at least they are not documented).
Note: although those parameters are part of Caikit-NLP and TGIS, up until two months ago they were not working. Querying the gRPC interface of Caikit-nlp by adding for example "temperature" parameter would yield an "unknown parameter" response. So this would also have to be checked up the stack.
Being able to set those parameters is a hard requirement. For text summarization you would need at near-zero temperature. For text creation you would need to set this higher, and so on...
Standard model parameters include temperature, top_k, top_p, typical_p, repetition_penalty,... Those parameters seems to be part of the caikit-nlp stack, or even the base TGIS. However those parameters are not exposed in the client (or at least they are not documented). Note: although those parameters are part of Caikit-NLP and TGIS, up until two months ago they were not working. Querying the gRPC interface of Caikit-nlp by adding for example "temperature" parameter would yield an "unknown parameter" response. So this would also have to be checked up the stack.
Being able to set those parameters is a hard requirement. For text summarization you would need at near-zero temperature. For text creation you would need to set this higher, and so on...