fintuning model router
Opened this issue · 1 comments
Teera21 commented
how can i fintunig this model in semantic-router to improve my performance and accuracy in decusion rout.
andreped commented
Hello, @Teera21! :]
What do you mean finetune? Finetune the embedding model? Or do you just want to enhance accuracy in the routing? I would try using HybridRouteLayer
and tuning the alpha
, as well as defining more clear routes by refining the utterances. There is also some gain to tune the aggregation
method and top_k
. At least in my experiments.
EDIT: Note that there is a PR on fine-tuning an embedding model here: #258
So there will likely be a notebook you could follow very soon if you want to fine-tune your own model.