[QUESTION] Protocol of adding a new model (Stella_en_<*>_v5 family) implementation with Candle

[AnubhabB]

Hi,

I have a working implementation of Stella_en_<*>_v5 family of models which is one of the top ranking model in the MTEB leaderboard for reranking and retrieval.

It's basically built on top of the candle-transformers::qwen2 implementation with the language modeling head swapped for their pre-trained dense layer.

I was hoping to open a pull request with candle along with an example.

Questions:

1. Would it be OK?
2. If it is, is there a protocol to follow for candle_transformers? Something akin to How to add a model to transformers
3. For embedding models are there any candle standard API implementations required? My implementations just spits out the logits from the forward pass.

Looking for some guidance. Thanks in advance.

CC: @EricLBuehler @LaurentMazare