(Optional) Self hosted embedding model solution

[wsxiaoys]

1. Need to design and implement standard interface like TextGeneration.
2. Consider add bert based embedding to upstream llama.cpp for integrate encoder-decoder model

Related:
https://github.com/ggerganov/llama.cpp/blob/master/examples/embedding/embedding.cpp
ggml-org/llama.cpp#2872

[wsxiaoys]

Candle provides a reference implementation: https://github.com/huggingface/candle/tree/main/candle-examples/examples/jina-bert

[wsxiaoys]

Related: https://github.com/FL33TW00D/embd

Benchmark: https://twitter.com/fleetwood___/status/1729001066069709255