consider adding `step_embed_llama()`

[EmilHvitfeldt]

basically an sentence encoder - people disagree on whether this is a good idea or not

There is an encoder method that is deterministic

https://github.com/McGill-NLP/llm2vec

https://stackoverflow.com/questions/76926025/sentence-embeddings-from-llama-2-huggingface-opensource

https://lmsys.org/blog/2023-03-30-vicuna/

https://huggingface.co/sentence-transformers/all-mpnet-base-v2

[jrosell]

I think sentence encoder from ollama, sentence_transformers or huggingface could be good addition, at last for people doing topic modeling.

Recently I had to do some readjustment of single topic assigment to a list of documents and I compared scores (0.2*nn_preservation + 0.3 * silhouette + 0.3 * spearman + 0.1 * single_doc_topic_count + 0.1 * dominance_topics_count) of different embeddings models and reduction methods with hdbscan clustering.

For hugggingface models I used https://r-text.org/articles/text.html#textembed-mapping-text-to-numbers-using-huggingface-language-models
For sentence_transformers I used reticulate and the python package.
If depending on python is not an option and you want to only use ollama embedings, I used these two options in the past:
https://hauselin.github.io/ollama-r/reference/embed.html
https://ragnar.tidyverse.org/reference/embed_ollama.html (more recently)