inferencing embeddings models

[arturborycki]

I create working prototype that allow to also inference embeddings models base on text data as input, I was interested to see if this is something commuity is interested in adding to the infera extension, As I am not part of the core team I wanted to see if there is an interest in that aditional direction?

Example:
select infera_load_text_model('gte_model', 'https://huggingface.co/thenlper/gte-base/resolve/main/onnx/model.onnx', 'https://huggingface.co/thenlper/gte-base/resolve/main/tokenizer.json', 512);

with embeddings as (
select 'sentence1' as text, infera_predict_text('gte_model', 'This is a test sentence for embedding generation.') as embedding
union all
select 'sentence2' as text, infera_predict_text('gte_model', 'The quick brown fox jumps over the lazy dog.') as embedding
)

┌───────────┬───────┬────────────────────────────────────── ───────────────────────┐
│ text │ dim │ first_5_values │
│ varchar │ int64 │ float[] │
├───────────┼───────┼──────────────────────────────────────────────────────────────┤
│ sentence1 │ 768 │ [0.16316502, -0.49085385, 0.35360435, 0.77152574, 1.2055475] │
│ sentence2 │ 768 │ [-0.382681, -0.74655575, 0.41404557, 0.2588208, 0.9010632] │
└───────────┴───────┴──────────────────────────────────────────────────────────────┘

[habedi]

Hi, this sounds like a very useful feature. Thanks for sharing it.