Is there a way to add a model with ONNX resources from HF to the program?
GrisaiaEvy opened this issue · 4 comments
I didn't find a way to add a model by myself. It seems that in the current version(3.5), you can only use the model defined in model.rs
.
Should we allow the user to add a model (which have ONNX file) on HF by themselves?
To use local models, you can use https://docs.rs/fastembed/latest/fastembed/struct.TextEmbedding.html#method.try_new_from_user_defined.
If you want some model from HF to be supported, you can raise a PR similar to #54.
To use local models, you can use https://docs.rs/fastembed/latest/fastembed/struct.TextEmbedding.html#method.try_new_from_user_defined.
If you want some model from HF to be supported, you can raise a PR similar to #54.
I mean not through PR, just pass the EmbeddingModel
parameter when creating an EmbeddingText
, so that it's flexible to use any model on HF.
let model = TextEmbedding::try_new(InitOptions { model_name: EmbeddingModel::new("a model on HF with ONNX file"), show_download_progress: true, ..Default::default() }).unwrap();
I apologise if it's a stupid quetsion because i jsut started learning this project a few days ago :)
Currently, only models listed in models.rs
are supported directly.
If you have the ONNX and tokenizer files locally, you can use the
https://docs.rs/fastembed/latest/fastembed/struct.TextEmbedding.html#method.try_new_from_user_defined method.
Currently, only models listed in
models.rs
are supported directly. If you have the ONNX and tokenizer files locally, you can use the https://docs.rs/fastembed/latest/fastembed/struct.TextEmbedding.html#method.try_new_from_user_defined method.
Ok, thank you for your answer.