How to ask questions using all-MiniLM-L6-V2?

all-MiniLM-L6-V2 is not a text generator (chat). It’s a tokenizer of input for sentence transformer libraries. It’s takes a text sentence and produces a vector space. See:

https://www.sbert.net

The use case is typically comparing two sentences, for grouping or search algorithms.

Note: the model supports up to 512 tokens input (most words are multiple tokens, so about 380 words)

Thank you for your question. 

This appears to be the same topic as your previous thread. I have already provided guidance there on loading the model, generating embeddings, and integrating all-MiniLM-L6-v2 into a .NET MAUI Android application. With that in mind, I would like to add the following comments based on the code shared here. 

Regarding the error at _session.Run(inputs), based on the code you shared, one possible cause is that the ONNX model expects additional input tensors that are not being provided. Many BERT-based models, including all-MiniLM-L6-v2 variants, commonly require inputs such as attention_mask and, in some cases, token_type_ids in addition to input_ids. 

For example: 

var inputs = new List 
{ 
    NamedOnnxValue.CreateFromTensor("input_ids", idTensor), 
    NamedOnnxValue.CreateFromTensor("attention_mask", attentionMaskTensor), 
    NamedOnnxValue.CreateFromTensor("token_type_ids", tokenTypeIdsTensor) 
}; 
using var results = _session.Run(inputs); 

The exact input names and required tensors depend on the ONNX model you are using. You may want to inspect the model inputs (for example, through _session.InputMetadata) and compare them against the tensors being supplied before calling _session.Run(). 

Please continue in the original thread and share the complete exception message together with the model input metadata if the issue persists. 

If you found my response helpful or informative, I would greatly appreciate it if you could follow this guide for your confirmation.  

Thank you.