Hey there ,
First of all, I found the provided examples and samples very clear thanks to the people who wrote them
I’m currently testing Sentis with a Sentence Similarity Hugging Face Model. The inference is working correctly (I wrote the Tokenizer in C#) which is a first win
My problem is in the post-processing, this model takes sentences [“I love dogs”, “I love cats”] and embed. The idea is to calculate the cos similarity of these embeds.
But first, I need to do a MeanPool of the embeds. And the first step of this MeanPool is doing the equivalent of torch.unsqueeze(-1).
Given this Unsqueeze method does not exist in Ops what I’ve done instead was using Expand but the problem is that this returns a Tensor not a TensorFloat. And the next step is a MatMul between two TensorFloat. So I need to cast my Tensor to a TensorFloat.
I don’t find how to cast a Tensor to TensorFloat in Sentis. Or methods in Tensor to transform a Tensor to a TensorFloat.
Thanks for your help