How can I fine tune sentence transfomer without any labels?

Question

I only have product descriptions and nothing else. I need to match similar products using cosine similarity. I have achieved this by taking embeddings from the Sentence Transformer. However, I need to fine-tune the Sentence Transformer on my product description data to obtain better embeddings. Can someone please explain how I can perform this task?

I have attempted to use the Sentence Transformer, but I need to fine-tune it on unlabelled data to obtain improved embeddings.