Note Input dataset for generating synthetic data. We use the `instruction` column as a starting point.

Note The dataset was generated from our pipeline. The `instruction` column from the input dataset becomes the anchor, alongside a generated positive and negative pair. This results in a triplets dataset we can use to train a Sentence Transformers model. You can find the code used here: https://github.com/davanstrien/awesome-synthetic-datasets

davanstrien/code-prompt-similarity-model

Sentence Similarity • 0.1B • Updated May 29, 2024 • 4 • 6

Note A fine-tuned Sentence Transformers model using the above dataset. You can see we get a nice bump in performance from minimal fine-tuning.