All Posts

We introduce Model augmented fine-tuning (Mafin) -- a novel approach for fine-tuning a black-box embedding model by augmenting it with a trainable embedding model.

We propose a practical acquisition function for prompt/completion pairs based on the predictive entropy of the language model and a measure of certainty of the implicit preference model optimized by DPO.

We introduce the fine-tuning augmented embedding (FAE), a method that can improve OpenAI's embedding model for your data.

We introduce a simple method to automatically determine the optimal chunk size to improve the summarization quality for large documents with large language models.