- Choosing an Embedding Loss Function
MNRL, triplet, CoSENT and batch-triplet losses: what each one optimises, which data it needs, and a playground that fine-tunes a model in your browser.
12 min - Fine-Tuning Embeddings with Contrastive Learning and MRL
Contrastive Learning and Matryoshka Representation Learning.
4 min - Small Language Models Found Their Lane
Why smaller LLMs became useful for routing, extraction, classification, and edge workflows.
5 min - QLoRA Made Fine-Tuning Feel Practical
How LoRA and QLoRA changed the economics of adapting LLMs.
8 min