Recipe recommendation provides a concrete setting for comparing two approaches to adapting a language model. Fine-tuning changes what the model learns from examples; retrieval-augmented generation (RAG) supplies relevant information when it produces a response.
In this project, I fine-tuned TinyLlama-1.1B on the HUMMUS dataset using low-rank adaptation (LoRA), integrated retrieval and compared the resulting outputs. LoRA trains a compact set of additional parameters rather than updating every model weight. The comparison brings together model adaptation and access to external information within the same application domain.
Fine-tuned TinyLlama for recipe recommendations, integrated retrieval, and compared fine-tuned and RAG outputs.
Performed model fine-tuning, RAG integration, and output comparison.
Used Python and TinyLlama-1.1B.
Fine-tuned TinyLlama-1.1B on the HUMMUS dataset.
Applied LoRA optimization.
Integrated Retrieval Augmented Generation.
Compared fine tuned vs RAG outputs.
To discuss fine-tuning and retrieval-augmented generation for recipe recommendations, get in touch.