Code: github.com/PSRahul/LLM-Explorer
A collection of small experiments with recent LLM tooling.
Retrieval-augmented generation
retrieval-augmented-generation.py builds a retrieval-augmented generation (RAG) pipeline with LangChain and Hugging Face models. Documents are retrieved for a question and passed to the model as context.
Synthetic data generation
generate_synthetic_data.py generates artificial tabular data from a few example rows, using the llamafile build of TinyLlama-1.1B. Given rows such as Rahul,India and Mark,America, the model continues the list with new name and region pairs. The output also shows the failure modes of a small model, such as near-duplicate names.
Model fetcher
model_fetcher.py downloads a model from the Hugging Face Hub and runs inference through it, as a quick way to try new checkpoints.