input text
quantized: no yes (4x smaller model, but currently the embeddings are inaccurate - see readme)
start