MistralLite is a fine-tuned variant of Mistral 7B optimized by Amazon for handling longer contexts, comfortably processing up to 32k tokens where the base model would struggle. It trades some general-purpose flexibility for stronger performance on tasks that require reading and reasoning over lengthy documents. A focused, efficient colleague that punches above its weight class for its size when documents get long.