Language models have evolved beyond simple text generation; they can now interact with functions and tools to perform tasks, answer questions, and even execute code. In this blog post, we explore the fascinating realm of using language models in conjunction with function tools to achieve powerful results.

Recent posts delve into the possibilities of using language models to generate structured responses and explain the functions within a given context.

Contrary to the traditional approach of fine-tuning language models, a new methodology has emerged. By leveraging grammar and a well-crafted system prompt that elucidates the functions, users can guide language models to produce specific, structured outputs without the need for fine-tuning.

One approach involves creating a grammar that enforces a specific JSON format and designing a system prompt that explains the desired functions. This methodology ensures a controlled and structured return from the language model.

However, it's crucial to note that a secondary program is often employed to parse the generated text and execute the corresponding functions. This extra layer of control ensures the safe execution of code and prevents potential security risks.

For those looking for an open-source solution, Functionary is a state-of-the-art general-purpose tool-use and function-calling language model. It comes with features like grammar sampling, parallel tool use, and automatic tool execution, integrated with ChatLab. Notably, Functionary can read tool outputs and generate model responses grounded in these outputs, enhancing the model's contextual understanding.

As the landscape of language models continues to evolve, the combination of grammar, system prompts, and function tools opens up exciting possibilities for users to interact with models in a structured and controlled manner. Whether you choose the fine-tuning route or opt for the grammar-based approach, the power to harness language models for diverse tasks is now more accessible than ever.

As the capabilities of language models continue to advance, it is conceivable that "one-size-fits-all" model will remain as the main paradigm. For instance, given the vast number of languages worldwide, many of which are low-resource, the prevalent practice is to pretrain a single model on multiple languages. In this paper, the researchers introduce the Sabiá: Portuguese Large Language Models and demonstrate that monolingual pretraining on the target language significantly improves models already extensively trained on diverse corpora. Few-shot evaluations … click here to read