Large Language Models (LLMs) have successfully addressed the difficult realm of artificial intelligence. LLMs are useful in any industry with their amazing ability to create unique and creative content with great linguistic accuracy and consistency. Large language models are often augmented by reasoning skills and the ability to use different tools. Extending basically refers to enhancing or expanding by adding additional elements or features. Enhanced LLM is the addition of external tools and skills to improve performance beyond innate ability.
Applications like Auto-GPT for autonomous task execution are made possible only by the Extended Language Model (ALM). Current ALM efforts primarily rely on a prompting paradigm that interweaves verbal reasoning with tool invocation, which, while effective, also imposes certain limitations. Connecting to external tools first requires LLM to run and pause periodically, which introduces delays and increases token usage. Second, LLM will generate tokens based on previous context and will resume token generation supplying all historical tokens if stopped for a tool response. The result is significant instant redundancy, leading to high costs in terms of token consumption for commercial LLM services.
To address this challenge, a team of researchers recently proposed ReWOO (Reasoning WithOut Observation), a modular paradigm to reduce token consumption. The idea behind ReWOO is to separate the LLM reasoning process from external observations, which can significantly reduce token consumption. By separating the inference process from external observation, ReWOO minimizes the computational load associated with repetitive prompts.
The main components of ALM are step-by-step reasoning, tool calling, and summarization, and ReWOO splits these into three separate modules: planner, worker, and solver. Planners break down tasks into interdependent blueprints of plans, each of which is assigned to a worker. Workers take external knowledge from tools to provide evidence, and solvers combine all plans and evidence to arrive at the final answer to the initial task to be completed.
To evaluate ReWOO’s performance, the team performed in-depth analysis across six open natural language processing (NLP) benchmarks and curated datasets. The results consistently show an improvement with the proposed methodology, with ReWOO achieving a 5x token efficiency improvement and a 4% accuracy improvement on the HotpotQA benchmark involving multi-step inference tasks. ReWOO has proven to be robust even in the presence of external tool failure issues.
Separating parametric modules from non-parametric tool calls not only improves prompt efficiency, but also allows fine tuning of instructions in ReWOO. GPT3.5 with 175B parameters can offload its inference capabilities to a smaller language model, 7B LLaMA, through fine-tuning, which greatly reduces the model parameters and develops effective and scalable ALM Possibilities are emphasized.
ReWOO therefore becomes a promising modular paradigm for ALM as it overcomes the challenges of redundant prompts and computational complexity for the first time.
please check out Paper and Github link. don’t forget to join 22,000+ ML SubReddit, Discord channeland email newsletterShare the latest AI research news, cool AI projects, and more. If you have any questions regarding the article above or missed something, feel free to email me. Asif@marktechpost.com
🚀 Check out 100’s of AI Tools at the AI Tools Club
Tanya Malhotra is a final year student at the University of Petroleum and Energy Research, Dehradun, with a Bachelor of Science in Computer Science Engineering with a specialization in Artificial Intelligence and Machine Learning.
A data science enthusiast with good analytical and critical thinking, she has a keen interest in learning new skills, leading groups, and managing work in an organized manner.
