@librarian-bot\n\t recommend

\n","updatedAt":"2024-01-27T08:52:47.105Z","author":{"_id":"60107b385ac3e86b3ea4fc34","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1627505688463-60107b385ac3e86b3ea4fc34.jpeg","fullname":"Daniel van Strien","name":"davanstrien","type":"user","isPro":true,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":835,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7918877601623535},"editors":["davanstrien"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1627505688463-60107b385ac3e86b3ea4fc34.jpeg"],"reactions":[],"isReport":false}},{"id":"65b4c469ae9d08540f5233cc","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false},"createdAt":"2024-01-27T08:52:57.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [AUTOACT: Automatic Agent Learning from Scratch via Self-Planning](https://huggingface.co/papers/2401.05268) (2024)\n* [Improving Small Language Models' Mathematical Reasoning via Mix Thoughts Distillation](https://huggingface.co/papers/2401.11864) (2024)\n* [ProTIP: Progressive Tool Retrieval Improves Planning](https://huggingface.co/papers/2312.10332) (2023)\n* [MLLM-Tool: A Multimodal Large Language Model For Tool Agent Learning](https://huggingface.co/papers/2401.10727) (2024)\n* [T-Eval: Evaluating the Tool Utilization Capability of Large Language Models Step by Step](https://huggingface.co/papers/2312.14033) (2023)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space","html":"

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

\n","updatedAt":"2024-01-27T08:52:57.657Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7312633395195007},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[{"reaction":"👍","users":["davanstrien"],"count":1}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2401.07324","authors":[{"_id":"65b20c3af32c79fea732fa56","name":"Weizhou Shen","hidden":false},{"_id":"65b20c3af32c79fea732fa57","name":"Chenliang Li","hidden":false},{"_id":"65b20c3af32c79fea732fa58","name":"Hongzhan Chen","hidden":false},{"_id":"65b20c3af32c79fea732fa59","user":{"_id":"64771cfdd7cf39f2e9381aa9","avatarUrl":"/avatars/48adf00c3b653df02628f80511639e19.svg","isPro":false,"fullname":"Ming","user":"MingYan123","type":"user"},"name":"Ming Yan","status":"claimed_verified","statusLastChangedAt":"2024-08-14T08:12:24.505Z","hidden":false},{"_id":"65b20c3af32c79fea732fa5a","name":"Xiaojun Quan","hidden":false},{"_id":"65b20c3af32c79fea732fa5b","name":"Hehong Chen","hidden":false},{"_id":"65b20c3af32c79fea732fa5c","name":"Ji Zhang","hidden":false},{"_id":"65b20c3af32c79fea732fa5d","name":"Fei Huang","hidden":false}],"publishedAt":"2024-01-14T16:17:07.000Z","title":"Small LLMs Are Weak Tool Learners: A Multi-LLM Agent","summary":"Large Language Model (LLM) agents significantly extend the capabilities of\nstandalone LLMs, empowering them to interact with external tools (e.g., APIs,\nfunctions) and complete complex tasks in a self-directed fashion. The challenge\nof tool use demands that LLMs not only understand user queries and generate\nanswers but also excel in task planning, memory management, tool invocation,\nand result summarization. While traditional approaches focus on training a\nsingle LLM with all these capabilities, performance limitations become\napparent, particularly with smaller models. Moreover, the entire LLM may\nrequire retraining when tools are updated. To overcome these challenges, we\npropose a novel strategy that decomposes the aforementioned capabilities into a\nplanner, caller, and summarizer. Each component is implemented by a single LLM\nthat focuses on a specific capability and collaborates with other components to\naccomplish the task. This modular framework facilitates individual updates and\nthe potential use of smaller LLMs for building each capability. To effectively\ntrain this framework, we introduce a two-stage training paradigm. First, we\nfine-tune a backbone LLM on the entire dataset without discriminating\nsub-tasks, providing the model with a comprehensive understanding of the task.\nSecond, the fine-tuned LLM is used to instantiate the planner, caller, and\nsummarizer respectively, which are continually fine-tuned on respective\nsub-tasks. Evaluation across various tool-use benchmarks illustrates that our\nproposed multi-LLM framework surpasses the traditional single-LLM approach,\nhighlighting its efficacy and advantages in tool learning.","upvotes":3,"discussionId":"65b20c3af32c79fea732fa79","githubRepo":"https://github.com/x-plug/multi-llm-agent","githubRepoAddedBy":"auto","ai_summary":"A modular multi-LLM framework enhances tool use by separating task planning, tool invocation, and result summarization, outperforming traditional single-LLM approaches.","ai_keywords":["Large Language Model (LLM)","tool use","task planning","memory management","tool invocation","result summarization","multi-LLM framework","planner","caller","summarizer","two-stage training paradigm","fine-tuning","backbone LLM","tool learning"],"githubStars":237},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6538119803519fddb4a17e10","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6538119803519fddb4a17e10/ffJMkdx-rM7VvLTCM6ri_.jpeg","isPro":false,"fullname":"samusenps","user":"samusenps","type":"user"},{"_id":"60107b385ac3e86b3ea4fc34","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1627505688463-60107b385ac3e86b3ea4fc34.jpeg","isPro":true,"fullname":"Daniel van Strien","user":"davanstrien","type":"user"},{"_id":"64771cfdd7cf39f2e9381aa9","avatarUrl":"/avatars/48adf00c3b653df02628f80511639e19.svg","isPro":false,"fullname":"Ming","user":"MingYan123","type":"user"}],"acceptLanguages":["*"]}">

Papers

arxiv:2401.07324

Small LLMs Are Weak Tool Learners: A Multi-LLM Agent

Published on Jan 14, 2024

Upvote

Authors:

Ming Yan ,

Abstract

A modular multi-LLM framework enhances tool use by separating task planning, tool invocation, and result summarization, outperforming traditional single-LLM approaches.

AI-generated summary

Large Language Model (LLM) agents significantly extend the capabilities of standalone LLMs, empowering them to interact with external tools (e.g., APIs, functions) and complete complex tasks in a self-directed fashion. The challenge of tool use demands that LLMs not only understand user queries and generate answers but also excel in task planning, memory management, tool invocation, and result summarization. While traditional approaches focus on training a single LLM with all these capabilities, performance limitations become apparent, particularly with smaller models. Moreover, the entire LLM may require retraining when tools are updated. To overcome these challenges, we propose a novel strategy that decomposes the aforementioned capabilities into a planner, caller, and summarizer. Each component is implemented by a single LLM that focuses on a specific capability and collaborates with other components to accomplish the task. This modular framework facilitates individual updates and the potential use of smaller LLMs for building each capability. To effectively train this framework, we introduce a two-stage training paradigm. First, we fine-tune a backbone LLM on the entire dataset without discriminating sub-tasks, providing the model with a comprehensive understanding of the task. Second, the fine-tuned LLM is used to instantiate the planner, caller, and summarizer respectively, which are continually fine-tuned on respective sub-tasks. Evaluation across various tool-use benchmarks illustrates that our proposed multi-LLM framework surpasses the traditional single-LLM approach, highlighting its efficacy and advantages in tool learning.