{"paper":{"title":"ToolAlpaca: Generalized Tool Learning for Language Models with 3000 Simulated Cases","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Compact language models can learn to use new real-world tools by training on simulated multi-agent interactions.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Boxi Cao, Hongyu Lin, Le Sun, Qiao Liang, Qiaoyu Tang, Xianpei Han, Ziliang Deng","submitted_at":"2023-06-08T15:46:32Z","abstract_excerpt":"Enabling large language models to utilize real-world tools effectively is crucial for achieving embodied intelligence. Existing approaches to tool learning have either primarily relied on extremely large language models, such as GPT-4, to attain generalized tool-use abilities in a zero-shot manner, or utilized supervised learning to train limited scopes of tools on compact models. However, it remains uncertain whether smaller language models can achieve generalized tool-use abilities without tool-specific training. To address this question, this paper introduces ToolAlpaca, a novel framework d"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experimental results demonstrate that ToolAlpaca achieves effective generalized tool-use capabilities comparable to those of extremely large language models like GPT-3.5, demonstrating that learning generalized tool-use ability is feasible for compact language models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The simulated multi-agent interactions produce training data whose distribution is close enough to real-world tool use that fine-tuned models generalize to unseen APIs without additional per-tool supervision.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ToolAlpaca trains 7B and 13B models on 3938 simulated tool-use cases to reach generalized tool-use performance comparable to GPT-3.5 on unseen APIs.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Compact language models can learn to use new real-world tools by training on simulated multi-agent interactions.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c03b8e867060ecd148fde9265f234cd5d4226ebfdcfd600b990dab9c82d2d9c4"},"source":{"id":"2306.05301","kind":"arxiv","version":2},"verdict":{"id":"dbd30bb2-7336-4641-8cf6-00bdd5144a9e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T22:59:52.482884Z","strongest_claim":"Experimental results demonstrate that ToolAlpaca achieves effective generalized tool-use capabilities comparable to those of extremely large language models like GPT-3.5, demonstrating that learning generalized tool-use ability is feasible for compact language models.","one_line_summary":"ToolAlpaca trains 7B and 13B models on 3938 simulated tool-use cases to reach generalized tool-use performance comparable to GPT-3.5 on unseen APIs.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The simulated multi-agent interactions produce training data whose distribution is close enough to real-world tool use that fine-tuned models generalize to unseen APIs without additional per-tool supervision.","pith_extraction_headline":"Compact language models can learn to use new real-world tools by training on simulated multi-agent interactions."},"references":{"count":34,"sample":[{"doi":"","year":2023,"title":"API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs","work_id":"a20d9332-ab34-485c-a060-1ba47cc98930","ref_index":1,"cited_arxiv_id":"2304.08244","is_internal_anchor":true},{"doi":"","year":2023,"title":"HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face","work_id":"f20ed1da-2676-4598-a11b-54549718735b","ref_index":2,"cited_arxiv_id":"2303.17580","is_internal_anchor":true},{"doi":"","year":2023,"title":"On the tool manipulation capability of open-source large language models.arXiv preprint arXiv:2305.16504(2023)","work_id":"18febc22-221f-4db2-8168-15e148f613c3","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Write a general overview of the API 's purpose and functionality","work_id":"4f28c4cb-f1aa-4915-97eb-fe1ba788466e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"List and briefly describe all features provided by the API, ensuring each feature has a clear and distinct purpose with low coupling between them","work_id":"ed0e44bb-d435-423f-a67e-eb9b69f7ea8e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":34,"snapshot_sha256":"25d4d992dc8065cdcfcf4029f83dbf3cdb6812c2d13360a181c9ed5fed450295","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"45df9e96c87c60b8217f373fb1c05b4802bb4e1df63fd97a244959507511895e"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}