TALM: Tool Augmented Language Models
By Aaron Parisi, Yao Zhao, Noah Fiedel. 2022
207 citations recorded by the source corpus.
Transformer based language models (LMs) demonstrate increasing performance with scale across a wide variety of tasks. Scale alone however cannot enable models to solve tasks that require access to ephemeral, changing, or private data that was unavailable at training time. Many useful tasks may also benefit from LMs being able to access APIs that read or modify state. In this work, we present Tool Augmented Language Models (TALM), combining a text-only approach to augment language models with non-differentiable tools, and an iterative"self-play"technique to bootstrap performance starting from few tool demonstrations. TALM exhibits strong performance on both a knowledge-heavy QA task and a reasoning oriented math task with simple tools. At a given model scale, TALM significantly outperforms non-augmented LMs. We further demonstrate that TALM successfully performs out-of-distribution inferences on both QA and math tasks, where non-augmented LMs fail. Our results suggest that Tool Augmented Language Models are a promising direction to enrich LMs' capabilities, with less dependence on scale.
Subfields: Artificial Intelligence, LLM Agents & Tool Use
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Mohit Bansal
- Luke S. Zettlemoyer
- Percy Liang
- Jianfeng Gao
- Yoshua Bengio
- Quoc V. Le
- Graham Neubig
- Sergey Levine
In LLM Agents & Tool Use
- Karthik Narasimhan
- Aleksandra Faust
- Carl Vondrick
- Luke S. Zettlemoyer
- Mustafa Safdari
- D'idac Sur'is
- Izzeddin Gur
- Sachit Menon
Citations in the corpus
- Toolformer: Language Models Can Teach Themselves to Use Tools (2023, methodology): “However, existing approaches either rely on large amounts of human annotations (Komeili et al., 2022; Thoppilan et al., 2022) or limit tool use to task-specific settings only (e.g., Gao et al., 2022; Parisi et al., 2022), hindering a more widespread adoption of tool use in LMs.”
- ViperGPT: Visual Inference via Python Execution for Reasoning (2023, methodology): “In the natural language processing domain, they have been aimed at using exter-nal tools [46, 40], or for structured reasoning using Codex [34, 54, 14, 10].”
- Tool Learning with Foundation Models (2023, unknown): “To this end, researchers recently show that with a few demonstrations, foundation models can teach themselves how to utilize a tool in a self-supervised manner (Parisi et al., 2022; Schick et al., 2023).”
- A Survey on Knowledge Distillation of Large Language Models (2024, methodology): “Commonly used methods mainly relied on human-curated data for training (Parisi et al., 2022; Nakano et al., 2022; Qin et al., 2023c; Song et al., 2023b) or prompt designing(Cai et al., 2023; Shen et al., 2023; Hao et al., 2024).”
- Teaching Algorithmic Reasoning via In-context Learning (2022, background): “We then show how algorithmic prompt can be used as tool use (Parisi et al., 2022), where a model queries another source for a particular type of information.”
- A Survey of Large Language Models (2023, unknown): “To teach LLMs to utilize tools, existing studies add exemplars using tools in context to elicit LLMs [68], or fine-tune LLMs on simulated data about tool utilization [69, 377].”
- A Comprehensive Overview of Large Language Models (2023, unknown): “Architecture details of pre-trained LLMs are available in Table V. Providing these details for instruction-tuned models is unnecessary because it fine-tunes pre-trained models for instruction datasets.”
- Cognitive Architectures for Language Agents (2023, background): “In particular, for NLP tasks that require augmentation of external knowledge or computation, stateless digital APIs (e.g., search, calculator, translator) are often packaged as “ tools ” (Parisi et al., 2022; Schick et al., 2023; Xu et al., 2023a; Tang et al., 2023b; Qin et al., 2023), which can be ”
- Mind's Eye: Grounded Language Model Reasoning through Simulation (2022, methodology): “To endow LMs with updated knowledge, TALM (Parisi et al., 2022) fine-tunes LMs interactively with augmented inputs from API calls, which is rather costly compared to prompt-based Mind’s Eye.”
- Tool learning with large language models: a survey (2024, unknown): “The methods adopted in the early work involved directly inserting the output of tools into the generated response [17, 26, 46, 54].”
- MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback (2023, methodology): “1, similar to Yao et al. 2022); (2) then either interact with tools by generating Python code and executing it through a Python interpreter (“Execute:” in Fig.”
- Chat-REC: Towards Interactive and Explainable LLMs-Augmented Recommender System (2023, background): “Reasoning is defined as breaking down complex tasks into simpler subtasks that the LM can solve more easily by itself or with the help of tools [9,15,13], while tools are external modules that the LM can call to augment its context.”
- Empowering Biomedical Discovery with AI Agents (2024, unknown): “To manage tasks from diverse environments, agents require tools to boost their capabilities [138].”
- Large Language Models as Tool Makers (2023, background): “…analogous to the evolution of human intelligence, recent research has unveiled the potential of augmenting LLMs with external tools , thereby significantly enhancing their problem-solving capacities and efficiencies (Yao et al., 2023; Liu et al., 2023; Parisi et al., 2022; Schick et al., 2023).”
- The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions (2024, background): “Modern large language models (LLMs) are no longer envisioned as simple autocomplete systems—they could instead empower agentic applications such as web agents, email secretaries, virtual assistants, and more (Nakano et al., 2021; Parisi et al., 2022; Schick et al., 2024; Shen et al., 2024).”
- ToolAlpaca: Generalized Tool Learning for Language Models with 3000 Simulated Cases (2023, background): “Concurrently, substantial research efforts are dedicated to fine-tuning smaller language models to acquire the capacity for tool usage (Komeili et al., 2022; Parisi et al., 2022; Schick et al., 2023).”
- ToRA: A Tool-Integrated Reasoning Agent for Mathematical Problem Solving (2023, methodology): “Tool-Augmented Language Models Augmenting LLMs with tools can largely alleviate LLMs’ limitations and improve reasoning and generation performance (Parisi et al., 2022; Mialon et al., 2023; Yao et al., 2023).”
- A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis (2023, methodology): “In Appendix E, we also test WebAgent on WebSRC (Chen et al., 2021b), a static HTML comprehension benchmark, compared to prior transformer models specialized for structured documents (Li et al., 2021b; Zhao et al., 2022).”
- A Survey on Evaluation of Large Language Models (2023, background): “, 2023) and TALM (Parisi et al., 2022), explore the utilization of tools to enhance language models.”