Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
ToolkenGPT: Augmenting Frozen Language Models with Massive Tools via Tool Embeddings
Authors: Shibo Hao, Tianyang Liu, Zhen Wang, Zhiting Hu
NeurIPS 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In diverse domains, including numerical reasoning, knowledge-based question answering, and embodied plan generation, our approach effectively augments LLMs with tools and substantially outperforms various latest baselines. |
| Researcher Affiliation | Academia | Shibo Hao1, Tianyang Liu1, Zhen Wang1, 2, Zhiting Hu1 1UC San Diego, 2Mohamed bin Zayed University of Artificial Intelligence |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1Code is available at https://github.com/Ber666/Toolken GPT |
| Open Datasets | Yes | To evaluate the tool-learning proficiency in numerical reasoning comprehensively, we curate two new test datasets: (1) GSM8K-XL, an enhanced version of the existing GSM8K [10] dataset. |
| Dataset Splits | Yes | We get 6,054 examples, of which 1,000 were allocated for validation, and 5,054 for the training data. |
| Hardware Specification | Yes | In terms of computational resources, we train and test Toolken GPT based on LLa MA-13B and LLa MA-33B using 2 and 4 Nvidia RTX 3090 GPUs, respectively. |
| Software Dependencies | No | The paper mentions using specific models like LLa MA-13B, Chat GPT (gpt-3.5-turbo), and Sentence RoBERTa-large, but does not provide specific version numbers for these or other software dependencies like deep learning frameworks or Python packages. |
| Experiment Setup | Yes | The embeddings were trained with a learning rate of 5e-4, performing early stopping based on the development set, with a maximum of 10 epochs. |