A Data Source for Reasoning Embodied Agents
Authors: Jack Lanchantin, Sainbayar Sukhbaatar, Gabriel Synnaeve, Yuxuan Sun, Kavya Srinet, Arthur Szlam
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show the results of several baseline models on instantiations of train sets. These include pre-trained language models fine-tuned on a text-formatted representation of the database, and graph-structured Transformers operating on a knowledgegraph representation of the database. |
| Researcher Affiliation | Industry | Meta AI {jacklanchantin, sainbar, gab, yuxuans, ksrinet, aszlam}@meta.com |
| Pseudocode | No | The paper includes mathematical equations for model operations and loss functions but does not provide any pseudocode or algorithm blocks. |
| Open Source Code | No | Code to generate the data and train the models will be released at github.com/facebookresearch/neuralmemory. |
| Open Datasets | No | In this work, to further pursue these advances, we introduce a new data generator for machine reasoning that integrates with an embodied agent. The generated data consists of templated text queries and answers, matched with world-states encoded into a database. ... We propose a context-question-answer data generator for embodied agents. ... With this data generation framework, we can create arbitrary amounts of simulated data. |
| Dataset Splits | Yes | Since we are generating the data, we vary the training samples from 1k to 1M, and use a validation set of 10k samples. |
| Hardware Specification | No | The paper does not specify any particular hardware used for running the experiments, such as specific GPU or CPU models. |
| Software Dependencies | No | The paper mentions using 'GPT-2 small model from the Hugging Face library' and 'Adam' optimizer, but it does not specify concrete version numbers for these or other software dependencies. |
| Experiment Setup | Yes | All our models are trained using Adam (Kingma and Ba 2014) for 5,000 epochs, where each epoch is over a chunk of 10,000 training samples. We use a linear warmup of 10,000 steps and cosine decay (Loshchilov and Hutter 2016). For the GPT2 model, we consider learning rates {1e-4, 5e-4, 1e-5} using a batch size of 32. For the structured model, we consider learning rates {1e-4, 5e-4, 1e-5}, batch size 32, layers {2, 3}, and embedding dimensions {256, 512}. |