MarioGPT: Open-Ended Text2Level Generation through Large Language Models

Authors: Shyam Sudhakaran, Miguel González-Duque, Matthias Freiberger, Claire Glanois, Elias Najarro, Sebastian Risi

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Here, we introduce Mario GPT, a fine-tuned GPT2 model trained to generate tile-based game levels, in our case Super Mario Bros levels. Mario GPT can not only generate diverse levels, but can be text-prompted for controllable level generation, addressing one of the key challenges of current PCG techniques. ... 4 Experiments and Results
Researcher Affiliation Collaboration Shyam Sudhakaran1, Miguel González-Duque 1, Matthias Freiberger 1, Claire Glanois1, Elias Najarro1, Sebastian Risi1,2 1IT University of Copenhagen, 2modl.ai, Copenhagen
Pseudocode No The paper describes processes and models (e.g., the novelty search setup and mutation operators in Figure 3) but does not include a formally labeled 'Pseudocode' or 'Algorithm' block with structured steps.
Open Source Code Yes Code available at https://github.com/shyamsn97/mario-gpt. ... To facilitate this, the code to run the experiments in this paper is publicly available at: https://github.com/shyamsn97/mario-gpt.
Open Datasets Yes Mario levels are represented similarly to previous works [45, 8, 35, 33, 34, 12], using the levels provided in the Video Game Level Corpus (VGLC) [40].
Dataset Splits No While Table 1 is titled 'Training Reconstruction Accuracy Validation Set', the paper does not explicitly provide specific percentages, absolute sample counts, or a detailed methodology for the training/validation/test dataset splits needed to reproduce the experiment.
Hardware Specification Yes Because the model is relatively small, it can be trained using a single Nvidia Ge Force RTX 2080 Ti GPU.
Software Dependencies No The paper mentions utilizing 'the open source transformers library' and 'distilgpt2' but does not specify concrete version numbers for these or other software dependencies required to replicate the experiment.
Experiment Setup Yes We train Mario GPT for 50,000 steps, sampling 4 random slices of levels at each iteration and optimize the model using the Adam optimizer [20]. ... In our case, when generating levels we use a temperature of 2.4-2.7.