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. |