metadata
license: apache-2.0
MarinbadGPT
MarinbadGPT is a language model based on HuggingFaceTB's SmolLM-135M architecture, finely trained on a corpus of Marinbad games. The aim of this model is to generate games of Marinbad and play them against a human player.
Model Training
The training of MarinbadGPT was conducted on a high-performance computing infrastructure utilizing NVIDIA H100 GPUs, renowned for their power in the field of deep learning.
Training Configuration:
- Infrastructure: 2x NVIDIA H100 (80GB HBM3)
- Duration: 1 hour
- Optimizer: AdamW
- Learning Rate: 3e-4
- Batch Size: Micro batch size of 128, with gradient accumulation steps of 8, resulting in an effective batch size of 1024
- Warmup Steps: 100