That's impressive progress in reducing training time for large-scale ChatGPT models! At CentML, we can further optimize training and inference workloads, potentially reducing training time and lowering costs for large-scale inference. Our software products offer ML compiler optimizations for inference and training, ML model deployment profiling, and tools for predicting compute time and cost on various hardware configurations. By incorporating CentML's optimizations, you can unlock even greater efficiency, allowing your business to scale effectively and sustainably while maintaining the best-in-class price-performance ratio. Let's connect to discuss how CentML can help you achieve even better results with all kinds of ML workloads, not only GPT-like models: https://centml.ai/benchmarks/