Large Language Models in AI such as GPT are quite heavy. We have to upload the weights and dataset to IPFS along with all config files. Then run the models on the blockchain. It was quite hard to coordinate the serialization and deserialization so that the nodes in the network could load the model and run it on time.