Chainbase open-sources the cryptocurrency-centric large language model Theia-Llama-3.1-8B

Decentralized full-chain data network Chainbase announced the open-source of Theia-Llama-3.1-8B, a large language model centered around cryptocurrency. The training data for the model comes from CoinMarketCap and detailed research reports. Chainbase claims that this model has lower perplexity and higher BERT scores compared to other mainstream models. LORA is used to efficiently fine-tune the model, making large pre-trained models suitable for specific tasks with lower computing capabilities. The model's memory usage is significantly reduced, and inference speed is accelerated while maintaining acceptable accuracy. In addition, Chainbase completed a $15 million Series A financing round in July this year, led by Tencent Investment Group and Matrix Partners China.