Chainbase, a company specializing in full-chain data, has open-sourced their large language model called Theia-Llama-3.1-8B, designed specifically for the crypto-encryption field. It excels mainstream models on measures such as perplexity and BERT scores. Furthermore, the team behind this achievement built the first professional dataset pertinent to Web3 and the top 2,000 crypto projects according to CoinMarketCap. With advanced tools, this model’s inference speed and memory requirements were considerably improved. The Theia-Llama-3.1-8B has already been utilized within the popular crypto application, TheiaChat, serving more than 300,000 active users daily.
Source: source