This is a repackaged software product wherein additional charges apply for seller support, as well as licensing fees paid to AWS. The additional cost is applicable to the extended support of 24-hour response time. The current Llama3.1-8B product uses a graphical interface to interact with users, no longer a DOS interface. The interface is clean, tidy, and easy to use. The following is the main introduction to the product

Meta recently released a new generation of open-source model series Llama3.1, which includes a 405B parameter version that performs similarly to or even surpasses closed source models such as GPT-4 in some benchmark tests. Llama3.1-8B-Instruction is an 8B parameter version in the series that supports English, German, French, Italian, Portuguese, Spanish, Hindi, and Thai, with a context length of up to 131072 tokens. The knowledge deadline has been updated until December 2023.

In order to enhance the ability of Llama3.1-8B-Instruction, Meta used over 25 million synthetic data generated by a larger 405B model during training. This enables Llama3.1-8B-Instruction to demonstrate cognitive and reasoning abilities similar to GPT3.5Turbo in code, mathematics, and other tests.

OpenBuddy utilizes the Llama3.1-8B-Instruction model and trains it on a small amount of Chinese data to release OpenBuddy-Llama3.1-8B-v22.1-131K, a new generation open-source cross language model with Chinese question answering and cross language translation capabilities. Although Llama3.1 itself does not have Chinese language proficiency, after training, the model is able to generate answers that are usually only possible with larger models on problems that are prone to conceptual confusion, demonstrating stronger cognitive potential.

However, due to limitations in the training dataset and time, OpenBuddy-Llama3.1-8B-v22.1 still has limitations in Chinese knowledge, especially in traditional cultural knowledge. Nevertheless, the model exhibits relatively stable performance in tasks such as comprehension of long texts, thanks to its inherent ability in long texts.

In the future, OpenBuddy plans to conduct larger scale training on the 8B and 70B models to enhance their Chinese knowledge reserve, long writing ability, and cognitive ability, and explore the possibility of fine-tuning the 405B model.