Reasons for selecting the Llama architecture
#2
by
torchtorchkimtorch
- opened
Thank you for providing new LLMs.
Is there any specific reason for using the Llama architecture?
Hello,
when we started development, the Llama architecture was accepted as the standard Transformer structure. Additionally, many various tools for training and inference of models with this architecture were provided, so no additional development or customization was necessary. We believed that high-quality data has a greater impact on model performance, so we adopted this standard architecture and focused on obtaining good data.