Which specific 1.3B Llama 2 base model was used for the quality scorer?

Dear Seed-Coder team,

Thank you for open-sourcing the Seed-Coder models. I was reading the technical report and noticed that the quality scorer was fine‑tuned from a “pretrained 1.3B Llama 2 model”.

To better understand and potentially reproduce your work, could you please clarify which exact base model this refers to? Is it an official Meta release, or a community variant based on the Llama 2 architecture (e.g., DeepSeek‑Coder‑1.3B, or another one)?

I would really appreciate your clarification. Thanks in advance for your time!