Hardware & runtime reference for TinyLlama‑1.1B

[apple-love]

Hi,

Could you share some reference information for reproduction:

1. How many GPUs are recommended for ptp_pregenerate (pre‑generate base‑model completions) and ptp_train respectively?
2. What is the approximate runtime for each step with TinyLlama‑1.1B (under your experimental setup)?