Update README.md

README.md

@@ -9,7 +9,7 @@ language:
 
 For the first time among Korean-targeted LLMs, we’re releasing **intermediate checkpoints** from the Tri family—**0.5B**, **1.9B**, and **7B**—to advance research on LLM training dynamics.
 
-Checkpoints are published **every 20,000 steps (≈40B tokens)**, and each step’s release is distinguished by its **branch name** so you can easily navigate between versions and analyze training progress at consistent intervals.
+Checkpoints are published **every 20,000 steps (≈20B tokens for 0.5B,  ≈40B tokens for 1.9B and 7B, ≈160B tokens for 70B)**, and each step’s release is distinguished by its **branch name** so you can easily navigate between versions and analyze training progress at consistent intervals.
 
 You can grab the **Tri-7B** model here: [https://huggingface.co/trillionlabs/Tri-7B](https://huggingface.co/trillionlabs/Tri-7B?utm_source=chatgpt.com).