0.5B base model trained on 60B tokens rivals Qwen3.5-2B trained on 30T+ tokens

by AZDR | View on Hacker News