I trained a 348M-active MoE on ~11B tokens. It scores 85.5% vs Qwen2.5-1.5B's 53.3% on shuffled fact chains. GGUF + eval set — please try to break it.

1. 

   消息推送

   I trained a 348M-active MoE on ~11B tokens. It scores 85.5% vs Qwen2.5-1.5B's 53.3% on shuffled fact chains. GGUF + eval set — please try to break it.

   23分钟前

   I trained a 348M-active MoE on ~11B tokens. It scores 85.5% vs Qwen2.5-1.5B's 53.3% on shuffled fact chains. GGUF + eval set — please try to break it.
   
   🌐 链接: https://www.reddit.com/r/LocalLLaMA/comments/1xssj9q/i_trained_a_348mactive_moe_on_11b_tokens_it/
   🏷️ 分组: Reddit｜AI·自托管·VPS
   🕒 时间: 2026-11-11 01:11:20

   Reddit

   

   From the LocalLLaMA community on Reddit: I trained a 348M-active MoE on ~11B tokens. It scores 85.5% vs Qwen2.5-1.5B's 53.3% on…

   Explore this post and more from the LocalLLaMA community

   23分钟前