🛠️ Maple – Sub-2-BPW Moe Pruning to Run Gemma 4 26B at 45 TPS on a 6GB GPU

Summary
Maple is a sub-2-BPW MoE pruning tool designed to run Gemma 4 26B at 45 TPS on a 6GB GPU.

Why it’s interesting
Achieves high token generation speeds for a large model on a modest 6GB GPU via sub-2-BPW MoE pruning.

Source metrics: Points 2 · Comments 0
HN discussion · Project
Source: #HackerNews / Show HN