#qwen3.8-flash-next

Running a 95 GiB MoE Model on a 32 GiB GPU: From 10 to ~100 Tokens/s