⚡️ You don't need a 4090 for Qwen3-8B. At Q4_K_M with an 8k context it's about… · @QwenAIFans · TwKit

@QwenAIFans

⚡️ You don't need a 4090 for Qwen3-8B. At Q4_K_M with an 8k context it's about 7.5 GB total. A 10 GB GPU or a 16 GB Mac can run it. The catch: stretch context to 128k and the KV cache alone adds ~19 GB on top of the weights. Size the window, not just the model. via llamaperf 👇 https://llamaperf.com/model/qwen3/vram

曝光 750 · 评论 1 · 点赞 6 · 书签 2