💥 Same Qwen3.8-27B. One build fits a 24 GB card. The other needs 80. Q4_K_M GG… · @QwenAIFans · TwKit

@QwenAIFans

💥 Same Qwen3.8-27B. One build fits a 24 GB card. The other needs 80. Q4_K_M GGUF on Ollama / LM Studio / llama.cpp: 17.2 GB at 8K. vLLM BF16: 52.8 GB. A 48 GB card still will not start it. Weights alone are about 50.9 GB. Grab FP8 (~30.6 GB) or stay on GGUF. Format picks the GPU bill. What card are you stuck on? via http://gpu.nz 👇 https://gpu.nz/en/guides/ollama-lm-studio-llama-cpp-vllm

曝光 576 · 评论 3 · 点赞 8 · 书签 4