标签： LLM

The best way to think about AI is as a force multiplier for the fields you already know about (or for the rate at which you want to learn about a new one). The 3rd category -no existing judgment, no interest in develo…

The best way to think about AI is as a force multiplier for the fields you already know about (or for the rate at which you want to learn about a new one). The 3rd category -no existing judgment, no interest in develo...

Box 的 CEO Aaron Levie 在 X 上发帖认为，AI 最有效的用法是作为已有专业领域的“力量倍增器”，只有专家才能驾驭 AI 产出高质量成果，而缺乏判断力和兴趣的“第三类人”只能制造大量低效的“垃圾内容”。这一观点与剑桥学者 Henry Shevlin 的“内化知识是 AI 时代的创造性原材料…

CUDA out of memory

CUDA out of memory

用户在 4x RTX 3070 上运行 Qwen/Qwen2.5-Coder-32B-Instruct-GPTQ-Int4 模型，从 vLLM 0.6.x 升级到 0.7.0 并启用 VLLM_USE_V1=1 后，原本可支持 12K context length，升级后仅能支撑 3K，超出即触发

No module named ‘libtpu’

No module named 'libtpu'

用户使用 vLLM 0.11.1rc6 加载 Qwen3-VL-2B-Instruct 多模态模型，设置 --dtype float16 和 --max-model-len 2048 ，在 AMD RX 7800 XT (gfx1101) 上启动时触发 OOM。即使已安装 flash_attn==2