Nice! Actual test of successfully deploying a 100-billion-level large model on DGX Spark - AntLingAGI
100 billion per machine, soaring speed! (See screen recording👇🏻) I have to say that the domestic open source ecosystem is really getting stronger and stronger now! It is strongly recommended that everyone learn about local model deployment: install Ling-3.0, Qwen 3.8 27B, Minimax H3, access multi-modality, and use various open source large models to build an exclusive local workflow, which can be easily implemented…
Qwen 3.8-27B setup guide: the free local model that matches Opus 4.6.
Here's what hardware ACTUALLY runs it. Two AI builders tested it honestly. No hype. The setup: → Easiest path: Ollama. One click. Paste a command in your terminal. Done. → LM Studio works too, if you
We put Qwen 3.8 27B on a stock office mini-pc with 32GB of memory with no GPU, and let it rip.
Nothing crazy but super dope what we can get done at the edge.
在M2Max的Claudecode中使用Qwen-3.8-27B，由于内存带宽只有400GB/s，经过不懈努力，暂时只能在16token/s的速度
在带缓存的情况下，本地还是太慢了，一次丢进去2w token 要处理半天
You have reached the end of the archive
All of qwen38