Unverified90% confidenceFactTime unknown
测试在一台36GB内存的MacBook上进行，模型以128K上下文大小和Q8 KV缓存加载
1
Sources
90%
Confidence
Medium-term (~90 days)
Relevance
5/31/2026
First Seen
Valid until: 8/29/2026
Sources
Qwen3.6本地部署实战：35B模型逆向LTE调制解调器击败Claude
bilibili比特光锥_BightCone
Related Entities
Related Claims
Unverified27B稠密模型在24GB M4 Pro Mac mini上使用4-bit量化可以运行，速度约为11 tok/s78% similarUnverifiedLlama 3.3 70B经过4-bit量化后模型大小约为40GB，可在配备64GB内存的Mac Studio或单张48GB显存的专业GPU上运行77% similarVerified一个原本需要60GB显存的30B参数模型，经过量化后能在16GB甚至8GB显存的消费级显卡上运行77% similarUnverified在128GB的MacBook Pro M3 Max上，独立评测显示DS4生成速度达每秒26.68个Token75% similarUnverified容量并非越大越好：64GB以下高速盘性价比低已淘汰，128GB是甜点容量；256GB以上更推荐移动固态。做Win/Mac启动盘32-64GB足够，装ISO镜像选128GB73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/13911API
curl https://kongchang.com/api/v1/knowledge/claims/13911MCP
get_claim(id=13911)