Unverified50% confidenceBenchmarkExact time
在对照测试中，35B模型速度约100 tokens/秒（Mac Studio），是9B模型16 tokens/秒的约五倍
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
Valid until: 10/11/2026
Sources
Related Claims
Unverified基于MLX框架，顶配Mac机型运行Llama-3-70B量化模型可达约20 tokens/秒的生成速度81% similarUnverified在4×RTX 3090 SXM4配置下，GLM-5.2量化版本的Token生成速度约为6.3-6.8 tokens/秒，提示评估速度约为10.6 tokens/秒74% similarUnverified本地部署DS4模型在STM32嵌入式开发测试中平均输出速度约为23 tokens/秒，推理时内存占用飙升至110GB，峰值速度达83.8 tokens/秒72% similarUnverifiedllama.cpp的GGUF格式4-bit量化版本使Llama 3.1 70B模型可在配备64GB统一内存的Apple M2 Ultra上运行，推理速度约20-30 tokens/秒71% similarUnverified用于文本补全任务的小型模型（1B–7B参数）在拥有8GB或以上统一内存的Mac上可流畅运行，推理延迟保持在几百毫秒以内。71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502659API
curl https://kongchang.com/api/v1/knowledge/claims/502659MCP
get_claim(id=502659)