Qwen3.5-27B

27B对话由 Qwen 发布 发起讨论

0

量化版本

205

实测记录

7

讨论数

量化版本

社区与官方发布的量化版本;「关联实测」只统计精确关联到该发布的实测记录。

还没有已收录的量化版本。

实测数据

共 50 组配置

硬件框架量化生成速度运行显存实测次数证据
NVIDIA RTX Sparkllama.cppQ6_K8 tok/s58.8 GB1L2 跨框架验证
NVIDIA RTX Sparkllama.cppQ8_06.8 tok/s67.3 GB1L2 跨框架验证
NVIDIA RTX SparkOllamaFP164.9 tok/s107 GB1L2 跨框架验证
NVIDIA RTX SparkOllamaQ6_K7 tok/s58.8 GB1L2 跨框架验证
NVIDIA RTX SparkOllamaQ8_06.9 tok/s67.3 GB1L2 跨框架验证
NVIDIA RTX 3090OllamaIQ1_S89.8 tok/s22.8 GB1L2 跨框架验证
NVIDIA RTX 5090OllamaIQ3_XS206.4 tok/s30.4 GB1L2 跨框架验证
NVIDIA RTX 5090StrataQ2_K376.5 tok/s30.4 GB2L2 跨框架验证
AMD Ryzen AI Max+ 395vLLMQ8_018.2 tok/s67.3 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395OllamaIQ3_S27.3 tok/s39.52 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395OllamaQ6_K16.1 tok/s58.8 GB1L2 跨框架验证
Apple M5 Maxllama.cppQ2_K65.6 tok/s33.29 GB1L2 跨框架验证
Apple M5 MaxOllamaIQ2_M64.3 tok/s31.58 GB1L2 跨框架验证
Apple M5 MaxOllamaIQ3_XS62.2 tok/s37.25 GB1L2 跨框架验证
Apple M5 Ultrallama.cppIQ1_S131.5 tok/s25.91 GB1L2 跨框架验证
Apple M5 UltraOllamaQ3_K_S84 tok/s38.96 GB1L2 跨框架验证
NVIDIA H20vLLMIQ2_XS63.2 tok/s29.32 GB1L2 跨框架验证
NVIDIA H20vLLMQ8_027.6 tok/s67.3 GB1L2 跨框架验证
NVIDIA H20StrataIQ3_S67.5 tok/s39.52 GB1L2 跨框架验证
NVIDIA RTX PRO 6000 Blackwell 96GBllama.cppQ4_K_M181.7 tok/s48.59 GB1L2 跨框架验证
NVIDIA RTX PRO 6000 Blackwell 96GBllama.cppQ6_K133.1 tok/s58.8 GB1L2 跨框架验证
NVIDIA RTX PRO 6000 Blackwell 96GBOllamaFP1679.6 tok/s91.2 GB1L2 跨框架验证
NVIDIA RTX PRO 6000 Blackwell 96GBOllamaQ3_K_L177.2 tok/s42.92 GB1L2 跨框架验证
Apple M2 Ultrallama.cppQ3_K_L82.5 tok/s42.92 GB1L2 跨框架验证
Apple M2 UltraOllamaIQ2_M105 tok/s31.58 GB1L2 跨框架验证
NVIDIA RTX Sparkllama.cppQ4_K_M10.2 tok/s48.59 GB2L2 跨框架验证
NVIDIA RTX SparkOllamaQ4_K_M9.8 tok/s48.59 GB2L2 跨框架验证
AMD Radeon RX 7900 XTXStrataIQ1_S136.7 tok/s22.8 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395OllamaIQ1_S109.5 tok/s25.91 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395OllamaQ3_K_S82.9 tok/s38.96 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395StrataQ3_K_L119.8 tok/s42.92 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395StrataQ4_K_M114.2 tok/s48.59 GB1L2 跨框架验证
Apple M5 Maxllama.cppIQ3_XS171.7 tok/s37.25 GB1L2 跨框架验证
Apple M5 Maxllama.cppQ4_K_S147 tok/s46.33 GB1L2 跨框架验证
Apple M5 MaxOllamaQ5_K_M119.4 tok/s54.83 GB1L2 跨框架验证
Apple M5 MaxOllamaQ6_K100.2 tok/s58.8 GB1L2 跨框架验证
NVIDIA DGX SparkOllamaIQ3_S120 tok/s39.52 GB1L2 跨框架验证
NVIDIA DGX SparkStrataQ4_K_S169.7 tok/s46.33 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBvLLMQ5_K_S40.6 tok/s53.13 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBOllamaIQ3_XS49.6 tok/s37.25 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)OllamaIQ3_S74.7 tok/s39.52 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)StrataIQ3_XS105.8 tok/s37.25 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)StrataQ5_K_M71 tok/s54.83 GB1L2 跨框架验证
2x AMD Strix HaloOllamaIQ2_M146.4 tok/s31.58 GB1L2 跨框架验证
2x AMD Strix HaloStrataIQ2_XS213.8 tok/s29.32 GB1L2 跨框架验证
Apple M1 ProOllamaIQ2_M70.6 tok/s30.4 GB1L2 跨框架验证
Apple M1 MaxvLLMQ6_K47.8 tok/s58.8 GB1L2 跨框架验证
Apple M2 UltraOllamaQ8_049.8 tok/s67.3 GB1L2 跨框架验证
Apple M2 UltraStrataQ4_K_M113.8 tok/s48.59 GB1L2 跨框架验证
NVIDIA RTX 4090vLLMIQ1_S153.3 tok/s22.8 GB1L2 跨框架验证

其中 205 条实测未关联到具体量化版本,完整数据见搜索页。

社区讨论

查看全部讨论