gemma-4-E2B-it

对话由 Google 发布 发起讨论

0

量化版本

53

实测记录

1

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 5090llama.cppQ3_K_L324.4 tok/s5.12 GB1L2 跨框架验证
NVIDIA RTX 5090StrataIQ3_S332.3 tok/s4.87 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395llama.cppQ3_K_M208.8 tok/s4.96 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395vLLMIQ1_S291.6 tok/s3.86 GB1L2 跨框架验证
Apple M5 MaxvLLMQ3_K_M285.3 tok/s4.96 GB1L2 跨框架验证
Apple M5 MaxStrataQ3_K_L329.6 tok/s5.12 GB1L2 跨框架验证
Apple M5 MaxStrataQ8_0323.1 tok/s6.93 GB1L2 跨框架验证
NVIDIA DGX SparkvLLMIQ2_M380.9 tok/s4.28 GB1L2 跨框架验证
NVIDIA DGX SparkvLLMIQ3_XS360 tok/s4.7 GB1L2 跨框架验证
NVIDIA DGX SparkStrataQ3_K_L389.3 tok/s5.12 GB1L2 跨框架验证
NVIDIA DGX SparkStrataQ4_K_M326.7 tok/s5.54 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBOllamaQ6_K70.7 tok/s6.3 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBStrataIQ2_M176.2 tok/s4.28 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBStrataIQ3_S167.9 tok/s4.87 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)OllamaFP1670.6 tok/s9.87 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)StrataQ3_K_S228.5 tok/s4.83 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)StrataQ6_K157.8 tok/s6.3 GB1L2 跨框架验证
2x AMD Strix HalovLLMQ3_K_S277.8 tok/s4.83 GB1L2 跨框架验证
2x AMD Strix HaloStrataIQ3_XS321.6 tok/s4.7 GB1L2 跨框架验证
AMD Strix Point (Radeon 890M)llama.cppIQ1_S145.7 tok/s3.86 GB1L2 跨框架验证
AMD Strix Point (Radeon 890M)llama.cppIQ4_XS106 tok/s5.21 GB1L2 跨框架验证
AMD Strix Point (Radeon 890M)OllamaFP1647.3 tok/s9.87 GB1L2 跨框架验证
AMD Strix Point (Radeon 890M)OllamaQ8_057.3 tok/s6.93 GB1L2 跨框架验证
Intel Arc A770 16GBOllamaQ6_K63.9 tok/s6.3 GB1L2 跨框架验证
Intel Arc A770 16GBStrataQ4_K_M120.1 tok/s5.54 GB1L2 跨框架验证
Intel Arc 140Vllama.cppQ4_K_M78.9 tok/s5.54 GB1L2 跨框架验证
Intel Arc 140Vllama.cppQ8_053.1 tok/s6.93 GB1L2 跨框架验证
Intel Arc 140VvLLMQ4_K_M93.7 tok/s5.54 GB1L2 跨框架验证
Apple M1llama.cppIQ2_XS125.4 tok/s4.12 GB1L2 跨框架验证
Apple M1llama.cppQ6_K62.5 tok/s6.3 GB1L2 跨框架验证
Apple M1OllamaIQ2_XS113.2 tok/s4.12 GB1L2 跨框架验证
Apple M1OllamaIQ3_S89.3 tok/s4.87 GB1L2 跨框架验证
Apple M1 Prollama.cppQ3_K_M141.1 tok/s4.96 GB1L2 跨框架验证
Apple M1 Prollama.cppQ6_K93.8 tok/s6.3 GB1L2 跨框架验证
Apple M1 ProOllamaQ3_K_L126.4 tok/s5.12 GB1L2 跨框架验证
Apple M1 ProOllamaQ5_K_M87.8 tok/s6.01 GB1L2 跨框架验证
Apple M1 MaxvLLMQ3_K_S189.6 tok/s4.83 GB1L2 跨框架验证
Apple M1 MaxvLLMQ8_098.8 tok/s6.93 GB1L2 跨框架验证
Apple M1 MaxStrataIQ1_S324.3 tok/s3.86 GB1L2 跨框架验证
Apple M2 ProvLLMQ4_K_M128.4 tok/s5.54 GB1L2 跨框架验证
Apple M2 ProOllamaIQ2_XS158 tok/s4.12 GB1L2 跨框架验证
Apple M2 ProOllamaQ3_K_S140 tok/s4.83 GB1L2 跨框架验证
Apple M2 Ultrallama.cppIQ2_XS255.2 tok/s4.12 GB1L2 跨框架验证
Apple M2 Ultrallama.cppQ3_K_S222.2 tok/s4.83 GB1L2 跨框架验证
Apple M2 UltraStrataQ3_K_L291.9 tok/s5.12 GB1L2 跨框架验证
Qualcomm Snapdragon X EliteOllamaQ4_K_M76.5 tok/s5.54 GB1L2 跨框架验证
Qualcomm Snapdragon X EliteStrataQ2_K154.6 tok/s4.41 GB1L2 跨框架验证
Qualcomm Snapdragon X EliteStrataQ3_K_L121.8 tok/s5.12 GB1L2 跨框架验证
Qualcomm Snapdragon 7 Gen 4vLLMQ3_K_M80.6 tok/s4.96 GB1L2 跨框架验证
Qualcomm Snapdragon 7 Gen 4vLLMQ8_046.5 tok/s6.93 GB1L2 跨框架验证

其中 53 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