gemma-4-E4B-it

8B对话由 Google 发布 发起讨论

0

量化版本

73

实测记录

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 4090OllamaQ3_K_M158.5 tok/s13.52 GB1L2 跨框架验证
NVIDIA RTX 4090StrataQ3_K_M243.8 tok/s13.52 GB1L2 跨框架验证
NVIDIA RTX 3090llama.cppIQ3_S121.9 tok/s13.19 GB1L2 跨框架验证
NVIDIA RTX 3090vLLMIQ4_XS102.6 tok/s14.53 GB1L2 跨框架验证
AMD Radeon RX 7900 XTXvLLMIQ3_XS133.2 tok/s12.52 GB1L2 跨框架验证
AMD Radeon RX 7900 XTXOllamaQ5_K_M71.3 tok/s17.72 GB1L2 跨框架验证
NVIDIA RTX 5080vLLMIQ2_XS301.3 tok/s10.16 GB1L2 跨框架验证
NVIDIA RTX 5080StrataIQ4_XS314.5 tok/s14.53 GB1L2 跨框架验证
NVIDIA RTX 3060 12GBvLLMIQ3_XS89 tok/s11.4 GB1L2 跨框架验证
NVIDIA RTX 3060 12GBOllamaQ3_K_S79.4 tok/s11.4 GB1L2 跨框架验证
NVIDIA RTX 4080 SuperOllamaQ2_K222.5 tok/s11.34 GB1L2 跨框架验证
Apple M5 Ultrallama.cppIQ3_S153.4 tok/s13.19 GB1L2 跨框架验证
Apple M5 UltraOllamaIQ2_M156.4 tok/s10.84 GB1L2 跨框架验证
NVIDIA H100 80GB HBM3llama.cppIQ3_XS297.4 tok/s12.52 GB1L2 跨框架验证
NVIDIA H100 80GB HBM3vLLMQ3_K_S343.3 tok/s13.02 GB1L2 跨框架验证
NVIDIA H20llama.cppIQ2_XS95.2 tok/s10.16 GB1L2 跨框架验证
NVIDIA H20vLLMQ3_K_M78.4 tok/s13.52 GB1L2 跨框架验证
NVIDIA RTX PRO 6000 Blackwell 96GBvLLMIQ3_XS344.9 tok/s12.52 GB1L2 跨框架验证
NVIDIA RTX PRO 6000 Blackwell 96GBOllamaIQ2_XS387.7 tok/s10.16 GB1L2 跨框架验证
Intel Arc B580llama.cppIQ3_XS70.8 tok/s11.4 GB1L2 跨框架验证
Intel Arc B580OllamaIQ3_S65.1 tok/s11.4 GB1L2 跨框架验证
NVIDIA RTX Sparkllama.cppIQ3_S105.3 tok/s13.19 GB1L2 跨框架验证
NVIDIA RTX SparkOllamaIQ2_M116.7 tok/s10.84 GB1L2 跨框架验证
NVIDIA RTX 5090vLLMQ3_K_L336.5 tok/s8.15 GB1L2 跨框架验证
NVIDIA RTX 5090vLLMQ3_K_S310.9 tok/s7.56 GB1L2 跨框架验证
NVIDIA RTX 5090StrataQ3_K_S326.1 tok/s7.56 GB1L2 跨框架验证
NVIDIA RTX 5090StrataQ5_K_M361.3 tok/s9.91 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395vLLMIQ4_XS142.3 tok/s8.32 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395vLLMQ3_K_S163.7 tok/s7.56 GB1L2 跨框架验证
AMD Ryzen AI Max+ 395StrataIQ2_XS273.7 tok/s6.13 GB1L2 跨框架验证
Apple M5 MaxOllamaIQ3_XS264.7 tok/s7.31 GB1L2 跨框架验证
Apple M5 MaxOllamaQ2_K303 tok/s6.72 GB1L2 跨框架验证
Apple M5 MaxStrataFP16155.1 tok/s17.64 GB1L2 跨框架验证
Apple M5 MaxStrataQ3_K_L374.8 tok/s8.15 GB1L2 跨框架验证
NVIDIA DGX Sparkllama.cppQ2_K258.2 tok/s6.72 GB1L2 跨框架验证
NVIDIA DGX SparkvLLMFP16107.3 tok/s17.64 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBvLLMIQ3_S89.5 tok/s7.64 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBvLLMQ8_046.8 tok/s11.76 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBOllamaIQ2_M83.5 tok/s6.47 GB1L2 跨框架验证
NVIDIA Jetson AGX Orin 64GBOllamaIQ3_S82 tok/s7.64 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)llama.cppIQ2_M142.2 tok/s6.47 GB1L2 跨框架验证
AMD Strix Halo (Radeon 8060S)StrataQ8_092.3 tok/s11.76 GB1L2 跨框架验证
2x AMD Strix HalovLLMIQ3_S197.3 tok/s7.64 GB1L2 跨框架验证
2x AMD Strix HalovLLMQ5_K_M149.2 tok/s9.91 GB1L2 跨框架验证
2x AMD Strix HaloStrataQ2_K311.8 tok/s6.72 GB1L2 跨框架验证
AMD Strix Point (Radeon 890M)vLLMIQ3_XS93.7 tok/s7.31 GB1L2 跨框架验证
AMD Strix Point (Radeon 890M)vLLMQ6_K56.3 tok/s10.5 GB1L2 跨框架验证
AMD Strix Point (Radeon 890M)OllamaQ2_K85.3 tok/s6.72 GB1L2 跨框架验证
Intel Arc A770 16GBllama.cppQ3_K_M76.3 tok/s7.81 GB1L2 跨框架验证
Intel Arc A770 16GBllama.cppQ4_K_S63.5 tok/s8.65 GB1L2 跨框架验证

其中 73 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

还没有讨论——发起第一个话题吧。

发起讨论