mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-09-27 21:46:57 +02:00
reflect the plamo2 graph fix, which changed its fusion pattern split (RMS_NORM+MUL 11->10, RMS_NORM+MUL+ADD 3->4; same total). Assisted-by: pi:llama.cpp/DeepSeek-V4-Flash-0731
11 KiB
11 KiB
| 1 | # test-fusion baseline for device MTL |
|---|---|
| 2 | # arch ,moe ,mode ,label , count |
| 3 | arcee ,0 ,any ,RMS_NORM+MUL , 5 |
| 4 | arctic ,0 ,any ,RMS_NORM+MUL , 7 |
| 5 | baichuan ,0 ,any ,RMS_NORM+MUL , 5 |
| 6 | bailingmoe ,1 ,any ,ADD+ADD , 2 |
| 7 | bailingmoe ,1 ,any ,RMS_NORM+MUL , 5 |
| 8 | bailingmoe2 ,1 ,any ,ADD+ADD , 1 |
| 9 | bailingmoe2 ,1 ,any ,RMS_NORM+MUL , 9 |
| 10 | bailingmoe3 ,1 ,any ,ADD+ADD , 1 |
| 11 | bailingmoe3 ,1 ,any ,GATED_DELTA_NET+CPY , 1 |
| 12 | bailingmoe3 ,1 ,any ,RMS_NORM+MUL , 8 |
| 13 | bloom ,0 ,any ,NORM+MUL+ADD , 6 |
| 14 | chatglm ,0 ,any ,RMS_NORM+MUL , 5 |
| 15 | codeshell ,0 ,any ,NORM+MUL+ADD , 5 |
| 16 | cogvlm ,0 ,any ,RMS_NORM+MUL , 5 |
| 17 | command-r ,0 ,any ,NORM+MUL , 3 |
| 18 | dbrx ,0 ,any ,NORM+MUL , 5 |
| 19 | deci ,0 ,any ,RMS_NORM+MUL , 5 |
| 20 | deepseek ,0 ,any ,ADD+ADD , 1 |
| 21 | deepseek ,0 ,any ,RMS_NORM+MUL , 5 |
| 22 | deepseek2 ,0 ,any ,ADD+ADD , 1 |
| 23 | deepseek2 ,0 ,any ,RMS_NORM+MUL , 9 |
| 24 | deepseek32 ,0 ,any ,ADD+ADD , 1 |
| 25 | deepseek32 ,0 ,any ,NORM+MUL+ADD , 2 |
| 26 | deepseek32 ,0 ,any ,RMS_NORM+MUL , 9 |
| 27 | deepseek4 ,0 ,any ,RMS_NORM+MUL , 20 |
| 28 | dots1 ,0 ,any ,ADD+ADD , 1 |
| 29 | dots1 ,0 ,any ,RMS_NORM+MUL , 9 |
| 30 | dream ,0 ,any ,RMS_NORM+MUL , 5 |
| 31 | ernie4_5-moe ,1 ,any ,ADD+ADD , 1 |
| 32 | ernie4_5-moe ,1 ,any ,RMS_NORM+MUL , 5 |
| 33 | ernie4_5 ,0 ,any ,RMS_NORM+MUL , 5 |
| 34 | exaone ,0 ,any ,RMS_NORM+MUL , 5 |
| 35 | exaone4 ,0 ,any ,RMS_NORM+MUL , 5 |
| 36 | exaone4 ,0 ,any ,RMS_NORM+MUL+ADD , 4 |
| 37 | falcon ,0 ,any ,ADD+ADD , 2 |
| 38 | falcon ,0 ,any ,NORM+MUL+ADD , 5 |
| 39 | falcon-h1 ,0 ,any ,ADD+ADD , 2 |
| 40 | falcon-h1 ,0 ,any ,RMS_NORM+MUL , 9 |
| 41 | gemma ,0 ,any ,RMS_NORM+MUL , 5 |
| 42 | gemma2 ,0 ,any ,RMS_NORM+MUL , 5 |
| 43 | gemma2 ,0 ,any ,RMS_NORM+MUL+ADD , 4 |
| 44 | glm-dsa ,0 ,any ,ADD+ADD , 1 |
| 45 | glm-dsa ,0 ,any ,NORM+MUL+ADD , 2 |
| 46 | glm-dsa ,0 ,any ,RMS_NORM+MUL , 9 |
| 47 | glm4 ,0 ,any ,RMS_NORM+MUL , 5 |
| 48 | glm4 ,0 ,any ,RMS_NORM+MUL+ADD , 4 |
| 49 | glm4moe ,1 ,any ,ADD+ADD , 1 |
| 50 | glm4moe ,1 ,any ,RMS_NORM+MUL , 9 |
| 51 | gpt-oss ,0 ,any ,RMS_NORM+MUL , 5 |
| 52 | gpt2 ,0 ,any ,NORM+MUL+ADD , 5 |
| 53 | gptneox ,0 ,any ,NORM+MUL+ADD , 5 |
| 54 | granite ,0 ,any ,RMS_NORM+MUL , 5 |
| 55 | granite ,0 ,any ,RMS_NORM+MUL , 5 |
| 56 | granitehybrid ,0 ,any ,RMS_NORM+MUL , 6 |
| 57 | granitemoe ,1 ,any ,RMS_NORM+MUL , 5 |
| 58 | granitemoe ,1 ,any ,RMS_NORM+MUL , 5 |
| 59 | grok ,0 ,any ,RMS_NORM+MUL , 5 |
| 60 | grok ,0 ,any ,RMS_NORM+MUL+ADD , 4 |
| 61 | grovemoe ,1 ,any ,ADD+ADD , 2 |
| 62 | grovemoe ,1 ,any ,RMS_NORM+MUL , 9 |
| 63 | hunyuan-dense ,0 ,any ,RMS_NORM+MUL , 9 |
| 64 | hunyuan-moe ,1 ,any ,ADD+ADD , 2 |
| 65 | hunyuan-moe ,1 ,any ,RMS_NORM+MUL , 9 |
| 66 | hunyuan_vl ,0 ,any ,RMS_NORM+MUL , 9 |
| 67 | hy_v3 ,0 ,any ,ADD+ADD , 2 |
| 68 | hy_v3 ,0 ,any ,RMS_NORM+MUL , 9 |
| 69 | internlm2 ,0 ,any ,RMS_NORM+MUL , 5 |
| 70 | jais ,0 ,any ,NORM+MUL+ADD , 5 |
| 71 | jais2 ,0 ,any ,NORM+MUL+ADD , 5 |
| 72 | jamba ,0 ,any ,RMS_NORM+MUL , 8 |
| 73 | kimi-k3 ,0 ,any ,GATED_DELTA_NET+CPY , 1 |
| 74 | kimi-k3 ,0 ,any ,RMS_NORM+MUL , 17 |
| 75 | kimi-linear ,0 ,any ,ADD+ADD , 1 |
| 76 | kimi-linear ,0 ,any ,GATED_DELTA_NET+CPY , 1 |
| 77 | kimi-linear ,0 ,any ,RMS_NORM+MUL , 7 |
| 78 | lfm2 ,0 ,any ,RMS_NORM+MUL , 7 |
| 79 | lfm2moe ,1 ,any ,RMS_NORM+MUL , 7 |
| 80 | llada ,0 ,any ,RMS_NORM+MUL , 5 |
| 81 | llada-moe ,1 ,any ,RMS_NORM+MUL , 9 |
| 82 | llama ,0 ,any ,RMS_NORM+MUL , 5 |
| 83 | llama ,0 ,any ,RMS_NORM+MUL , 5 |
| 84 | llama4 ,0 ,any ,ADD+ADD , 2 |
| 85 | llama4 ,0 ,any ,RMS_NORM+MUL , 9 |
| 86 | maincoder ,0 ,any ,RMS_NORM+MUL , 9 |
| 87 | mamba ,0 ,any ,RMS_NORM+MUL , 3 |
| 88 | mamba2 ,0 ,any ,RMS_NORM+MUL , 5 |
| 89 | minicpm ,0 ,any ,RMS_NORM+MUL , 5 |
| 90 | minicpm ,0 ,any ,RMS_NORM+MUL , 5 |
| 91 | minicpm3 ,0 ,any ,RMS_NORM+MUL , 9 |
| 92 | minimax-01 ,0 ,any ,RMS_NORM+MUL , 6 |
| 93 | minimax-m2 ,0 ,any ,RMS_NORM+MUL , 9 |
| 94 | minimax-m3 ,0 ,any ,ADD+ADD , 1 |
| 95 | minimax-m3 ,0 ,any ,RMS_NORM+MUL , 11 |
| 96 | mistral3 ,0 ,any ,RMS_NORM+MUL , 5 |
| 97 | mistral3 ,0 ,any ,RMS_NORM+MUL , 5 |
| 98 | mistral4 ,0 ,any ,ADD+ADD , 1 |
| 99 | mistral4 ,0 ,any ,RMS_NORM+MUL , 9 |
| 100 | mpt ,0 ,any ,NORM+MUL+ADD , 5 |
| 101 | nanbeige ,0 ,any ,RMS_NORM+MUL , 5 |
| 102 | nemotron ,0 ,any ,NORM+MUL+ADD , 5 |
| 103 | nemotron_h ,0 ,any ,RMS_NORM+MUL , 5 |
| 104 | nemotron_h_moe ,1 ,any ,RMS_NORM+MUL , 5 |
| 105 | olmoe ,1 ,any ,RMS_NORM+MUL , 9 |
| 106 | openelm ,0 ,any ,RMS_NORM+MUL , 9 |
| 107 | orion ,0 ,any ,NORM+MUL+ADD , 5 |
| 108 | paddleocr ,0 ,any ,RMS_NORM+MUL , 5 |
| 109 | pangu-embedded ,0 ,any ,RMS_NORM+MUL , 5 |
| 110 | phi2 ,0 ,any ,ADD+ADD , 2 |
| 111 | phi2 ,0 ,any ,NORM+MUL+ADD , 3 |
| 112 | phi3 ,0 ,any ,RMS_NORM+MUL , 5 |
| 113 | phimoe ,1 ,any ,RMS_NORM+MUL+ADD , 5 |
| 114 | plamo ,0 ,any ,ADD+ADD , 2 |
| 115 | plamo ,0 ,any ,RMS_NORM+MUL , 3 |
| 116 | plamo2 ,0 ,any ,RMS_NORM+MUL , 10 |
| 117 | plamo2 ,0 ,any ,RMS_NORM+MUL+ADD , 4 |
| 118 | pockettts ,0 ,any ,NORM+MUL+ADD , 5 |
| 119 | qwen ,0 ,any ,RMS_NORM+MUL , 5 |
| 120 | qwen2 ,0 ,any ,RMS_NORM+MUL , 5 |
| 121 | qwen2moe ,1 ,any ,ADD+ADD , 2 |
| 122 | qwen2moe ,1 ,any ,RMS_NORM+MUL , 5 |
| 123 | qwen2vl ,0 ,any ,RMS_NORM+MUL , 5 |
| 124 | qwen3 ,0 ,any ,RMS_NORM+MUL , 9 |
| 125 | qwen35 ,0 ,any ,GATED_DELTA_NET+CPY , 1 |
| 126 | qwen35 ,0 ,any ,RMS_NORM+MUL , 8 |
| 127 | qwen35moe ,1 ,any ,ADD+ADD , 2 |
| 128 | qwen35moe ,1 ,any ,GATED_DELTA_NET+CPY , 1 |
| 129 | qwen35moe ,1 ,any ,RMS_NORM+MUL , 8 |
| 130 | qwen3moe ,1 ,any ,RMS_NORM+MUL , 9 |
| 131 | qwen3next ,0 ,any ,ADD+ADD , 2 |
| 132 | qwen3next ,0 ,any ,GATED_DELTA_NET+CPY , 1 |
| 133 | qwen3next ,0 ,any ,RMS_NORM+MUL , 8 |
| 134 | qwen3tts ,0 ,any ,RMS_NORM+MUL , 9 |
| 135 | qwen3vl ,0 ,any ,RMS_NORM+MUL , 9 |
| 136 | qwen3vlmoe ,1 ,any ,RMS_NORM+MUL , 9 |
| 137 | qwen4exp ,0 ,any ,ADD+ADD+ADD , 5 |
| 138 | qwen4exp ,0 ,any ,ADD+ADD+ADD+ADD+ADD+ADD+ADD , 9 |
| 139 | qwen4exp ,0 ,any ,GATED_DELTA_NET+CPY , 1 |
| 140 | qwen4exp ,0 ,any ,RMS_NORM+MUL , 5 |
| 141 | refact ,0 ,any ,RMS_NORM+MUL , 5 |
| 142 | refact ,0 ,any ,RMS_NORM+MUL , 5 |
| 143 | rnd1 ,0 ,any ,RMS_NORM+MUL , 9 |
| 144 | seed_oss ,0 ,any ,RMS_NORM+MUL , 5 |
| 145 | smallthinker ,0 ,any ,RMS_NORM+MUL , 5 |
| 146 | smollm3 ,0 ,any ,RMS_NORM+MUL , 5 |
| 147 | stablelm ,0 ,any ,NORM+MUL , 4 |
| 148 | stablelm ,0 ,any ,NORM+MUL+ADD , 5 |
| 149 | starcoder ,0 ,any ,NORM+MUL+ADD , 5 |
| 150 | starcoder2 ,0 ,any ,NORM+MUL+ADD , 5 |
| 151 | talkie ,0 ,any ,ADD+ADD , 2 |
| 152 | xverse ,0 ,any ,RMS_NORM+MUL , 5 |