Files
Niklas Wenzel 79bce0c881 tests : enable Gemma 4 in test-llama-archs (#31401)
* tests : enable Gemma 4 in `test-llama-archs`

* tests : use GQA for Gemma 4 in `test-llama-archs`
2026-10-31 23:55:02 +01:00

21 KiB

1# test-fusion baseline for device MTL
2# arch ,moe ,mode ,label , count
3afmoe ,1 ,any ,MUL+ADD , 2
4afmoe ,1 ,any ,RMS_NORM+MUL , 10
5afmoe ,1 ,any ,RMS_NORM+MUL+ADD , 3
6arcee ,0 ,any ,RMS_NORM+MUL , 5
7arctic ,0 ,any ,MUL+ADD , 4
8arctic ,0 ,any ,RMS_NORM+MUL , 7
9arctic ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
10baichuan ,0 ,any ,RMS_NORM+MUL , 5
11bailingmoe ,1 ,any ,ADD+ADD , 2
12bailingmoe ,1 ,any ,MUL+ADD , 4
13bailingmoe ,1 ,any ,RMS_NORM+MUL , 5
14bailingmoe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS , 2
15bailingmoe2 ,1 ,any ,ADD+ADD , 1
16bailingmoe2 ,1 ,any ,MUL+ADD , 2
17bailingmoe2 ,1 ,any ,RMS_NORM+MUL , 9
18bailingmoe3 ,1 ,any ,ADD+ADD , 1
19bailingmoe3 ,1 ,any ,GATED_DELTA_NET+CPY , 1
20bailingmoe3 ,1 ,any ,MUL+ADD , 2
21bailingmoe3 ,1 ,any ,RMS_NORM+MUL , 8
22bailingmoe3 ,1 ,any ,RMS_NORM+SCALE , 2
23beam ,0 ,any ,MUL+ADD , 6
24beam ,0 ,any ,RMS_NORM+MUL , 25
25bloom ,0 ,any ,NORM+MUL+ADD , 6
26chatglm ,0 ,any ,RMS_NORM+MUL , 5
27codeshell ,0 ,any ,NORM+MUL+ADD , 5
28cogvlm ,0 ,any ,RMS_NORM+MUL , 5
29cohere2 ,0 ,any ,ADD+ADD , 2
30cohere2 ,0 ,any ,NORM+MUL , 3
31cohere2moe ,1 ,any ,ADD+ADD , 2
32cohere2moe ,1 ,any ,MUL+ADD , 2
33cohere2moe ,1 ,any ,RMS_NORM+MUL , 3
34command-r ,0 ,any ,NORM+MUL , 3
35dbrx ,0 ,any ,MUL+ADD , 4
36dbrx ,0 ,any ,NORM+MUL , 5
37dbrx ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
38deci ,0 ,any ,RMS_NORM+MUL , 5
39deepseek ,0 ,any ,ADD+ADD , 1
40deepseek ,0 ,any ,MUL+ADD , 2
41deepseek ,0 ,any ,RMS_NORM+MUL , 5
42deepseek ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS , 1
43deepseek2 ,0 ,any ,ADD+ADD , 1
44deepseek2 ,0 ,any ,MUL+ADD , 2
45deepseek2 ,0 ,any ,RMS_NORM+MUL , 9
46deepseek32 ,0 ,any ,ADD+ADD , 1
47deepseek32 ,0 ,any ,MUL+ADD , 2
48deepseek32 ,0 ,any ,NORM+MUL+ADD , 2
49deepseek32 ,0 ,any ,RMS_NORM+MUL , 9
50deepseek4 ,0 ,any ,MUL+ADD , 8
51deepseek4 ,0 ,any ,RMS_NORM+MUL , 20
52deepseek41 ,0 ,any ,MUL+ADD , 12
53deepseek41 ,0 ,any ,RMS_NORM+MUL , 29
54dots1 ,0 ,any ,ADD+ADD , 1
55dots1 ,0 ,any ,MUL+ADD , 2
56dots1 ,0 ,any ,RMS_NORM+MUL , 9
57dots3note ,0 ,any ,ADD+ADD , 1
58dots3note ,0 ,any ,MUL+ADD , 2
59dots3note ,0 ,any ,NORM+MUL+ADD , 1
60dots3note ,0 ,any ,RMS_NORM+MUL , 11
61dream ,0 ,any ,RMS_NORM+MUL , 5
62ernie4_5-moe ,1 ,any ,ADD+ADD , 1
63ernie4_5-moe ,1 ,any ,MUL+ADD , 2
64ernie4_5-moe ,1 ,any ,RMS_NORM+MUL , 5
65ernie4_5 ,0 ,any ,RMS_NORM+MUL , 5
66exaone ,0 ,any ,RMS_NORM+MUL , 5
67exaone-moe ,1 ,any ,ADD+ADD , 1
68exaone-moe ,1 ,any ,MUL+ADD , 2
69exaone-moe ,1 ,any ,RMS_NORM+MUL , 9
70exaone4 ,0 ,any ,RMS_NORM+MUL , 5
71exaone4 ,0 ,any ,RMS_NORM+MUL+ADD , 4
72falcon ,0 ,any ,ADD+ADD , 2
73falcon ,0 ,any ,NORM+MUL+ADD , 5
74falcon-h1 ,0 ,any ,ADD+ADD , 2
75falcon-h1 ,0 ,any ,RMS_NORM+MUL , 9
76flint ,0 ,any ,ADD+ADD , 1
77flint ,0 ,any ,MUL+ADD , 2
78flint ,0 ,any ,RMS_NORM+MUL , 9
79gemma ,0 ,any ,RMS_NORM+MUL , 5
80gemma2 ,0 ,any ,RMS_NORM+MUL , 5
81gemma2 ,0 ,any ,RMS_NORM+MUL+ADD , 4
82gemma3 ,0 ,any ,RMS_NORM+MUL , 9
83gemma3 ,0 ,any ,RMS_NORM+MUL+ADD , 4
84gemma4 ,0 ,any ,RMS_NORM+MUL , 47
85gemma4 ,0 ,any ,RMS_NORM+MUL+ADD , 20
86glm-dsa ,0 ,any ,ADD+ADD , 1
87glm-dsa ,0 ,any ,MUL+ADD , 2
88glm-dsa ,0 ,any ,NORM+MUL+ADD , 2
89glm-dsa ,0 ,any ,RMS_NORM+MUL , 9
90glm4 ,0 ,any ,RMS_NORM+MUL , 5
91glm4 ,0 ,any ,RMS_NORM+MUL+ADD , 4
92glm4moe ,1 ,any ,ADD+ADD , 1
93glm4moe ,1 ,any ,MUL+ADD , 2
94glm4moe ,1 ,any ,RMS_NORM+MUL , 9
95glm5-next ,0 ,any ,ADD+ADD+ADD , 1
96glm5-next ,0 ,any ,GATED_DELTA_NET+CPY , 1
97glm5-next ,0 ,any ,MUL+ADD , 2
98glm5-next ,0 ,any ,NORM+MUL+ADD , 1
99glm5-next ,0 ,any ,RMS_NORM+MUL , 8
100glm5-next ,0 ,any ,RMS_NORM+SCALE , 2
101gpt-oss ,0 ,any ,MUL+ADD , 4
102gpt-oss ,0 ,any ,RMS_NORM+MUL , 5
103gpt2 ,0 ,any ,NORM+MUL+ADD , 5
104gptneox ,0 ,any ,NORM+MUL+ADD , 5
105granite ,0 ,any ,RMS_NORM+MUL , 5
106granite ,0 ,any ,MUL+ADD , 4
107granite ,0 ,any ,RMS_NORM+MUL , 5
108granite ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
109granite_swa ,0 ,any ,RMS_NORM+MUL , 5
110granitehybrid ,0 ,any ,RMS_NORM+MUL , 6
111granitemoe ,1 ,any ,RMS_NORM+MUL , 5
112granitemoe ,1 ,any ,MUL+ADD , 4
113granitemoe ,1 ,any ,RMS_NORM+MUL , 5
114granitemoe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
115grok ,0 ,any ,MUL+ADD , 4
116grok ,0 ,any ,RMS_NORM+MUL , 5
117grok ,0 ,any ,RMS_NORM+MUL+ADD , 4
118grok ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
119grovemoe ,1 ,any ,ADD+ADD , 2
120grovemoe ,1 ,any ,MUL+ADD , 8
121grovemoe ,1 ,any ,RMS_NORM+MUL , 9
122hunyuan-dense ,0 ,any ,RMS_NORM+MUL , 9
123hunyuan-moe ,1 ,any ,ADD+ADD , 2
124hunyuan-moe ,1 ,any ,MUL+ADD , 4
125hunyuan-moe ,1 ,any ,RMS_NORM+MUL , 9
126hunyuan-moe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
127hunyuan_vl ,0 ,any ,RMS_NORM+MUL , 9
128hy_v3 ,0 ,any ,ADD+ADD , 2
129hy_v3 ,0 ,any ,MUL+ADD , 4
130hy_v3 ,0 ,any ,RMS_NORM+MUL , 9
131hy_v4 ,0 ,any ,MUL+ADD , 2
132hy_v4 ,0 ,any ,NORM+MUL+ADD , 1
133hy_v4 ,0 ,any ,RMS_NORM+MUL , 9
134internlm2 ,0 ,any ,RMS_NORM+MUL , 5
135jais ,0 ,any ,NORM+MUL+ADD , 5
136jais2 ,0 ,any ,NORM+MUL+ADD , 5
137jamba ,0 ,any ,RMS_NORM+MUL , 8
138k2-horizon ,0 ,any ,RMS_NORM+MUL , 4
139k2-horizon ,0 ,any ,ADD+ADD , 1
140k2-horizon ,0 ,any ,MUL+ADD , 2
141k2-horizon ,0 ,any ,RMS_NORM+MUL , 4
142kimi-k3 ,0 ,any ,GATED_DELTA_NET+CPY , 1
143kimi-k3 ,0 ,any ,MUL+ADD , 2
144kimi-k3 ,0 ,any ,RMS_NORM+MUL , 17
145kimi-k3 ,0 ,any ,RMS_NORM+SCALE , 2
146kimi-linear ,0 ,any ,ADD+ADD , 1
147kimi-linear ,0 ,any ,GATED_DELTA_NET+CPY , 1
148kimi-linear ,0 ,any ,MUL+ADD , 2
149kimi-linear ,0 ,any ,RMS_NORM+MUL , 7
150kimi-linear ,0 ,any ,RMS_NORM+SCALE , 2
151kolibri1 ,0 ,any ,MUL+ADD , 4
152kolibri1 ,0 ,any ,RMS_NORM+MUL , 10
153kolibri1 ,0 ,any ,RMS_NORM+MUL+ADD , 3
154laguna ,0 ,any ,ADD+ADD , 1
155laguna ,0 ,any ,MUL+ADD , 2
156laguna ,0 ,any ,RMS_NORM+MUL , 9
157lfm2 ,0 ,any ,RMS_NORM+MUL , 7
158lfm2moe ,1 ,any ,MUL+ADD , 2
159lfm2moe ,1 ,any ,RMS_NORM+MUL , 7
160lfm3 ,0 ,any ,RMS_NORM+MUL , 17
161llada ,0 ,any ,RMS_NORM+MUL , 5
162llada-moe ,1 ,any ,MUL+ADD , 4
163llada-moe ,1 ,any ,RMS_NORM+MUL , 9
164llada-moe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS , 2
165llama ,0 ,any ,RMS_NORM+MUL , 5
166llama ,0 ,any ,MUL+ADD , 4
167llama ,0 ,any ,RMS_NORM+MUL , 5
168llama ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
169llama4 ,0 ,any ,ADD+ADD , 2
170llama4 ,0 ,any ,RMS_NORM+MUL , 9
171maincoder ,0 ,any ,RMS_NORM+MUL , 9
172maion-coder ,0 ,any ,MUL+ADD , 4
173maion-coder ,0 ,any ,RMS_NORM+MUL , 17
174maion-coder ,0 ,any ,RMS_NORM+MUL+ADD , 8
175maion-coder ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
176mamba ,0 ,any ,RMS_NORM+MUL , 3
177mamba2 ,0 ,any ,RMS_NORM+MUL , 5
178maple ,0 ,any ,MUL+ADD , 4
179maple ,0 ,any ,RMS_NORM+MUL , 9
180maple ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
181mellum ,0 ,any ,MUL+ADD , 4
182mellum ,0 ,any ,RMS_NORM+MUL , 9
183mellum ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
184mimo2 ,0 ,any ,MUL+ADD , 4
185mimo2 ,0 ,any ,RMS_NORM+MUL , 5
186minicpm ,0 ,any ,RMS_NORM+MUL , 5
187minicpm ,0 ,any ,MUL+ADD , 4
188minicpm ,0 ,any ,RMS_NORM+MUL , 5
189minicpm ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
190minicpm3 ,0 ,any ,RMS_NORM+MUL , 9
191minimax-01 ,0 ,any ,MUL+ADD , 4
192minimax-01 ,0 ,any ,RMS_NORM+MUL , 6
193minimax-01 ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
194minimax-m2 ,0 ,any ,MUL+ADD , 4
195minimax-m2 ,0 ,any ,RMS_NORM+MUL , 9
196minimax-m3 ,0 ,any ,ADD+ADD , 1
197minimax-m3 ,0 ,any ,MUL+ADD , 2
198minimax-m3 ,0 ,any ,RMS_NORM+MUL , 11
199mistral-large4 ,0 ,any ,ADD+ADD , 3
200mistral-large4 ,0 ,any ,MUL+ADD , 6
201mistral-large4 ,0 ,any ,NORM+MUL+ADD , 6
202mistral-large4 ,0 ,any ,RMS_NORM+MUL , 12
203mistral3 ,0 ,any ,RMS_NORM+MUL , 5
204mistral3 ,0 ,any ,MUL+ADD , 4
205mistral3 ,0 ,any ,RMS_NORM+MUL , 5
206mistral3 ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
207mistral4 ,0 ,any ,ADD+ADD , 1
208mistral4 ,0 ,any ,MUL+ADD , 2
209mistral4 ,0 ,any ,RMS_NORM+MUL , 9
210mpt ,0 ,any ,NORM+MUL+ADD , 5
211muse-glimmer ,0 ,any ,RMS_NORM+MUL , 10
212muse-glimmer ,0 ,any ,RMS_NORM+MUL+ADD , 3
213muse-spark ,0 ,any ,MUL+ADD , 4
214muse-spark ,0 ,any ,RMS_NORM+MUL , 14
215nanbeige ,0 ,any ,RMS_NORM+MUL , 5
216nemotron ,0 ,any ,NORM+MUL+ADD , 5
217nemotron_h ,0 ,any ,RMS_NORM+MUL , 5
218nemotron_h_moe ,1 ,any ,RMS_NORM+MUL , 5
219olmo2 ,0 ,any ,RMS_NORM+MUL , 5
220olmo2 ,0 ,any ,RMS_NORM+MUL+ADD , 4
221olmoe ,1 ,any ,MUL+ADD , 4
222olmoe ,1 ,any ,RMS_NORM+MUL , 9
223olmoe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS , 2
224openelm ,0 ,any ,RMS_NORM+MUL , 9
225orion ,0 ,any ,NORM+MUL+ADD , 5
226paddleocr ,0 ,any ,RMS_NORM+MUL , 5
227pangu-embedded ,0 ,any ,RMS_NORM+MUL , 5
228phi2 ,0 ,any ,ADD+ADD , 2
229phi2 ,0 ,any ,NORM+MUL+ADD , 3
230phi3 ,0 ,any ,RMS_NORM+MUL , 5
231phimoe ,1 ,any ,MUL+ADD , 4
232phimoe ,1 ,any ,RMS_NORM+MUL+ADD , 5
233phimoe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
234plamo ,0 ,any ,ADD+ADD , 2
235plamo ,0 ,any ,RMS_NORM+MUL , 3
236plamo2 ,0 ,any ,RMS_NORM+MUL , 10
237plamo2 ,0 ,any ,RMS_NORM+MUL+ADD , 4
238plamo2 ,0 ,any ,SSM_CONV+UNARY , 1
239plamo3 ,0 ,any ,RMS_NORM+MUL , 9
240plamo3 ,0 ,any ,RMS_NORM+MUL+ADD , 4
241pockettts ,0 ,any ,NORM+MUL+ADD , 5
242qwen ,0 ,any ,RMS_NORM+MUL , 5
243qwen2 ,0 ,any ,RMS_NORM+MUL , 5
244qwen2moe ,1 ,any ,ADD+ADD , 2
245qwen2moe ,1 ,any ,MUL+ADD , 4
246qwen2moe ,1 ,any ,RMS_NORM+MUL , 5
247qwen2moe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS , 2
248qwen2vl ,0 ,any ,RMS_NORM+MUL , 5
249qwen3 ,0 ,any ,RMS_NORM+MUL , 9
250qwen35 ,0 ,any ,GATED_DELTA_NET+CPY , 1
251qwen35 ,0 ,any ,RMS_NORM+MUL , 8
252qwen35 ,0 ,any ,RMS_NORM+SCALE , 2
253qwen35 ,0 ,any ,SSM_CONV+UNARY , 1
254qwen35moe ,1 ,any ,ADD+ADD , 2
255qwen35moe ,1 ,any ,GATED_DELTA_NET+CPY , 1
256qwen35moe ,1 ,any ,MUL+ADD , 4
257qwen35moe ,1 ,any ,RMS_NORM+MUL , 8
258qwen35moe ,1 ,any ,RMS_NORM+SCALE , 2
259qwen35moe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
260qwen35moe ,1 ,any ,SSM_CONV+UNARY , 1
261qwen3moe ,1 ,any ,MUL+ADD , 4
262qwen3moe ,1 ,any ,RMS_NORM+MUL , 9
263qwen3moe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
264qwen3next ,0 ,any ,ADD+ADD , 2
265qwen3next ,0 ,any ,GATED_DELTA_NET+CPY , 1
266qwen3next ,0 ,any ,MUL+ADD , 4
267qwen3next ,0 ,any ,RMS_NORM+MUL , 8
268qwen3next ,0 ,any ,RMS_NORM+SCALE , 2
269qwen3next ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
270qwen3next ,0 ,any ,SSM_CONV+UNARY , 1
271qwen3tts ,0 ,any ,RMS_NORM+MUL , 9
272qwen3vl ,0 ,any ,RMS_NORM+MUL , 9
273qwen3vlmoe ,1 ,any ,MUL+ADD , 4
274qwen3vlmoe ,1 ,any ,RMS_NORM+MUL , 9
275qwen3vlmoe ,1 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
276qwen4exp ,0 ,any ,ADD+ADD+ADD , 2
277qwen4exp ,0 ,any ,GATED_DELTA_NET+CPY , 1
278qwen4exp ,0 ,any ,MUL+ADD , 4
279qwen4exp ,0 ,any ,RMS_NORM+MUL , 13
280qwen4exp ,0 ,any ,RMS_NORM+SCALE , 2
281qwen4exp ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
282qwen4exp ,0 ,any ,SSM_CONV+UNARY , 1
283refact ,0 ,any ,RMS_NORM+MUL , 5
284refact ,0 ,any ,RMS_NORM+MUL , 5
285rnd1 ,0 ,any ,MUL+ADD , 4
286rnd1 ,0 ,any ,RMS_NORM+MUL , 9
287rnd1 ,0 ,any ,SOFT_MAX+ARGSORT+GET_ROWS+SUM_ROWS+CLAMP+DIV, 2
288seed_oss ,0 ,any ,RMS_NORM+MUL , 5
289smallthinker ,0 ,any ,MUL+ADD , 4
290smallthinker ,0 ,any ,RMS_NORM+MUL , 5
291smollm3 ,0 ,any ,RMS_NORM+MUL , 5
292spark2_5 ,0 ,any ,RMS_NORM+MUL , 5
293stablelm ,0 ,any ,NORM+MUL , 4
294stablelm ,0 ,any ,NORM+MUL+ADD , 5
295starcoder ,0 ,any ,NORM+MUL+ADD , 5
296starcoder2 ,0 ,any ,NORM+MUL+ADD , 5
297talkie ,0 ,any ,ADD+ADD , 2
298xverse ,0 ,any ,RMS_NORM+MUL , 5