AIKGBench Dynamic Shape 算子生成成功率报告

概述

AIKG 在 AIKGBench Dynamic Shape 基准测试中的算子生成成功率统计(pass@4)。

成功率统计

• Ascend: 161/200 (80.5%)
• CUDA: 待测试

详细结果

序号	算子名称	Ascend	CUDA
1	attention_BNSD	✗
2	attention_BSH	✗
3	attention_flash_score	✗
4	attention_flash	✗
5	attention_flash_with_causal	✗
6	attention_flash_with_dropout	✗
7	attention_gqa	✗
8	attention_mqa	✗
9	attention_rope	✗
10	attention_scaled_dot_product_large	✗
11	attention_scaled_dot_product_small	✗
12	attention_scaled_dot_product	✗
13	attention_TH	✗
14	elemwise_abs	✓
15	elemwise_add_001	✓
16	elemwise_add_002	✓
17	elemwise_add_003	✓
18	elemwise_addcdiv	✓
19	elemwise_addcmul	✓
20	elemwise_atan	✓
21	elemwise_atan2	✓
22	elemwise_bitwise_and	✓
23	elemwise_bitwise_not	✓
24	elemwise_bitwise_or	✓
25	elemwise_bitwise_xor	✓
26	elemwise_broadcast	✓
27	elemwise_cast_full	✓
28	elemwise_cast	✓
29	elemwise_cdiv	✓
30	elemwise_ceil	✓
31	elemwise_clamp	✓
32	elemwise_cos	✓
33	elemwise_count_dim0	✓
34	elemwise_count_dim1	✓
35	elemwise_div_001	✓
36	elemwise_equal	✓
37	elemwise_exp_001	✓
38	elemwise_expand	✓
39	elemwise_floor_div	✓
40	elemwise_floor	✓
41	elemwise_gelu_001	✓
42	elemwise_greater_equal	✓
43	elemwise_greater	✓
44	elemwise_hard_sigmoid_001	✓
45	elemwise_hard_swish_001	✓
46	elemwise_i2f	✓
47	elemwise_is_finite	✓
48	elemwise_is_inf	✓
49	elemwise_is_nan	✓
50	elemwise_leaky_relu_001	✓
51	elemwise_less_equal	✓
52	elemwise_less	✓
53	elemwise_linspace	✓
54	elemwise_log	✓
55	elemwise_log2	✓
56	elemwise_logical_and	✓
57	elemwise_logical_not	✓
58	elemwise_logical_or	✓
59	elemwise_maximum	✓
60	elemwise_minimum	✓
61	elemwise_mul_001	✓
62	elemwise_ne	✓
63	elemwise_neg	✓
64	elemwise_pow_001	✓
65	elemwise_pow_002	✓
66	elemwise_precise_div	✓
67	elemwise_relu_001	✓
68	elemwise_rope_001	✗
69	elemwise_round	✓
70	elemwise_rsqrt	✓
71	elemwise_sigmoid_001	✓
72	elemwise_sigmoid	✓
73	elemwise_silu_001	✓
74	elemwise_silu	✓
75	elemwise_sin	✓
76	elemwise_softplus_001	✓
77	elemwise_softsign_001	✓
78	elemwise_sqrt	✓
79	elemwise_sub_001	✓
80	elemwise_sub_002	✓
81	elemwise_swish_001	✓
82	elemwise_tan	✓
83	elemwise_tanh_001	✓
84	fused_add_layer_norm	✓
85	fused_add_rms_norm	✓
86	fused_apply_rotary_pos_emb	✗
87	fused_ffn	✗
88	fused_gelu_and_mul	✓
89	fused_moe_gating_topk_softmax	✗
90	fused_silu_and_mul	✓
91	index_feeds_repeat	✗
92	index_gather	✓
93	index_masked_select	✗
94	index_scatter_add_with_sorted	✓
95	index_scatter_elements	✗
96	index_select	✓
97	index_topk	✗
98	norm_add_layer_quant	✗
99	norm_add_layer	✓
100	norm_add_rms_cast	✓
101	norm_add_rms_dynamic_quant	✓
102	norm_add_rms_quant	✓
103	norm_add_rms	✓
104	norm_batch	✓
105	norm_deep	✓
106	norm_dua_quantize_add_layer	✓
107	norm_group_silu	✓
108	norm_group_swish	✓
109	norm_group	✓
110	norm_inplace_add_layer	✗
111	norm_inplace_add_rms	✗
112	norm_instance	✓
113	norm_layer_001_large	✓
114	norm_layer_001_small	✓
115	norm_layer_001	✓
116	norm_layer	✓
117	norm_quantize_add_layer	✓
118	norm_rms_001_large_std	✓
119	norm_rms_001_large	✓
120	norm_rms_001_small_std	✓
121	norm_rms_001_small	✓
122	norm_rms_001_std	✓
123	norm_rms_001	✓
124	norm_rms	✓
125	reduction_amax	✓
126	reduction_amin	✓
127	reduction_argmax_001	✓
128	reduction_argmax_002	✓
129	reduction_argmax	✓
130	reduction_argmin_001	✓
131	reduction_argmin_002	✓
132	reduction_count_nonzero	✓
133	reduction_count_vector	✓
134	reduction_cumprod	✓
135	reduction_cumsum_001	✓
136	reduction_cumsum_002	✓
137	reduction_cumsum	✓
138	reduction_dot	✓
139	reduction_fast_softmax_001	✓
140	reduction_fast_softmax_grad_001	✗
141	reduction_max_001	✓
142	reduction_max_002	✓
143	reduction_max_003	✓
144	reduction_max_large	✓
145	reduction_max_small	✓
146	reduction_max	✓
147	reduction_mean_001	✓
148	reduction_mean_002	✓
149	reduction_mean_003	✓
150	reduction_mean_004	✓
151	reduction_mean_005	✓
152	reduction_mean_006	✓
153	reduction_mean	✓
154	reduction_mean_vector	✓
155	reduction_min_001	✓
156	reduction_min_002	✓
157	reduction_min_003	✓
158	reduction_min	✓
159	reduction_prod_001	✗
160	reduction_prod_002	✓
161	reduction_prod	✓
162	reduction_softmax_001	✓
163	reduction_sum_001	✓
164	reduction_sum_002	✓
165	reduction_sum_005	✓
166	reduction_sum_006	✓
167	reduction_sum_dim1	✓
168	reduction_sum	✓
169	reduction_top2_gating_argmax	✗
170	sorting_isin	✗
171	sorting_sort	✗
172	sorting_topk_001	✗
173	sorting_topk_top_p_sampling_001	✗
174	sorting_topk	✗
175	tensor_manipulation_cat_001	✓
176	tensor_manipulation_cat_002	✓
177	tensor_manipulation_cat_dim	✗
178	tensor_manipulation_cat	✓
179	tensor_manipulation_flip	✓
180	tensor_manipulation_gather	✓
181	tensor_manipulation_hstack	✓
182	tensor_manipulation_index_select	✓
183	tensor_manipulation_pad_001	✗
184	tensor_manipulation_permute	✗
185	tensor_manipulation_reshape	✓
186	tensor_manipulation_scatter_add	✗
187	tensor_manipulation_scatter_multiply	✗
188	tensor_manipulation_scatter	✗
189	tensor_manipulation_split_001	✗
190	tensor_manipulation_split	✓
191	tensor_manipulation_squeeze	✓
192	tensor_manipulation_stack_001	✓
193	tensor_manipulation_stack_002	✓
194	tensor_manipulation_stack	✓
195	tensor_manipulation_transpose_001	✓
196	tensor_manipulation_transpose_2d	✓
197	tensor_manipulation_transpose	✓
198	tensor_manipulation_unpad_001	✓
199	tensor_manipulation_unsqueeze_001	✓
200	tensor_manipulation_unsqueeze	✓

说明

• ✓ 表示生成成功
• ✗ 表示生成失败
• 更新时间：2025年9月26日
• 测试方法：pass@4（4次尝试中至少1次成功）