<
   ir_version: 10,
   opset_import: ["" : 23],
   producer_name: "pytorch"
>
main_graph (uint8[batch,384,384,3] image) => (float[batch,1152] image_embedding) 
   <
      float[batch,729,1152] add_1007
      float[batch,729,1152] add_1068
      float[batch,729,1152] add_107
      float[batch,729,1152] add_1097
      float[batch,729,1152] add_1158
      float[batch,729,1152] add_1187
      float[batch,729,1152] add_1248
      float[batch,729,1152] add_1277
      float[batch,729,1152] add_13
      float[batch,729,1152] add_1338
      float[batch,729,1152] add_1367
      float[batch,729,1152] add_1428
      float[batch,729,1152] add_1457
      float[batch,729,1152] add_1518
      float[batch,729,1152] add_1547
      float[batch,729,1152] add_1608
      float[batch,729,1152] add_1637
      float[batch,729,1152] add_168
      float[batch,729,1152] add_1698
      float[batch,729,1152] add_1727
      float[batch,729,1152] add_1788
      float[batch,729,1152] add_1817
      float[batch,729,1152] add_1878
      float[batch,729,1152] add_1907
      float[batch,729,1152] add_1968
      float[batch,729,1152] add_197
      float[batch,729,1152] add_1997
      float[batch,729,1152] add_2058
      float[batch,729,1152] add_2087
      float[batch,729,1152] add_2148
      float[batch,729,1152] add_2177
      float[batch,729,1152] add_2238
      float[batch,729,1152] add_2267
      float[batch,729,1152] add_2328
      float[batch,729,1152] add_2357
      float[batch,729,1152] add_2418
      float[batch,729,1152] add_2447
      float[batch,1,1152] add_2545
      float[batch,729,1152] add_258
      float[batch,729,1152] add_287
      float[batch,729,1152] add_348
      float[batch,729,1152] add_377
      float[batch,729,1152] add_438
      float[batch,729,1152] add_467
      float[batch,729,1152] add_528
      float[batch,729,1152] add_557
      float[batch,729,1152] add_618
      float[batch,729,1152] add_647
      float[batch,729,1152] add_708
      float[batch,729,1152] add_737
      float[batch,729,1152] add_78
      float[batch,729,1152] add_798
      float[batch,729,1152] add_827
      float[batch,729,1152] add_888
      float[batch,729,1152] add_917
      float[batch,729,1152] add_978
      float[batch,1] clamp_min
      float[batch,1152,27,27] conv2d
      float[batch,729,4304] gelu
      float[batch,729,4304] gelu_1
      float[batch,729,4304] gelu_10
      float[batch,729,4304] gelu_11
      float[batch,729,4304] gelu_12
      float[batch,729,4304] gelu_13
      float[batch,729,4304] gelu_14
      float[batch,729,4304] gelu_15
      float[batch,729,4304] gelu_16
      float[batch,729,4304] gelu_17
      float[batch,729,4304] gelu_18
      float[batch,729,4304] gelu_19
      float[batch,729,4304] gelu_2
      float[batch,729,4304] gelu_20
      float[batch,729,4304] gelu_21
      float[batch,729,4304] gelu_22
      float[batch,729,4304] gelu_23
      float[batch,729,4304] gelu_24
      float[batch,729,4304] gelu_25
      float[batch,729,4304] gelu_26
      float[batch,1,4304] gelu_27
      float[batch,729,4304] gelu_3
      float[batch,729,4304] gelu_4
      float[batch,729,4304] gelu_5
      float[batch,729,4304] gelu_6
      float[batch,729,4304] gelu_7
      float[batch,729,4304] gelu_8
      float[batch,729,4304] gelu_9
      float[batch,3,384,384] image_chw
      float[batch] image_ez
      float[batch] image_ez_r
      float[batch,1,1,1] image_ez_s
      float[batch,384,384,3] image_f32
      float[batch,729,1152] layer_norm
      float[batch,729,1152] layer_norm_1
      float[batch,729,1152] layer_norm_10
      float[batch,729,1152] layer_norm_11
      float[batch,729,1152] layer_norm_12
      float[batch,729,1152] layer_norm_13
      float[batch,729,1152] layer_norm_14
      float[batch,729,1152] layer_norm_15
      float[batch,729,1152] layer_norm_16
      float[batch,729,1152] layer_norm_17
      float[batch,729,1152] layer_norm_18
      float[batch,729,1152] layer_norm_19
      float[batch,729,1152] layer_norm_2
      float[batch,729,1152] layer_norm_20
      float[batch,729,1152] layer_norm_21
      float[batch,729,1152] layer_norm_22
      float[batch,729,1152] layer_norm_23
      float[batch,729,1152] layer_norm_24
      float[batch,729,1152] layer_norm_25
      float[batch,729,1152] layer_norm_26
      float[batch,729,1152] layer_norm_27
      float[batch,729,1152] layer_norm_28
      float[batch,729,1152] layer_norm_29
      float[batch,729,1152] layer_norm_3
      float[batch,729,1152] layer_norm_30
      float[batch,729,1152] layer_norm_31
      float[batch,729,1152] layer_norm_32
      float[batch,729,1152] layer_norm_33
      float[batch,729,1152] layer_norm_34
      float[batch,729,1152] layer_norm_35
      float[batch,729,1152] layer_norm_36
      float[batch,729,1152] layer_norm_37
      float[batch,729,1152] layer_norm_38
      float[batch,729,1152] layer_norm_39
      float[batch,729,1152] layer_norm_4
      float[batch,729,1152] layer_norm_40
      float[batch,729,1152] layer_norm_41
      float[batch,729,1152] layer_norm_42
      float[batch,729,1152] layer_norm_43
      float[batch,729,1152] layer_norm_44
      float[batch,729,1152] layer_norm_45
      float[batch,729,1152] layer_norm_46
      float[batch,729,1152] layer_norm_47
      float[batch,729,1152] layer_norm_48
      float[batch,729,1152] layer_norm_49
      float[batch,729,1152] layer_norm_5
      float[batch,729,1152] layer_norm_50
      float[batch,729,1152] layer_norm_51
      float[batch,729,1152] layer_norm_52
      float[batch,729,1152] layer_norm_53
      float[batch,729,1152] layer_norm_54
      float[batch,1,1152] layer_norm_55
      float[batch,729,1152] layer_norm_6
      float[batch,729,1152] layer_norm_7
      float[batch,729,1152] layer_norm_8
      float[batch,729,1152] layer_norm_9
      float[batch,1] linalg_vector_norm
      float[batch,729,3456] linear
      float[batch,729,1152] linear_1
      float[batch,729,4304] linear_10
      float[batch,729,3456] linear_100
      float[batch,729,1152] linear_101
      float[batch,729,4304] linear_102
      float[batch,729,1152] linear_103
      float[batch,729,3456] linear_104
      float[batch,729,1152] linear_105
      float[batch,729,4304] linear_106
      float[batch,729,1152] linear_107
      float[batch,729,2304] linear_109
      float[batch,729,1152] linear_11
      float[batch,1,1152] linear_110
      float[batch,1,4304] linear_111
      float[batch,1,1152] linear_112
      float[batch,729,3456] linear_12
      float[batch,729,1152] linear_13
      float[batch,729,4304] linear_14
      float[batch,729,1152] linear_15
      float[batch,729,3456] linear_16
      float[batch,729,1152] linear_17
      float[batch,729,4304] linear_18
      float[batch,729,1152] linear_19
      float[batch,729,4304] linear_2
      float[batch,729,3456] linear_20
      float[batch,729,1152] linear_21
      float[batch,729,4304] linear_22
      float[batch,729,1152] linear_23
      float[batch,729,3456] linear_24
      float[batch,729,1152] linear_25
      float[batch,729,4304] linear_26
      float[batch,729,1152] linear_27
      float[batch,729,3456] linear_28
      float[batch,729,1152] linear_29
      float[batch,729,1152] linear_3
      float[batch,729,4304] linear_30
      float[batch,729,1152] linear_31
      float[batch,729,3456] linear_32
      float[batch,729,1152] linear_33
      float[batch,729,4304] linear_34
      float[batch,729,1152] linear_35
      float[batch,729,3456] linear_36
      float[batch,729,1152] linear_37
      float[batch,729,4304] linear_38
      float[batch,729,1152] linear_39
      float[batch,729,3456] linear_4
      float[batch,729,3456] linear_40
      float[batch,729,1152] linear_41
      float[batch,729,4304] linear_42
      float[batch,729,1152] linear_43
      float[batch,729,3456] linear_44
      float[batch,729,1152] linear_45
      float[batch,729,4304] linear_46
      float[batch,729,1152] linear_47
      float[batch,729,3456] linear_48
      float[batch,729,1152] linear_49
      float[batch,729,1152] linear_5
      float[batch,729,4304] linear_50
      float[batch,729,1152] linear_51
      float[batch,729,3456] linear_52
      float[batch,729,1152] linear_53
      float[batch,729,4304] linear_54
      float[batch,729,1152] linear_55
      float[batch,729,3456] linear_56
      float[batch,729,1152] linear_57
      float[batch,729,4304] linear_58
      float[batch,729,1152] linear_59
      float[batch,729,4304] linear_6
      float[batch,729,3456] linear_60
      float[batch,729,1152] linear_61
      float[batch,729,4304] linear_62
      float[batch,729,1152] linear_63
      float[batch,729,3456] linear_64
      float[batch,729,1152] linear_65
      float[batch,729,4304] linear_66
      float[batch,729,1152] linear_67
      float[batch,729,3456] linear_68
      float[batch,729,1152] linear_69
      float[batch,729,1152] linear_7
      float[batch,729,4304] linear_70
      float[batch,729,1152] linear_71
      float[batch,729,3456] linear_72
      float[batch,729,1152] linear_73
      float[batch,729,4304] linear_74
      float[batch,729,1152] linear_75
      float[batch,729,3456] linear_76
      float[batch,729,1152] linear_77
      float[batch,729,4304] linear_78
      float[batch,729,1152] linear_79
      float[batch,729,3456] linear_8
      float[batch,729,3456] linear_80
      float[batch,729,1152] linear_81
      float[batch,729,4304] linear_82
      float[batch,729,1152] linear_83
      float[batch,729,3456] linear_84
      float[batch,729,1152] linear_85
      float[batch,729,4304] linear_86
      float[batch,729,1152] linear_87
      float[batch,729,3456] linear_88
      float[batch,729,1152] linear_89
      float[batch,729,1152] linear_9
      float[batch,729,4304] linear_90
      float[batch,729,1152] linear_91
      float[batch,729,3456] linear_92
      float[batch,729,1152] linear_93
      float[batch,729,4304] linear_94
      float[batch,729,1152] linear_95
      float[batch,729,3456] linear_96
      float[batch,729,1152] linear_97
      float[batch,729,4304] linear_98
      float[batch,729,1152] linear_99
      float[batch,729,1152] node_scaled_dot_product_attention_10_k
      float[batch,729,1152] node_scaled_dot_product_attention_10_q
      float[batch,729,1152] node_scaled_dot_product_attention_10_v
      float[batch,729,1152] node_scaled_dot_product_attention_11_k
      float[batch,729,1152] node_scaled_dot_product_attention_11_q
      float[batch,729,1152] node_scaled_dot_product_attention_11_v
      float[batch,729,1152] node_scaled_dot_product_attention_12_k
      float[batch,729,1152] node_scaled_dot_product_attention_12_q
      float[batch,729,1152] node_scaled_dot_product_attention_12_v
      float[batch,729,1152] node_scaled_dot_product_attention_13_k
      float[batch,729,1152] node_scaled_dot_product_attention_13_q
      float[batch,729,1152] node_scaled_dot_product_attention_13_v
      float[batch,729,1152] node_scaled_dot_product_attention_14_k
      float[batch,729,1152] node_scaled_dot_product_attention_14_q
      float[batch,729,1152] node_scaled_dot_product_attention_14_v
      float[batch,729,1152] node_scaled_dot_product_attention_15_k
      float[batch,729,1152] node_scaled_dot_product_attention_15_q
      float[batch,729,1152] node_scaled_dot_product_attention_15_v
      float[batch,729,1152] node_scaled_dot_product_attention_16_k
      float[batch,729,1152] node_scaled_dot_product_attention_16_q
      float[batch,729,1152] node_scaled_dot_product_attention_16_v
      float[batch,729,1152] node_scaled_dot_product_attention_17_k
      float[batch,729,1152] node_scaled_dot_product_attention_17_q
      float[batch,729,1152] node_scaled_dot_product_attention_17_v
      float[batch,729,1152] node_scaled_dot_product_attention_18_k
      float[batch,729,1152] node_scaled_dot_product_attention_18_q
      float[batch,729,1152] node_scaled_dot_product_attention_18_v
      float[batch,729,1152] node_scaled_dot_product_attention_19_k
      float[batch,729,1152] node_scaled_dot_product_attention_19_q
      float[batch,729,1152] node_scaled_dot_product_attention_19_v
      float[batch,729,1152] node_scaled_dot_product_attention_1_k
      float[batch,729,1152] node_scaled_dot_product_attention_1_q
      float[batch,729,1152] node_scaled_dot_product_attention_1_v
      float[batch,729,1152] node_scaled_dot_product_attention_20_k
      float[batch,729,1152] node_scaled_dot_product_attention_20_q
      float[batch,729,1152] node_scaled_dot_product_attention_20_v
      float[batch,729,1152] node_scaled_dot_product_attention_21_k
      float[batch,729,1152] node_scaled_dot_product_attention_21_q
      float[batch,729,1152] node_scaled_dot_product_attention_21_v
      float[batch,729,1152] node_scaled_dot_product_attention_22_k
      float[batch,729,1152] node_scaled_dot_product_attention_22_q
      float[batch,729,1152] node_scaled_dot_product_attention_22_v
      float[batch,729,1152] node_scaled_dot_product_attention_23_k
      float[batch,729,1152] node_scaled_dot_product_attention_23_q
      float[batch,729,1152] node_scaled_dot_product_attention_23_v
      float[batch,729,1152] node_scaled_dot_product_attention_24_k
      float[batch,729,1152] node_scaled_dot_product_attention_24_q
      float[batch,729,1152] node_scaled_dot_product_attention_24_v
      float[batch,729,1152] node_scaled_dot_product_attention_25_k
      float[batch,729,1152] node_scaled_dot_product_attention_25_q
      float[batch,729,1152] node_scaled_dot_product_attention_25_v
      float[batch,729,1152] node_scaled_dot_product_attention_26_k
      float[batch,729,1152] node_scaled_dot_product_attention_26_q
      float[batch,729,1152] node_scaled_dot_product_attention_26_v
      float[batch,729,1152] node_scaled_dot_product_attention_27_k
      float[batch,1,1152] node_scaled_dot_product_attention_27_q
      float[batch,1,1] node_scaled_dot_product_attention_27_q_col_out
      float[batch,729,1152] node_scaled_dot_product_attention_27_v
      float[batch,729,1152] node_scaled_dot_product_attention_2_k
      float[batch,729,1152] node_scaled_dot_product_attention_2_q
      float[batch,729,1152] node_scaled_dot_product_attention_2_v
      float[batch,729,1152] node_scaled_dot_product_attention_3_k
      float[batch,729,1152] node_scaled_dot_product_attention_3_q
      float[batch,729,1152] node_scaled_dot_product_attention_3_v
      float[batch,729,1152] node_scaled_dot_product_attention_4_k
      float[batch,729,1152] node_scaled_dot_product_attention_4_q
      float[batch,729,1152] node_scaled_dot_product_attention_4_v
      float[batch,729,1152] node_scaled_dot_product_attention_5_k
      float[batch,729,1152] node_scaled_dot_product_attention_5_q
      float[batch,729,1152] node_scaled_dot_product_attention_5_v
      float[batch,729,1152] node_scaled_dot_product_attention_6_k
      float[batch,729,1152] node_scaled_dot_product_attention_6_q
      float[batch,729,1152] node_scaled_dot_product_attention_6_v
      float[batch,729,1152] node_scaled_dot_product_attention_7_k
      float[batch,729,1152] node_scaled_dot_product_attention_7_q
      float[batch,729,1152] node_scaled_dot_product_attention_7_v
      float[batch,729,1152] node_scaled_dot_product_attention_8_k
      float[batch,729,1152] node_scaled_dot_product_attention_8_q
      float[batch,729,1152] node_scaled_dot_product_attention_8_v
      float[batch,729,1152] node_scaled_dot_product_attention_9_k
      float[batch,729,1152] node_scaled_dot_product_attention_9_q
      float[batch,729,1152] node_scaled_dot_product_attention_9_v
      float[batch,729,1152] node_scaled_dot_product_attention_k
      float[batch,729,1152] node_scaled_dot_product_attention_q
      float[batch,729,1152] node_scaled_dot_product_attention_v
      float[batch,729,1152] scaled_dot_product_attention
      float[batch,729,1152] scaled_dot_product_attention_1
      float[batch,729,1152] scaled_dot_product_attention_10
      float[batch,729,1152] scaled_dot_product_attention_11
      float[batch,729,1152] scaled_dot_product_attention_12
      float[batch,729,1152] scaled_dot_product_attention_13
      float[batch,729,1152] scaled_dot_product_attention_14
      float[batch,729,1152] scaled_dot_product_attention_15
      float[batch,729,1152] scaled_dot_product_attention_16
      float[batch,729,1152] scaled_dot_product_attention_17
      float[batch,729,1152] scaled_dot_product_attention_18
      float[batch,729,1152] scaled_dot_product_attention_19
      float[batch,729,1152] scaled_dot_product_attention_2
      float[batch,729,1152] scaled_dot_product_attention_20
      float[batch,729,1152] scaled_dot_product_attention_21
      float[batch,729,1152] scaled_dot_product_attention_22
      float[batch,729,1152] scaled_dot_product_attention_23
      float[batch,729,1152] scaled_dot_product_attention_24
      float[batch,729,1152] scaled_dot_product_attention_25
      float[batch,729,1152] scaled_dot_product_attention_26
      float[batch,1,1152] scaled_dot_product_attention_27
      float[batch,729,1152] scaled_dot_product_attention_3
      float[batch,729,1152] scaled_dot_product_attention_4
      float[batch,729,1152] scaled_dot_product_attention_5
      float[batch,729,1152] scaled_dot_product_attention_6
      float[batch,729,1152] scaled_dot_product_attention_7
      float[batch,729,1152] scaled_dot_product_attention_8
      float[batch,729,1152] scaled_dot_product_attention_9
      float[batch,1152] select
      float[batch,729,1152] transpose
      float[batch,729,3456] val_115
      float[batch,729,1152] val_116
      float[batch,729,4304] val_117
      float[batch,729,1152] val_118
      float[batch,729,3456] val_119
      float[batch,729,1152] val_120
      float[batch,729,4304] val_121
      float[batch,729,1152] val_122
      float[batch,729,3456] val_123
      float[batch,729,1152] val_124
      float[batch,729,4304] val_125
      float[batch,729,1152] val_126
      float[batch,729,3456] val_127
      float[batch,729,1152] val_128
      float[batch,729,4304] val_129
      float[batch,729,1152] val_130
      float[batch,729,3456] val_131
      float[batch,729,1152] val_132
      float[batch,729,4304] val_133
      float[batch,729,1152] val_134
      float[batch,729,3456] val_135
      float[batch,729,1152] val_136
      float[batch,729,4304] val_137
      float[batch,729,1152] val_138
      float[batch,729,3456] val_139
      float[batch,729,1152] val_140
      float[batch,729,4304] val_141
      float[batch,729,1152] val_142
      float[batch,729,3456] val_143
      float[batch,729,1152] val_144
      float[batch,729,4304] val_145
      float[batch,729,1152] val_146
      float[batch,729,3456] val_147
      float[batch,729,1152] val_148
      float[batch,729,4304] val_149
      float[batch,729,1152] val_150
      float[batch,729,3456] val_151
      float[batch,729,1152] val_152
      float[batch,729,4304] val_153
      float[batch,729,1152] val_154
      float[batch,729,3456] val_155
      float[batch,729,1152] val_156
      float[batch,729,4304] val_157
      float[batch,729,1152] val_158
      float[batch,729,3456] val_159
      float[batch,729,1152] val_160
      float[batch,729,4304] val_161
      float[batch,729,1152] val_162
      float[batch,729,3456] val_163
      float[batch,729,1152] val_164
      float[batch,729,4304] val_165
      float[batch,729,1152] val_166
      float[batch,729,3456] val_167
      float[batch,729,1152] val_168
      float[batch,729,4304] val_169
      float[batch,729,1152] val_170
      float[batch,729,3456] val_171
      float[batch,729,1152] val_172
      float[batch,729,4304] val_173
      float[batch,729,1152] val_174
      float[batch,729,3456] val_175
      float[batch,729,1152] val_176
      float[batch,729,4304] val_177
      float[batch,729,1152] val_178
      float[batch,729,3456] val_179
      float[batch,729,1152] val_180
      float[batch,729,4304] val_181
      float[batch,729,1152] val_182
      float[batch,729,3456] val_183
      float[batch,729,1152] val_184
      float[batch,729,4304] val_185
      float[batch,729,1152] val_186
      float[batch,729,3456] val_187
      float[batch,729,1152] val_188
      float[batch,729,4304] val_189
      float[batch,729,1152] val_190
      float[batch,729,3456] val_191
      float[batch,729,1152] val_192
      float[batch,729,4304] val_193
      float[batch,729,1152] val_194
      float[batch,729,3456] val_195
      float[batch,729,1152] val_196
      float[batch,729,4304] val_197
      float[batch,729,1152] val_198
      float[batch,729,3456] val_199
      float[batch,729,1152] val_200
      float[batch,729,4304] val_201
      float[batch,729,1152] val_202
      float[batch,729,3456] val_203
      float[batch,729,1152] val_204
      float[batch,729,4304] val_205
      float[batch,729,1152] val_206
      float[batch,729,3456] val_207
      float[batch,729,1152] val_208
      float[batch,729,4304] val_209
      float[batch,729,1152] val_210
      float[batch,729,3456] val_211
      float[batch,729,1152] val_212
      float[batch,729,4304] val_213
      float[batch,729,1152] val_214
      float[batch,729,3456] val_215
      float[batch,729,1152] val_216
      float[batch,729,4304] val_217
      float[batch,729,1152] val_218
      float[batch,729,3456] val_219
      float[batch,729,1152] val_220
      float[batch,729,4304] val_221
      float[batch,729,1152] val_222
      float[batch,729,2304] val_223
      float[batch,1,1152] val_224
      float[batch,1,4304] val_225
      float[batch,1,1152] val_226
      float[batch,1152,729] view
   >
{
   [pre_cast] image_f32 = Cast <to: int = 1> (image)
   [pre_nhwc_to_nchw] image_chw = Transpose <perm: ints = [0, 3, 1, 2]> (image_f32)
   [node_conv2d] conv2d = Conv <auto_pad: string = "NOTSET", dilations: ints = [1, 1], group: int = 1, pads: ints = [0, 0, 0, 0], strides: ints = [14, 14]> (image_chw, "visual.trunk.patch_embed.proj.weight", "visual.trunk.patch_embed.proj.bias")
   view = Reshape <allowzero: int = 1> (conv2d, view_target)
   [node_transpose] transpose = Transpose <perm: ints = [0, 2, 1]> (view)
   add_13 = Add (transpose, "visual.trunk.pos_embed")
   [node_layer_norm] layer_norm = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_13, "visual.trunk.blocks.0.norm1.weight", "visual.trunk.blocks.0.norm1.bias")
   val_115 = MatMul (layer_norm, val_3)
   [node_linear] linear = Add (val_115, "visual.trunk.blocks.0.attn.qkv.bias")
   [node_scaled_dot_product_attention_qkv_split] node_scaled_dot_product_attention_q, node_scaled_dot_product_attention_k, node_scaled_dot_product_attention_v = Split <axis: int = -1> (linear, attn3d_split_3x1152)
   [node_scaled_dot_product_attention] scaled_dot_product_attention = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_q, node_scaled_dot_product_attention_k, node_scaled_dot_product_attention_v)
   val_116 = MatMul (scaled_dot_product_attention, val_4)
   linear_1 = Add (val_116, "visual.trunk.blocks.0.attn.proj.bias")
   add_78 = Add (add_13, linear_1)
   layer_norm_1 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_78, "visual.trunk.blocks.0.norm2.weight", "visual.trunk.blocks.0.norm2.bias")
   val_117 = MatMul (layer_norm_1, val_5)
   linear_2 = Add (val_117, "visual.trunk.blocks.0.mlp.fc1.bias")
   [node_gelu] gelu = Gelu <approximate: string = "none"> (linear_2)
   val_118 = MatMul (gelu, val_6)
   linear_3 = Add (val_118, "visual.trunk.blocks.0.mlp.fc2.bias")
   add_107 = Add (add_78, linear_3)
   layer_norm_2 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_107, "visual.trunk.blocks.1.norm1.weight", "visual.trunk.blocks.1.norm1.bias")
   val_119 = MatMul (layer_norm_2, val_7)
   linear_4 = Add (val_119, "visual.trunk.blocks.1.attn.qkv.bias")
   [node_scaled_dot_product_attention_1_qkv_split] node_scaled_dot_product_attention_1_q, node_scaled_dot_product_attention_1_k, node_scaled_dot_product_attention_1_v = Split <axis: int = -1> (linear_4, attn3d_split_3x1152)
   scaled_dot_product_attention_1 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_1_q, node_scaled_dot_product_attention_1_k, node_scaled_dot_product_attention_1_v)
   val_120 = MatMul (scaled_dot_product_attention_1, val_8)
   linear_5 = Add (val_120, "visual.trunk.blocks.1.attn.proj.bias")
   add_168 = Add (add_107, linear_5)
   layer_norm_3 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_168, "visual.trunk.blocks.1.norm2.weight", "visual.trunk.blocks.1.norm2.bias")
   val_121 = MatMul (layer_norm_3, val_9)
   linear_6 = Add (val_121, "visual.trunk.blocks.1.mlp.fc1.bias")
   gelu_1 = Gelu <approximate: string = "none"> (linear_6)
   val_122 = MatMul (gelu_1, val_10)
   linear_7 = Add (val_122, "visual.trunk.blocks.1.mlp.fc2.bias")
   add_197 = Add (add_168, linear_7)
   layer_norm_4 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_197, "visual.trunk.blocks.2.norm1.weight", "visual.trunk.blocks.2.norm1.bias")
   val_123 = MatMul (layer_norm_4, val_11)
   linear_8 = Add (val_123, "visual.trunk.blocks.2.attn.qkv.bias")
   [node_scaled_dot_product_attention_2_qkv_split] node_scaled_dot_product_attention_2_q, node_scaled_dot_product_attention_2_k, node_scaled_dot_product_attention_2_v = Split <axis: int = -1> (linear_8, attn3d_split_3x1152)
   scaled_dot_product_attention_2 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_2_q, node_scaled_dot_product_attention_2_k, node_scaled_dot_product_attention_2_v)
   val_124 = MatMul (scaled_dot_product_attention_2, val_12)
   linear_9 = Add (val_124, "visual.trunk.blocks.2.attn.proj.bias")
   add_258 = Add (add_197, linear_9)
   layer_norm_5 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_258, "visual.trunk.blocks.2.norm2.weight", "visual.trunk.blocks.2.norm2.bias")
   val_125 = MatMul (layer_norm_5, val_13)
   linear_10 = Add (val_125, "visual.trunk.blocks.2.mlp.fc1.bias")
   gelu_2 = Gelu <approximate: string = "none"> (linear_10)
   val_126 = MatMul (gelu_2, val_14)
   linear_11 = Add (val_126, "visual.trunk.blocks.2.mlp.fc2.bias")
   add_287 = Add (add_258, linear_11)
   layer_norm_6 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_287, "visual.trunk.blocks.3.norm1.weight", "visual.trunk.blocks.3.norm1.bias")
   val_127 = MatMul (layer_norm_6, val_15)
   linear_12 = Add (val_127, "visual.trunk.blocks.3.attn.qkv.bias")
   [node_scaled_dot_product_attention_3_qkv_split] node_scaled_dot_product_attention_3_q, node_scaled_dot_product_attention_3_k, node_scaled_dot_product_attention_3_v = Split <axis: int = -1> (linear_12, attn3d_split_3x1152)
   scaled_dot_product_attention_3 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_3_q, node_scaled_dot_product_attention_3_k, node_scaled_dot_product_attention_3_v)
   val_128 = MatMul (scaled_dot_product_attention_3, val_16)
   linear_13 = Add (val_128, "visual.trunk.blocks.3.attn.proj.bias")
   add_348 = Add (add_287, linear_13)
   layer_norm_7 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_348, "visual.trunk.blocks.3.norm2.weight", "visual.trunk.blocks.3.norm2.bias")
   val_129 = MatMul (layer_norm_7, val_17)
   linear_14 = Add (val_129, "visual.trunk.blocks.3.mlp.fc1.bias")
   gelu_3 = Gelu <approximate: string = "none"> (linear_14)
   val_130 = MatMul (gelu_3, val_18)
   linear_15 = Add (val_130, "visual.trunk.blocks.3.mlp.fc2.bias")
   add_377 = Add (add_348, linear_15)
   layer_norm_8 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_377, "visual.trunk.blocks.4.norm1.weight", "visual.trunk.blocks.4.norm1.bias")
   val_131 = MatMul (layer_norm_8, val_19)
   linear_16 = Add (val_131, "visual.trunk.blocks.4.attn.qkv.bias")
   [node_scaled_dot_product_attention_4_qkv_split] node_scaled_dot_product_attention_4_q, node_scaled_dot_product_attention_4_k, node_scaled_dot_product_attention_4_v = Split <axis: int = -1> (linear_16, attn3d_split_3x1152)
   scaled_dot_product_attention_4 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_4_q, node_scaled_dot_product_attention_4_k, node_scaled_dot_product_attention_4_v)
   val_132 = MatMul (scaled_dot_product_attention_4, val_20)
   linear_17 = Add (val_132, "visual.trunk.blocks.4.attn.proj.bias")
   add_438 = Add (add_377, linear_17)
   layer_norm_9 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_438, "visual.trunk.blocks.4.norm2.weight", "visual.trunk.blocks.4.norm2.bias")
   val_133 = MatMul (layer_norm_9, val_21)
   linear_18 = Add (val_133, "visual.trunk.blocks.4.mlp.fc1.bias")
   gelu_4 = Gelu <approximate: string = "none"> (linear_18)
   val_134 = MatMul (gelu_4, val_22)
   linear_19 = Add (val_134, "visual.trunk.blocks.4.mlp.fc2.bias")
   add_467 = Add (add_438, linear_19)
   layer_norm_10 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_467, "visual.trunk.blocks.5.norm1.weight", "visual.trunk.blocks.5.norm1.bias")
   val_135 = MatMul (layer_norm_10, val_23)
   linear_20 = Add (val_135, "visual.trunk.blocks.5.attn.qkv.bias")
   [node_scaled_dot_product_attention_5_qkv_split] node_scaled_dot_product_attention_5_q, node_scaled_dot_product_attention_5_k, node_scaled_dot_product_attention_5_v = Split <axis: int = -1> (linear_20, attn3d_split_3x1152)
   scaled_dot_product_attention_5 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_5_q, node_scaled_dot_product_attention_5_k, node_scaled_dot_product_attention_5_v)
   val_136 = MatMul (scaled_dot_product_attention_5, val_24)
   linear_21 = Add (val_136, "visual.trunk.blocks.5.attn.proj.bias")
   add_528 = Add (add_467, linear_21)
   layer_norm_11 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_528, "visual.trunk.blocks.5.norm2.weight", "visual.trunk.blocks.5.norm2.bias")
   val_137 = MatMul (layer_norm_11, val_25)
   linear_22 = Add (val_137, "visual.trunk.blocks.5.mlp.fc1.bias")
   gelu_5 = Gelu <approximate: string = "none"> (linear_22)
   val_138 = MatMul (gelu_5, val_26)
   linear_23 = Add (val_138, "visual.trunk.blocks.5.mlp.fc2.bias")
   add_557 = Add (add_528, linear_23)
   layer_norm_12 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_557, "visual.trunk.blocks.6.norm1.weight", "visual.trunk.blocks.6.norm1.bias")
   val_139 = MatMul (layer_norm_12, val_27)
   linear_24 = Add (val_139, "visual.trunk.blocks.6.attn.qkv.bias")
   [node_scaled_dot_product_attention_6_qkv_split] node_scaled_dot_product_attention_6_q, node_scaled_dot_product_attention_6_k, node_scaled_dot_product_attention_6_v = Split <axis: int = -1> (linear_24, attn3d_split_3x1152)
   scaled_dot_product_attention_6 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_6_q, node_scaled_dot_product_attention_6_k, node_scaled_dot_product_attention_6_v)
   val_140 = MatMul (scaled_dot_product_attention_6, val_28)
   linear_25 = Add (val_140, "visual.trunk.blocks.6.attn.proj.bias")
   add_618 = Add (add_557, linear_25)
   layer_norm_13 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_618, "visual.trunk.blocks.6.norm2.weight", "visual.trunk.blocks.6.norm2.bias")
   val_141 = MatMul (layer_norm_13, val_29)
   linear_26 = Add (val_141, "visual.trunk.blocks.6.mlp.fc1.bias")
   gelu_6 = Gelu <approximate: string = "none"> (linear_26)
   val_142 = MatMul (gelu_6, val_30)
   linear_27 = Add (val_142, "visual.trunk.blocks.6.mlp.fc2.bias")
   add_647 = Add (add_618, linear_27)
   layer_norm_14 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_647, "visual.trunk.blocks.7.norm1.weight", "visual.trunk.blocks.7.norm1.bias")
   val_143 = MatMul (layer_norm_14, val_31)
   linear_28 = Add (val_143, "visual.trunk.blocks.7.attn.qkv.bias")
   [node_scaled_dot_product_attention_7_qkv_split] node_scaled_dot_product_attention_7_q, node_scaled_dot_product_attention_7_k, node_scaled_dot_product_attention_7_v = Split <axis: int = -1> (linear_28, attn3d_split_3x1152)
   scaled_dot_product_attention_7 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_7_q, node_scaled_dot_product_attention_7_k, node_scaled_dot_product_attention_7_v)
   val_144 = MatMul (scaled_dot_product_attention_7, val_32)
   linear_29 = Add (val_144, "visual.trunk.blocks.7.attn.proj.bias")
   add_708 = Add (add_647, linear_29)
   layer_norm_15 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_708, "visual.trunk.blocks.7.norm2.weight", "visual.trunk.blocks.7.norm2.bias")
   val_145 = MatMul (layer_norm_15, val_33)
   linear_30 = Add (val_145, "visual.trunk.blocks.7.mlp.fc1.bias")
   gelu_7 = Gelu <approximate: string = "none"> (linear_30)
   val_146 = MatMul (gelu_7, val_34)
   linear_31 = Add (val_146, "visual.trunk.blocks.7.mlp.fc2.bias")
   add_737 = Add (add_708, linear_31)
   layer_norm_16 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_737, "visual.trunk.blocks.8.norm1.weight", "visual.trunk.blocks.8.norm1.bias")
   val_147 = MatMul (layer_norm_16, val_35)
   linear_32 = Add (val_147, "visual.trunk.blocks.8.attn.qkv.bias")
   [node_scaled_dot_product_attention_8_qkv_split] node_scaled_dot_product_attention_8_q, node_scaled_dot_product_attention_8_k, node_scaled_dot_product_attention_8_v = Split <axis: int = -1> (linear_32, attn3d_split_3x1152)
   scaled_dot_product_attention_8 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_8_q, node_scaled_dot_product_attention_8_k, node_scaled_dot_product_attention_8_v)
   val_148 = MatMul (scaled_dot_product_attention_8, val_36)
   linear_33 = Add (val_148, "visual.trunk.blocks.8.attn.proj.bias")
   add_798 = Add (add_737, linear_33)
   layer_norm_17 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_798, "visual.trunk.blocks.8.norm2.weight", "visual.trunk.blocks.8.norm2.bias")
   val_149 = MatMul (layer_norm_17, val_37)
   linear_34 = Add (val_149, "visual.trunk.blocks.8.mlp.fc1.bias")
   gelu_8 = Gelu <approximate: string = "none"> (linear_34)
   val_150 = MatMul (gelu_8, val_38)
   linear_35 = Add (val_150, "visual.trunk.blocks.8.mlp.fc2.bias")
   add_827 = Add (add_798, linear_35)
   layer_norm_18 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_827, "visual.trunk.blocks.9.norm1.weight", "visual.trunk.blocks.9.norm1.bias")
   val_151 = MatMul (layer_norm_18, val_39)
   linear_36 = Add (val_151, "visual.trunk.blocks.9.attn.qkv.bias")
   [node_scaled_dot_product_attention_9_qkv_split] node_scaled_dot_product_attention_9_q, node_scaled_dot_product_attention_9_k, node_scaled_dot_product_attention_9_v = Split <axis: int = -1> (linear_36, attn3d_split_3x1152)
   scaled_dot_product_attention_9 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_9_q, node_scaled_dot_product_attention_9_k, node_scaled_dot_product_attention_9_v)
   val_152 = MatMul (scaled_dot_product_attention_9, val_40)
   linear_37 = Add (val_152, "visual.trunk.blocks.9.attn.proj.bias")
   add_888 = Add (add_827, linear_37)
   layer_norm_19 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_888, "visual.trunk.blocks.9.norm2.weight", "visual.trunk.blocks.9.norm2.bias")
   val_153 = MatMul (layer_norm_19, val_41)
   linear_38 = Add (val_153, "visual.trunk.blocks.9.mlp.fc1.bias")
   gelu_9 = Gelu <approximate: string = "none"> (linear_38)
   val_154 = MatMul (gelu_9, val_42)
   linear_39 = Add (val_154, "visual.trunk.blocks.9.mlp.fc2.bias")
   add_917 = Add (add_888, linear_39)
   layer_norm_20 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_917, "visual.trunk.blocks.10.norm1.weight", "visual.trunk.blocks.10.norm1.bias")
   val_155 = MatMul (layer_norm_20, val_43)
   linear_40 = Add (val_155, "visual.trunk.blocks.10.attn.qkv.bias")
   [node_scaled_dot_product_attention_10_qkv_split] node_scaled_dot_product_attention_10_q, node_scaled_dot_product_attention_10_k, node_scaled_dot_product_attention_10_v = Split <axis: int = -1> (linear_40, attn3d_split_3x1152)
   scaled_dot_product_attention_10 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_10_q, node_scaled_dot_product_attention_10_k, node_scaled_dot_product_attention_10_v)
   val_156 = MatMul (scaled_dot_product_attention_10, val_44)
   linear_41 = Add (val_156, "visual.trunk.blocks.10.attn.proj.bias")
   add_978 = Add (add_917, linear_41)
   layer_norm_21 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_978, "visual.trunk.blocks.10.norm2.weight", "visual.trunk.blocks.10.norm2.bias")
   val_157 = MatMul (layer_norm_21, val_45)
   linear_42 = Add (val_157, "visual.trunk.blocks.10.mlp.fc1.bias")
   gelu_10 = Gelu <approximate: string = "none"> (linear_42)
   val_158 = MatMul (gelu_10, val_46)
   linear_43 = Add (val_158, "visual.trunk.blocks.10.mlp.fc2.bias")
   add_1007 = Add (add_978, linear_43)
   layer_norm_22 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1007, "visual.trunk.blocks.11.norm1.weight", "visual.trunk.blocks.11.norm1.bias")
   val_159 = MatMul (layer_norm_22, val_47)
   linear_44 = Add (val_159, "visual.trunk.blocks.11.attn.qkv.bias")
   [node_scaled_dot_product_attention_11_qkv_split] node_scaled_dot_product_attention_11_q, node_scaled_dot_product_attention_11_k, node_scaled_dot_product_attention_11_v = Split <axis: int = -1> (linear_44, attn3d_split_3x1152)
   scaled_dot_product_attention_11 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_11_q, node_scaled_dot_product_attention_11_k, node_scaled_dot_product_attention_11_v)
   val_160 = MatMul (scaled_dot_product_attention_11, val_48)
   linear_45 = Add (val_160, "visual.trunk.blocks.11.attn.proj.bias")
   add_1068 = Add (add_1007, linear_45)
   layer_norm_23 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1068, "visual.trunk.blocks.11.norm2.weight", "visual.trunk.blocks.11.norm2.bias")
   val_161 = MatMul (layer_norm_23, val_49)
   linear_46 = Add (val_161, "visual.trunk.blocks.11.mlp.fc1.bias")
   gelu_11 = Gelu <approximate: string = "none"> (linear_46)
   val_162 = MatMul (gelu_11, val_50)
   linear_47 = Add (val_162, "visual.trunk.blocks.11.mlp.fc2.bias")
   add_1097 = Add (add_1068, linear_47)
   layer_norm_24 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1097, "visual.trunk.blocks.12.norm1.weight", "visual.trunk.blocks.12.norm1.bias")
   val_163 = MatMul (layer_norm_24, val_51)
   linear_48 = Add (val_163, "visual.trunk.blocks.12.attn.qkv.bias")
   [node_scaled_dot_product_attention_12_qkv_split] node_scaled_dot_product_attention_12_q, node_scaled_dot_product_attention_12_k, node_scaled_dot_product_attention_12_v = Split <axis: int = -1> (linear_48, attn3d_split_3x1152)
   scaled_dot_product_attention_12 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_12_q, node_scaled_dot_product_attention_12_k, node_scaled_dot_product_attention_12_v)
   val_164 = MatMul (scaled_dot_product_attention_12, val_52)
   linear_49 = Add (val_164, "visual.trunk.blocks.12.attn.proj.bias")
   add_1158 = Add (add_1097, linear_49)
   layer_norm_25 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1158, "visual.trunk.blocks.12.norm2.weight", "visual.trunk.blocks.12.norm2.bias")
   val_165 = MatMul (layer_norm_25, val_53)
   linear_50 = Add (val_165, "visual.trunk.blocks.12.mlp.fc1.bias")
   gelu_12 = Gelu <approximate: string = "none"> (linear_50)
   val_166 = MatMul (gelu_12, val_54)
   linear_51 = Add (val_166, "visual.trunk.blocks.12.mlp.fc2.bias")
   add_1187 = Add (add_1158, linear_51)
   layer_norm_26 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1187, "visual.trunk.blocks.13.norm1.weight", "visual.trunk.blocks.13.norm1.bias")
   val_167 = MatMul (layer_norm_26, val_55)
   linear_52 = Add (val_167, "visual.trunk.blocks.13.attn.qkv.bias")
   [node_scaled_dot_product_attention_13_qkv_split] node_scaled_dot_product_attention_13_q, node_scaled_dot_product_attention_13_k, node_scaled_dot_product_attention_13_v = Split <axis: int = -1> (linear_52, attn3d_split_3x1152)
   scaled_dot_product_attention_13 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_13_q, node_scaled_dot_product_attention_13_k, node_scaled_dot_product_attention_13_v)
   val_168 = MatMul (scaled_dot_product_attention_13, val_56)
   linear_53 = Add (val_168, "visual.trunk.blocks.13.attn.proj.bias")
   add_1248 = Add (add_1187, linear_53)
   layer_norm_27 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1248, "visual.trunk.blocks.13.norm2.weight", "visual.trunk.blocks.13.norm2.bias")
   val_169 = MatMul (layer_norm_27, val_57)
   linear_54 = Add (val_169, "visual.trunk.blocks.13.mlp.fc1.bias")
   gelu_13 = Gelu <approximate: string = "none"> (linear_54)
   val_170 = MatMul (gelu_13, val_58)
   linear_55 = Add (val_170, "visual.trunk.blocks.13.mlp.fc2.bias")
   add_1277 = Add (add_1248, linear_55)
   layer_norm_28 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1277, "visual.trunk.blocks.14.norm1.weight", "visual.trunk.blocks.14.norm1.bias")
   val_171 = MatMul (layer_norm_28, val_59)
   linear_56 = Add (val_171, "visual.trunk.blocks.14.attn.qkv.bias")
   [node_scaled_dot_product_attention_14_qkv_split] node_scaled_dot_product_attention_14_q, node_scaled_dot_product_attention_14_k, node_scaled_dot_product_attention_14_v = Split <axis: int = -1> (linear_56, attn3d_split_3x1152)
   scaled_dot_product_attention_14 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_14_q, node_scaled_dot_product_attention_14_k, node_scaled_dot_product_attention_14_v)
   val_172 = MatMul (scaled_dot_product_attention_14, val_60)
   linear_57 = Add (val_172, "visual.trunk.blocks.14.attn.proj.bias")
   add_1338 = Add (add_1277, linear_57)
   layer_norm_29 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1338, "visual.trunk.blocks.14.norm2.weight", "visual.trunk.blocks.14.norm2.bias")
   val_173 = MatMul (layer_norm_29, val_61)
   linear_58 = Add (val_173, "visual.trunk.blocks.14.mlp.fc1.bias")
   gelu_14 = Gelu <approximate: string = "none"> (linear_58)
   val_174 = MatMul (gelu_14, val_62)
   linear_59 = Add (val_174, "visual.trunk.blocks.14.mlp.fc2.bias")
   add_1367 = Add (add_1338, linear_59)
   layer_norm_30 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1367, "visual.trunk.blocks.15.norm1.weight", "visual.trunk.blocks.15.norm1.bias")
   val_175 = MatMul (layer_norm_30, val_63)
   linear_60 = Add (val_175, "visual.trunk.blocks.15.attn.qkv.bias")
   [node_scaled_dot_product_attention_15_qkv_split] node_scaled_dot_product_attention_15_q, node_scaled_dot_product_attention_15_k, node_scaled_dot_product_attention_15_v = Split <axis: int = -1> (linear_60, attn3d_split_3x1152)
   scaled_dot_product_attention_15 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_15_q, node_scaled_dot_product_attention_15_k, node_scaled_dot_product_attention_15_v)
   val_176 = MatMul (scaled_dot_product_attention_15, val_64)
   linear_61 = Add (val_176, "visual.trunk.blocks.15.attn.proj.bias")
   add_1428 = Add (add_1367, linear_61)
   layer_norm_31 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1428, "visual.trunk.blocks.15.norm2.weight", "visual.trunk.blocks.15.norm2.bias")
   val_177 = MatMul (layer_norm_31, val_65)
   linear_62 = Add (val_177, "visual.trunk.blocks.15.mlp.fc1.bias")
   gelu_15 = Gelu <approximate: string = "none"> (linear_62)
   val_178 = MatMul (gelu_15, val_66)
   linear_63 = Add (val_178, "visual.trunk.blocks.15.mlp.fc2.bias")
   add_1457 = Add (add_1428, linear_63)
   layer_norm_32 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1457, "visual.trunk.blocks.16.norm1.weight", "visual.trunk.blocks.16.norm1.bias")
   val_179 = MatMul (layer_norm_32, val_67)
   linear_64 = Add (val_179, "visual.trunk.blocks.16.attn.qkv.bias")
   [node_scaled_dot_product_attention_16_qkv_split] node_scaled_dot_product_attention_16_q, node_scaled_dot_product_attention_16_k, node_scaled_dot_product_attention_16_v = Split <axis: int = -1> (linear_64, attn3d_split_3x1152)
   scaled_dot_product_attention_16 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_16_q, node_scaled_dot_product_attention_16_k, node_scaled_dot_product_attention_16_v)
   val_180 = MatMul (scaled_dot_product_attention_16, val_68)
   linear_65 = Add (val_180, "visual.trunk.blocks.16.attn.proj.bias")
   add_1518 = Add (add_1457, linear_65)
   layer_norm_33 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1518, "visual.trunk.blocks.16.norm2.weight", "visual.trunk.blocks.16.norm2.bias")
   val_181 = MatMul (layer_norm_33, val_69)
   linear_66 = Add (val_181, "visual.trunk.blocks.16.mlp.fc1.bias")
   gelu_16 = Gelu <approximate: string = "none"> (linear_66)
   val_182 = MatMul (gelu_16, val_70)
   linear_67 = Add (val_182, "visual.trunk.blocks.16.mlp.fc2.bias")
   add_1547 = Add (add_1518, linear_67)
   layer_norm_34 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1547, "visual.trunk.blocks.17.norm1.weight", "visual.trunk.blocks.17.norm1.bias")
   val_183 = MatMul (layer_norm_34, val_71)
   linear_68 = Add (val_183, "visual.trunk.blocks.17.attn.qkv.bias")
   [node_scaled_dot_product_attention_17_qkv_split] node_scaled_dot_product_attention_17_q, node_scaled_dot_product_attention_17_k, node_scaled_dot_product_attention_17_v = Split <axis: int = -1> (linear_68, attn3d_split_3x1152)
   scaled_dot_product_attention_17 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_17_q, node_scaled_dot_product_attention_17_k, node_scaled_dot_product_attention_17_v)
   val_184 = MatMul (scaled_dot_product_attention_17, val_72)
   linear_69 = Add (val_184, "visual.trunk.blocks.17.attn.proj.bias")
   add_1608 = Add (add_1547, linear_69)
   layer_norm_35 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1608, "visual.trunk.blocks.17.norm2.weight", "visual.trunk.blocks.17.norm2.bias")
   val_185 = MatMul (layer_norm_35, val_73)
   linear_70 = Add (val_185, "visual.trunk.blocks.17.mlp.fc1.bias")
   gelu_17 = Gelu <approximate: string = "none"> (linear_70)
   val_186 = MatMul (gelu_17, val_74)
   linear_71 = Add (val_186, "visual.trunk.blocks.17.mlp.fc2.bias")
   add_1637 = Add (add_1608, linear_71)
   layer_norm_36 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1637, "visual.trunk.blocks.18.norm1.weight", "visual.trunk.blocks.18.norm1.bias")
   val_187 = MatMul (layer_norm_36, val_75)
   linear_72 = Add (val_187, "visual.trunk.blocks.18.attn.qkv.bias")
   [node_scaled_dot_product_attention_18_qkv_split] node_scaled_dot_product_attention_18_q, node_scaled_dot_product_attention_18_k, node_scaled_dot_product_attention_18_v = Split <axis: int = -1> (linear_72, attn3d_split_3x1152)
   scaled_dot_product_attention_18 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_18_q, node_scaled_dot_product_attention_18_k, node_scaled_dot_product_attention_18_v)
   val_188 = MatMul (scaled_dot_product_attention_18, val_76)
   linear_73 = Add (val_188, "visual.trunk.blocks.18.attn.proj.bias")
   add_1698 = Add (add_1637, linear_73)
   layer_norm_37 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1698, "visual.trunk.blocks.18.norm2.weight", "visual.trunk.blocks.18.norm2.bias")
   val_189 = MatMul (layer_norm_37, val_77)
   linear_74 = Add (val_189, "visual.trunk.blocks.18.mlp.fc1.bias")
   gelu_18 = Gelu <approximate: string = "none"> (linear_74)
   val_190 = MatMul (gelu_18, val_78)
   linear_75 = Add (val_190, "visual.trunk.blocks.18.mlp.fc2.bias")
   add_1727 = Add (add_1698, linear_75)
   layer_norm_38 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1727, "visual.trunk.blocks.19.norm1.weight", "visual.trunk.blocks.19.norm1.bias")
   val_191 = MatMul (layer_norm_38, val_79)
   linear_76 = Add (val_191, "visual.trunk.blocks.19.attn.qkv.bias")
   [node_scaled_dot_product_attention_19_qkv_split] node_scaled_dot_product_attention_19_q, node_scaled_dot_product_attention_19_k, node_scaled_dot_product_attention_19_v = Split <axis: int = -1> (linear_76, attn3d_split_3x1152)
   scaled_dot_product_attention_19 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_19_q, node_scaled_dot_product_attention_19_k, node_scaled_dot_product_attention_19_v)
   val_192 = MatMul (scaled_dot_product_attention_19, val_80)
   linear_77 = Add (val_192, "visual.trunk.blocks.19.attn.proj.bias")
   add_1788 = Add (add_1727, linear_77)
   layer_norm_39 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1788, "visual.trunk.blocks.19.norm2.weight", "visual.trunk.blocks.19.norm2.bias")
   val_193 = MatMul (layer_norm_39, val_81)
   linear_78 = Add (val_193, "visual.trunk.blocks.19.mlp.fc1.bias")
   gelu_19 = Gelu <approximate: string = "none"> (linear_78)
   val_194 = MatMul (gelu_19, val_82)
   linear_79 = Add (val_194, "visual.trunk.blocks.19.mlp.fc2.bias")
   add_1817 = Add (add_1788, linear_79)
   layer_norm_40 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1817, "visual.trunk.blocks.20.norm1.weight", "visual.trunk.blocks.20.norm1.bias")
   val_195 = MatMul (layer_norm_40, val_83)
   linear_80 = Add (val_195, "visual.trunk.blocks.20.attn.qkv.bias")
   [node_scaled_dot_product_attention_20_qkv_split] node_scaled_dot_product_attention_20_q, node_scaled_dot_product_attention_20_k, node_scaled_dot_product_attention_20_v = Split <axis: int = -1> (linear_80, attn3d_split_3x1152)
   scaled_dot_product_attention_20 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_20_q, node_scaled_dot_product_attention_20_k, node_scaled_dot_product_attention_20_v)
   val_196 = MatMul (scaled_dot_product_attention_20, val_84)
   linear_81 = Add (val_196, "visual.trunk.blocks.20.attn.proj.bias")
   add_1878 = Add (add_1817, linear_81)
   layer_norm_41 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1878, "visual.trunk.blocks.20.norm2.weight", "visual.trunk.blocks.20.norm2.bias")
   val_197 = MatMul (layer_norm_41, val_85)
   linear_82 = Add (val_197, "visual.trunk.blocks.20.mlp.fc1.bias")
   gelu_20 = Gelu <approximate: string = "none"> (linear_82)
   val_198 = MatMul (gelu_20, val_86)
   linear_83 = Add (val_198, "visual.trunk.blocks.20.mlp.fc2.bias")
   add_1907 = Add (add_1878, linear_83)
   layer_norm_42 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1907, "visual.trunk.blocks.21.norm1.weight", "visual.trunk.blocks.21.norm1.bias")
   val_199 = MatMul (layer_norm_42, val_87)
   linear_84 = Add (val_199, "visual.trunk.blocks.21.attn.qkv.bias")
   [node_scaled_dot_product_attention_21_qkv_split] node_scaled_dot_product_attention_21_q, node_scaled_dot_product_attention_21_k, node_scaled_dot_product_attention_21_v = Split <axis: int = -1> (linear_84, attn3d_split_3x1152)
   scaled_dot_product_attention_21 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_21_q, node_scaled_dot_product_attention_21_k, node_scaled_dot_product_attention_21_v)
   val_200 = MatMul (scaled_dot_product_attention_21, val_88)
   linear_85 = Add (val_200, "visual.trunk.blocks.21.attn.proj.bias")
   add_1968 = Add (add_1907, linear_85)
   layer_norm_43 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1968, "visual.trunk.blocks.21.norm2.weight", "visual.trunk.blocks.21.norm2.bias")
   val_201 = MatMul (layer_norm_43, val_89)
   linear_86 = Add (val_201, "visual.trunk.blocks.21.mlp.fc1.bias")
   gelu_21 = Gelu <approximate: string = "none"> (linear_86)
   val_202 = MatMul (gelu_21, val_90)
   linear_87 = Add (val_202, "visual.trunk.blocks.21.mlp.fc2.bias")
   add_1997 = Add (add_1968, linear_87)
   layer_norm_44 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_1997, "visual.trunk.blocks.22.norm1.weight", "visual.trunk.blocks.22.norm1.bias")
   val_203 = MatMul (layer_norm_44, val_91)
   linear_88 = Add (val_203, "visual.trunk.blocks.22.attn.qkv.bias")
   [node_scaled_dot_product_attention_22_qkv_split] node_scaled_dot_product_attention_22_q, node_scaled_dot_product_attention_22_k, node_scaled_dot_product_attention_22_v = Split <axis: int = -1> (linear_88, attn3d_split_3x1152)
   scaled_dot_product_attention_22 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_22_q, node_scaled_dot_product_attention_22_k, node_scaled_dot_product_attention_22_v)
   val_204 = MatMul (scaled_dot_product_attention_22, val_92)
   linear_89 = Add (val_204, "visual.trunk.blocks.22.attn.proj.bias")
   add_2058 = Add (add_1997, linear_89)
   layer_norm_45 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2058, "visual.trunk.blocks.22.norm2.weight", "visual.trunk.blocks.22.norm2.bias")
   val_205 = MatMul (layer_norm_45, val_93)
   linear_90 = Add (val_205, "visual.trunk.blocks.22.mlp.fc1.bias")
   gelu_22 = Gelu <approximate: string = "none"> (linear_90)
   val_206 = MatMul (gelu_22, val_94)
   linear_91 = Add (val_206, "visual.trunk.blocks.22.mlp.fc2.bias")
   add_2087 = Add (add_2058, linear_91)
   layer_norm_46 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2087, "visual.trunk.blocks.23.norm1.weight", "visual.trunk.blocks.23.norm1.bias")
   val_207 = MatMul (layer_norm_46, val_95)
   linear_92 = Add (val_207, "visual.trunk.blocks.23.attn.qkv.bias")
   [node_scaled_dot_product_attention_23_qkv_split] node_scaled_dot_product_attention_23_q, node_scaled_dot_product_attention_23_k, node_scaled_dot_product_attention_23_v = Split <axis: int = -1> (linear_92, attn3d_split_3x1152)
   scaled_dot_product_attention_23 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_23_q, node_scaled_dot_product_attention_23_k, node_scaled_dot_product_attention_23_v)
   val_208 = MatMul (scaled_dot_product_attention_23, val_96)
   linear_93 = Add (val_208, "visual.trunk.blocks.23.attn.proj.bias")
   add_2148 = Add (add_2087, linear_93)
   layer_norm_47 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2148, "visual.trunk.blocks.23.norm2.weight", "visual.trunk.blocks.23.norm2.bias")
   val_209 = MatMul (layer_norm_47, val_97)
   linear_94 = Add (val_209, "visual.trunk.blocks.23.mlp.fc1.bias")
   gelu_23 = Gelu <approximate: string = "none"> (linear_94)
   val_210 = MatMul (gelu_23, val_98)
   linear_95 = Add (val_210, "visual.trunk.blocks.23.mlp.fc2.bias")
   add_2177 = Add (add_2148, linear_95)
   layer_norm_48 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2177, "visual.trunk.blocks.24.norm1.weight", "visual.trunk.blocks.24.norm1.bias")
   val_211 = MatMul (layer_norm_48, val_99)
   linear_96 = Add (val_211, "visual.trunk.blocks.24.attn.qkv.bias")
   [node_scaled_dot_product_attention_24_qkv_split] node_scaled_dot_product_attention_24_q, node_scaled_dot_product_attention_24_k, node_scaled_dot_product_attention_24_v = Split <axis: int = -1> (linear_96, attn3d_split_3x1152)
   scaled_dot_product_attention_24 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_24_q, node_scaled_dot_product_attention_24_k, node_scaled_dot_product_attention_24_v)
   val_212 = MatMul (scaled_dot_product_attention_24, val_100)
   linear_97 = Add (val_212, "visual.trunk.blocks.24.attn.proj.bias")
   add_2238 = Add (add_2177, linear_97)
   layer_norm_49 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2238, "visual.trunk.blocks.24.norm2.weight", "visual.trunk.blocks.24.norm2.bias")
   val_213 = MatMul (layer_norm_49, val_101)
   linear_98 = Add (val_213, "visual.trunk.blocks.24.mlp.fc1.bias")
   gelu_24 = Gelu <approximate: string = "none"> (linear_98)
   val_214 = MatMul (gelu_24, val_102)
   linear_99 = Add (val_214, "visual.trunk.blocks.24.mlp.fc2.bias")
   add_2267 = Add (add_2238, linear_99)
   layer_norm_50 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2267, "visual.trunk.blocks.25.norm1.weight", "visual.trunk.blocks.25.norm1.bias")
   val_215 = MatMul (layer_norm_50, val_103)
   linear_100 = Add (val_215, "visual.trunk.blocks.25.attn.qkv.bias")
   [node_scaled_dot_product_attention_25_qkv_split] node_scaled_dot_product_attention_25_q, node_scaled_dot_product_attention_25_k, node_scaled_dot_product_attention_25_v = Split <axis: int = -1> (linear_100, attn3d_split_3x1152)
   scaled_dot_product_attention_25 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_25_q, node_scaled_dot_product_attention_25_k, node_scaled_dot_product_attention_25_v)
   val_216 = MatMul (scaled_dot_product_attention_25, val_104)
   linear_101 = Add (val_216, "visual.trunk.blocks.25.attn.proj.bias")
   add_2328 = Add (add_2267, linear_101)
   layer_norm_51 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2328, "visual.trunk.blocks.25.norm2.weight", "visual.trunk.blocks.25.norm2.bias")
   val_217 = MatMul (layer_norm_51, val_105)
   linear_102 = Add (val_217, "visual.trunk.blocks.25.mlp.fc1.bias")
   gelu_25 = Gelu <approximate: string = "none"> (linear_102)
   val_218 = MatMul (gelu_25, val_106)
   linear_103 = Add (val_218, "visual.trunk.blocks.25.mlp.fc2.bias")
   add_2357 = Add (add_2328, linear_103)
   layer_norm_52 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2357, "visual.trunk.blocks.26.norm1.weight", "visual.trunk.blocks.26.norm1.bias")
   val_219 = MatMul (layer_norm_52, val_107)
   linear_104 = Add (val_219, "visual.trunk.blocks.26.attn.qkv.bias")
   [node_scaled_dot_product_attention_26_qkv_split] node_scaled_dot_product_attention_26_q, node_scaled_dot_product_attention_26_k, node_scaled_dot_product_attention_26_v = Split <axis: int = -1> (linear_104, attn3d_split_3x1152)
   scaled_dot_product_attention_26 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_26_q, node_scaled_dot_product_attention_26_k, node_scaled_dot_product_attention_26_v)
   val_220 = MatMul (scaled_dot_product_attention_26, val_108)
   linear_105 = Add (val_220, "visual.trunk.blocks.26.attn.proj.bias")
   add_2418 = Add (add_2357, linear_105)
   layer_norm_53 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2418, "visual.trunk.blocks.26.norm2.weight", "visual.trunk.blocks.26.norm2.bias")
   val_221 = MatMul (layer_norm_53, val_109)
   linear_106 = Add (val_221, "visual.trunk.blocks.26.mlp.fc1.bias")
   gelu_26 = Gelu <approximate: string = "none"> (linear_106)
   val_222 = MatMul (gelu_26, val_110)
   linear_107 = Add (val_222, "visual.trunk.blocks.26.mlp.fc2.bias")
   add_2447 = Add (add_2418, linear_107)
   layer_norm_54 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (add_2447, "visual.trunk.norm.weight", "visual.trunk.norm.bias")
   [ez_slice] image_ez_s = Slice (image_f32, image_ez_starts, image_ez_ends, image_ez_axes)
   [ez_flatten] image_ez_r = Reshape (image_ez_s, val_0)
   [ez_mul] image_ez = Mul (image_ez_r, image_ez_zero)
   val_223 = MatMul (layer_norm_54, val_111)
   linear_109 = Add (val_223, "visual.trunk.attn_pool.kv.bias")
   [node_scaled_dot_product_attention_27_qkv_split] node_scaled_dot_product_attention_27_k, node_scaled_dot_product_attention_27_v = Split <axis: int = -1> (linear_109, attn3d_split_2x1152)
   [node_scaled_dot_product_attention_27_q_col] node_scaled_dot_product_attention_27_q_col_out = Unsqueeze (image_ez, node_scaled_dot_product_attention_27_q_col_axes)
   [node_scaled_dot_product_attention_27_q_bcast] node_scaled_dot_product_attention_27_q = Add (node_scaled_dot_product_attention_27_q3, node_scaled_dot_product_attention_27_q_col_out)
   scaled_dot_product_attention_27 = Attention <is_causal: int = 0, kv_num_heads: int = 16, q_num_heads: int = 16, qk_matmul_output_mode: int = 0, softcap: float = 0> (node_scaled_dot_product_attention_27_q, node_scaled_dot_product_attention_27_k, node_scaled_dot_product_attention_27_v)
   val_224 = MatMul (scaled_dot_product_attention_27, val_112)
   linear_110 = Add (val_224, "visual.trunk.attn_pool.proj.bias")
   layer_norm_55 = LayerNormalization <axis: int = -1, epsilon: float = 1e-06, stash_type: int = 1> (linear_110, "visual.trunk.attn_pool.norm.weight", "visual.trunk.attn_pool.norm.bias")
   val_225 = MatMul (layer_norm_55, val_113)
   linear_111 = Add (val_225, "visual.trunk.attn_pool.mlp.fc1.bias")
   gelu_27 = Gelu <approximate: string = "none"> (linear_111)
   val_226 = MatMul (gelu_27, val_114)
   linear_112 = Add (val_226, "visual.trunk.attn_pool.mlp.fc2.bias")
   add_2545 = Add (linear_110, linear_112)
   select = Squeeze (add_2545, val_2)
   [node_linalg_vector_norm] linalg_vector_norm = ReduceL2 <keepdims: int = 1, noop_with_empty_axes: int = 0> (select, val_0)
   [node_clamp_min] clamp_min = Clip (linalg_vector_norm, val_1)
   [node_div] image_embedding = Div (select, clamp_min)
}

weights:
attn3d_split_2x1152 INT64[2] 414304bef26c
attn3d_split_3x1152 INT64[3] 125b254aeb28
image_ez_axes INT64[3] e2e2033ae7e1
image_ez_ends INT64[3] 605390e5a369
image_ez_starts INT64[3] 9d908ecfb6b2
image_ez_zero FLOAT[] df3f619804a9
node_scaled_dot_product_attention_27_q3 FLOAT[1,1,1152] b5b3f2332217
node_scaled_dot_product_attention_27_q_col_axes INT64[2] 0c730b69905c
val_0 INT64[1] 12a3ae445661
val_1 FLOAT[] 6708d9be4956
val_10 FLOAT[4304,1152] 014d13f758b0
val_100 FLOAT[1152,1152] 648f236a31cd
val_101 FLOAT[1152,4304] b8fef72e2fdc
val_102 FLOAT[4304,1152] e301247ee262
val_103 FLOAT[1152,3456] 5c6cce6ea44c
val_104 FLOAT[1152,1152] f1c7ec954448
val_105 FLOAT[1152,4304] d9955c52c4bd
val_106 FLOAT[4304,1152] 1beb4cb38778
val_107 FLOAT[1152,3456] e6bab83fb4df
val_108 FLOAT[1152,1152] 856ab16af8c3
val_109 FLOAT[1152,4304] 6897690fb2e6
val_11 FLOAT[1152,3456] c6b299d8eadb
val_110 FLOAT[4304,1152] c95c20fd4e6d
val_111 FLOAT[1152,2304] 15a00dbfdce5
val_112 FLOAT[1152,1152] b1b5761d6c57
val_113 FLOAT[1152,4304] e7afada6a12e
val_114 FLOAT[4304,1152] 51ccef2276f5
val_12 FLOAT[1152,1152] 12c4419a7f12
val_13 FLOAT[1152,4304] fdc74382043c
val_14 FLOAT[4304,1152] 1de8f56b6aa4
val_15 FLOAT[1152,3456] 3badb12d1195
val_16 FLOAT[1152,1152] f11d068f1447
val_17 FLOAT[1152,4304] 8f060f24d57b
val_18 FLOAT[4304,1152] 67281f70d099
val_19 FLOAT[1152,3456] 3f56aa63e847
val_2 INT64[1] 7c9fa136d441
val_20 FLOAT[1152,1152] dabcac77ed62
val_21 FLOAT[1152,4304] 9064d1644292
val_22 FLOAT[4304,1152] 24cd961740aa
val_23 FLOAT[1152,3456] 29ebe8f26723
val_24 FLOAT[1152,1152] 35004161f828
val_25 FLOAT[1152,4304] cb8f4afd79a7
val_26 FLOAT[4304,1152] 5180388c2de1
val_27 FLOAT[1152,3456] dcd3e97b40ba
val_28 FLOAT[1152,1152] 3293018abf7b
val_29 FLOAT[1152,4304] 6883788e404a
val_3 FLOAT[1152,3456] 460db31a4a73
val_30 FLOAT[4304,1152] 0ad7ddb683e2
val_31 FLOAT[1152,3456] 86a9e1efaa6b
val_32 FLOAT[1152,1152] 7c18ee90232f
val_33 FLOAT[1152,4304] 997951b2b591
val_34 FLOAT[4304,1152] 870dc3609d55
val_35 FLOAT[1152,3456] bc48b36a652e
val_36 FLOAT[1152,1152] 72364694d3b8
val_37 FLOAT[1152,4304] f8d407a7c463
val_38 FLOAT[4304,1152] bfb467e9f376
val_39 FLOAT[1152,3456] b8b993a088c5
val_4 FLOAT[1152,1152] eb4995b1151b
val_40 FLOAT[1152,1152] 3b16404a39df
val_41 FLOAT[1152,4304] c8fe0fd4c1f2
val_42 FLOAT[4304,1152] 8a1235250e8f
val_43 FLOAT[1152,3456] 5ae54a17e5e8
val_44 FLOAT[1152,1152] aaf5bd1ada81
val_45 FLOAT[1152,4304] def7b2df4483
val_46 FLOAT[4304,1152] 2c76b7ddc686
val_47 FLOAT[1152,3456] bfd7c64171a3
val_48 FLOAT[1152,1152] 31f5fa773681
val_49 FLOAT[1152,4304] 9f82d0c3546f
val_5 FLOAT[1152,4304] f2584823c302
val_50 FLOAT[4304,1152] ce289a0b6a23
val_51 FLOAT[1152,3456] 6f0adb6b8af9
val_52 FLOAT[1152,1152] e24ed2ac7140
val_53 FLOAT[1152,4304] 1a40ba074fd6
val_54 FLOAT[4304,1152] ee44ef45d83d
val_55 FLOAT[1152,3456] 218179d294f1
val_56 FLOAT[1152,1152] e37eb81d99b9
val_57 FLOAT[1152,4304] 670ea5b9abaf
val_58 FLOAT[4304,1152] b0a723589dce
val_59 FLOAT[1152,3456] ab3b3673e787
val_6 FLOAT[4304,1152] 4ab277a82919
val_60 FLOAT[1152,1152] 7ed0b165d1a6
val_61 FLOAT[1152,4304] c7b17b021597
val_62 FLOAT[4304,1152] bc1d0a36da81
val_63 FLOAT[1152,3456] a20b31e03616
val_64 FLOAT[1152,1152] a1744c43aedb
val_65 FLOAT[1152,4304] 76f0805770aa
val_66 FLOAT[4304,1152] 17d6db3d760d
val_67 FLOAT[1152,3456] 91958bdb5746
val_68 FLOAT[1152,1152] 2a8d3fb6052c
val_69 FLOAT[1152,4304] ecdb54dc0f06
val_7 FLOAT[1152,3456] 89766cedda26
val_70 FLOAT[4304,1152] f8fdb20beae3
val_71 FLOAT[1152,3456] 617130c41a0e
val_72 FLOAT[1152,1152] 2782f05c0ac0
val_73 FLOAT[1152,4304] 39bbfd512031
val_74 FLOAT[4304,1152] b3ba1f4949be
val_75 FLOAT[1152,3456] b206122e61b6
val_76 FLOAT[1152,1152] f3c05e892fa9
val_77 FLOAT[1152,4304] 249a00d4990a
val_78 FLOAT[4304,1152] 1e29abcf59a8
val_79 FLOAT[1152,3456] 1505df61b5b4
val_8 FLOAT[1152,1152] e1511353e0ec
val_80 FLOAT[1152,1152] da2c527fea18
val_81 FLOAT[1152,4304] 637cec58406d
val_82 FLOAT[4304,1152] cd9524b7b3f0
val_83 FLOAT[1152,3456] 82ee0d75ddd8
val_84 FLOAT[1152,1152] fec53aeeb301
val_85 FLOAT[1152,4304] f0490c063d80
val_86 FLOAT[4304,1152] f860b00306e8
val_87 FLOAT[1152,3456] 1a6ab7486386
val_88 FLOAT[1152,1152] 84debe571415
val_89 FLOAT[1152,4304] 06243e1a7f1c
val_9 FLOAT[1152,4304] 27b784a9ee35
val_90 FLOAT[4304,1152] ae800cda403b
val_91 FLOAT[1152,3456] 7db1ae5083fc
val_92 FLOAT[1152,1152] fea170d35e45
val_93 FLOAT[1152,4304] 7905274423bd
val_94 FLOAT[4304,1152] dd76c0816c98
val_95 FLOAT[1152,3456] bd1415d2d238
val_96 FLOAT[1152,1152] 78b449d95823
val_97 FLOAT[1152,4304] b7f8334e35f4
val_98 FLOAT[4304,1152] 98900e629b0e
val_99 FLOAT[1152,3456] 82d1c0492622
view_target INT64[3] 8af5235ecd35
visual.trunk.attn_pool.kv.bias FLOAT[2304] 27053adab941
visual.trunk.attn_pool.mlp.fc1.bias FLOAT[4304] af910f60b95c
visual.trunk.attn_pool.mlp.fc2.bias FLOAT[1152] 23d0e2d4d4d6
visual.trunk.attn_pool.norm.bias FLOAT[1152] 94b04aba7e8d
visual.trunk.attn_pool.norm.weight FLOAT[1152] 3857a1b51745
visual.trunk.attn_pool.proj.bias FLOAT[1152] d847abb576d0
visual.trunk.blocks.0.attn.proj.bias FLOAT[1152] c0a44182b8ba
visual.trunk.blocks.0.attn.qkv.bias FLOAT[3456] ad88b0e65929
visual.trunk.blocks.0.mlp.fc1.bias FLOAT[4304] 4a9214d9a31b
visual.trunk.blocks.0.mlp.fc2.bias FLOAT[1152] df6f4396928b
visual.trunk.blocks.0.norm1.bias FLOAT[1152] 156595ad141e
visual.trunk.blocks.0.norm1.weight FLOAT[1152] 7def83f120b4
visual.trunk.blocks.0.norm2.bias FLOAT[1152] 49764a213326
visual.trunk.blocks.0.norm2.weight FLOAT[1152] a07f33bd30de
visual.trunk.blocks.1.attn.proj.bias FLOAT[1152] aec41a88d80f
visual.trunk.blocks.1.attn.qkv.bias FLOAT[3456] 3892bc0d8e83
visual.trunk.blocks.1.mlp.fc1.bias FLOAT[4304] ed64e399e35a
visual.trunk.blocks.1.mlp.fc2.bias FLOAT[1152] fef28089ffa0
visual.trunk.blocks.1.norm1.bias FLOAT[1152] 140edd9590a3
visual.trunk.blocks.1.norm1.weight FLOAT[1152] 110f954b979a
visual.trunk.blocks.1.norm2.bias FLOAT[1152] 0a91a5ec2b80
visual.trunk.blocks.1.norm2.weight FLOAT[1152] 554a4aee33d8
visual.trunk.blocks.10.attn.proj.bias FLOAT[1152] d66b87899718
visual.trunk.blocks.10.attn.qkv.bias FLOAT[3456] 5d8d9f03589d
visual.trunk.blocks.10.mlp.fc1.bias FLOAT[4304] 7cc0c881b556
visual.trunk.blocks.10.mlp.fc2.bias FLOAT[1152] 3b8d6af2580c
visual.trunk.blocks.10.norm1.bias FLOAT[1152] 1f9f95190024
visual.trunk.blocks.10.norm1.weight FLOAT[1152] 988af5c77630
visual.trunk.blocks.10.norm2.bias FLOAT[1152] 35cdd987f011
visual.trunk.blocks.10.norm2.weight FLOAT[1152] 5c5dc88b42f7
visual.trunk.blocks.11.attn.proj.bias FLOAT[1152] ec871e0c7f71
visual.trunk.blocks.11.attn.qkv.bias FLOAT[3456] 7b67a074cef0
visual.trunk.blocks.11.mlp.fc1.bias FLOAT[4304] 282ae3b2096a
visual.trunk.blocks.11.mlp.fc2.bias FLOAT[1152] 4111764164e5
visual.trunk.blocks.11.norm1.bias FLOAT[1152] 4988d3424f37
visual.trunk.blocks.11.norm1.weight FLOAT[1152] 66b6e3052177
visual.trunk.blocks.11.norm2.bias FLOAT[1152] b60b23460366
visual.trunk.blocks.11.norm2.weight FLOAT[1152] 73c597d8e130
visual.trunk.blocks.12.attn.proj.bias FLOAT[1152] 5451649311e4
visual.trunk.blocks.12.attn.qkv.bias FLOAT[3456] 7e4e79493010
visual.trunk.blocks.12.mlp.fc1.bias FLOAT[4304] 1c5cf483513f
visual.trunk.blocks.12.mlp.fc2.bias FLOAT[1152] 532904717664
visual.trunk.blocks.12.norm1.bias FLOAT[1152] d96644f02b14
visual.trunk.blocks.12.norm1.weight FLOAT[1152] 2e757e443495
visual.trunk.blocks.12.norm2.bias FLOAT[1152] 1a9af94779dd
visual.trunk.blocks.12.norm2.weight FLOAT[1152] 764880e8b24c
visual.trunk.blocks.13.attn.proj.bias FLOAT[1152] 606a14d9d8a0
visual.trunk.blocks.13.attn.qkv.bias FLOAT[3456] 8d9eabc7333b
visual.trunk.blocks.13.mlp.fc1.bias FLOAT[4304] 06404555cee1
visual.trunk.blocks.13.mlp.fc2.bias FLOAT[1152] 7b4c6a0a87f4
visual.trunk.blocks.13.norm1.bias FLOAT[1152] f60b1fc1a8ff
visual.trunk.blocks.13.norm1.weight FLOAT[1152] f9bf46dbaeda
visual.trunk.blocks.13.norm2.bias FLOAT[1152] 7afae91754e7
visual.trunk.blocks.13.norm2.weight FLOAT[1152] 28e6c59f2308
visual.trunk.blocks.14.attn.proj.bias FLOAT[1152] df84144a11a7
visual.trunk.blocks.14.attn.qkv.bias FLOAT[3456] 558bd3b8f686
visual.trunk.blocks.14.mlp.fc1.bias FLOAT[4304] f01de232ecaa
visual.trunk.blocks.14.mlp.fc2.bias FLOAT[1152] bc019f8bded3
visual.trunk.blocks.14.norm1.bias FLOAT[1152] 9e81c4e74ad0
visual.trunk.blocks.14.norm1.weight FLOAT[1152] 04a78b6f5fd9
visual.trunk.blocks.14.norm2.bias FLOAT[1152] 4b86887ed751
visual.trunk.blocks.14.norm2.weight FLOAT[1152] 33ad38f1510c
visual.trunk.blocks.15.attn.proj.bias FLOAT[1152] 9792f89dde39
visual.trunk.blocks.15.attn.qkv.bias FLOAT[3456] 91e2509501fb
visual.trunk.blocks.15.mlp.fc1.bias FLOAT[4304] 4659e7d4e3f6
visual.trunk.blocks.15.mlp.fc2.bias FLOAT[1152] b9bb6a18fcab
visual.trunk.blocks.15.norm1.bias FLOAT[1152] 9ee314c7ce44
visual.trunk.blocks.15.norm1.weight FLOAT[1152] 50c959ee8e09
visual.trunk.blocks.15.norm2.bias FLOAT[1152] 99aacda67c4b
visual.trunk.blocks.15.norm2.weight FLOAT[1152] 7c51b164a447
visual.trunk.blocks.16.attn.proj.bias FLOAT[1152] 0a3aba1bef7e
visual.trunk.blocks.16.attn.qkv.bias FLOAT[3456] 43973d33048c
visual.trunk.blocks.16.mlp.fc1.bias FLOAT[4304] 8b46a2648226
visual.trunk.blocks.16.mlp.fc2.bias FLOAT[1152] ab53665f5a1e
visual.trunk.blocks.16.norm1.bias FLOAT[1152] 59009383d287
visual.trunk.blocks.16.norm1.weight FLOAT[1152] 9cbb23a2d89b
visual.trunk.blocks.16.norm2.bias FLOAT[1152] aa6017e7e2c8
visual.trunk.blocks.16.norm2.weight FLOAT[1152] 5bbc4d0238ca
visual.trunk.blocks.17.attn.proj.bias FLOAT[1152] d9986151b3f6
visual.trunk.blocks.17.attn.qkv.bias FLOAT[3456] f6cd8067c3dc
visual.trunk.blocks.17.mlp.fc1.bias FLOAT[4304] 2670cea97d69
visual.trunk.blocks.17.mlp.fc2.bias FLOAT[1152] 403934f0e2a0
visual.trunk.blocks.17.norm1.bias FLOAT[1152] 3fe4da65f0c2
visual.trunk.blocks.17.norm1.weight FLOAT[1152] 5c3f4f2829dd
visual.trunk.blocks.17.norm2.bias FLOAT[1152] b47c70c7a6f9
visual.trunk.blocks.17.norm2.weight FLOAT[1152] 30cba3aeeaec
visual.trunk.blocks.18.attn.proj.bias FLOAT[1152] c21c90290234
visual.trunk.blocks.18.attn.qkv.bias FLOAT[3456] 86ce4aa97e01
visual.trunk.blocks.18.mlp.fc1.bias FLOAT[4304] b6b1d9387c9d
visual.trunk.blocks.18.mlp.fc2.bias FLOAT[1152] dbfa39a531a9
visual.trunk.blocks.18.norm1.bias FLOAT[1152] e764db9b37e2
visual.trunk.blocks.18.norm1.weight FLOAT[1152] 5c8fa1eca11f
visual.trunk.blocks.18.norm2.bias FLOAT[1152] 8f1b07e4d759
visual.trunk.blocks.18.norm2.weight FLOAT[1152] a015154c06f8
visual.trunk.blocks.19.attn.proj.bias FLOAT[1152] 702aad8f162d
visual.trunk.blocks.19.attn.qkv.bias FLOAT[3456] 49764bd15e7c
visual.trunk.blocks.19.mlp.fc1.bias FLOAT[4304] ebefbf01df40
visual.trunk.blocks.19.mlp.fc2.bias FLOAT[1152] 582fed5a718e
visual.trunk.blocks.19.norm1.bias FLOAT[1152] c2fd3dff24f6
visual.trunk.blocks.19.norm1.weight FLOAT[1152] 3416fd606de2
visual.trunk.blocks.19.norm2.bias FLOAT[1152] 107d6cc5b300
visual.trunk.blocks.19.norm2.weight FLOAT[1152] 1f6e0bd8687a
visual.trunk.blocks.2.attn.proj.bias FLOAT[1152] 627e55c59924
visual.trunk.blocks.2.attn.qkv.bias FLOAT[3456] a4dcdec00e42
visual.trunk.blocks.2.mlp.fc1.bias FLOAT[4304] c7a698a63010
visual.trunk.blocks.2.mlp.fc2.bias FLOAT[1152] e5684bcbd17c
visual.trunk.blocks.2.norm1.bias FLOAT[1152] ca736f11630f
visual.trunk.blocks.2.norm1.weight FLOAT[1152] 6dfdcb120e76
visual.trunk.blocks.2.norm2.bias FLOAT[1152] 7f628fae8c67
visual.trunk.blocks.2.norm2.weight FLOAT[1152] e2bc54fb61c4
visual.trunk.blocks.20.attn.proj.bias FLOAT[1152] 223fb4d33f8e
visual.trunk.blocks.20.attn.qkv.bias FLOAT[3456] acb290d49f81
visual.trunk.blocks.20.mlp.fc1.bias FLOAT[4304] fbf10da9e64b
visual.trunk.blocks.20.mlp.fc2.bias FLOAT[1152] dce3c39bac0c
visual.trunk.blocks.20.norm1.bias FLOAT[1152] f86fd4efa6e6
visual.trunk.blocks.20.norm1.weight FLOAT[1152] c7bcd3f68b85
visual.trunk.blocks.20.norm2.bias FLOAT[1152] 962b75fde85a
visual.trunk.blocks.20.norm2.weight FLOAT[1152] 81e475f83bc1
visual.trunk.blocks.21.attn.proj.bias FLOAT[1152] c398e0dfd51a
visual.trunk.blocks.21.attn.qkv.bias FLOAT[3456] e1f70ceb09ec
visual.trunk.blocks.21.mlp.fc1.bias FLOAT[4304] ea861be8a984
visual.trunk.blocks.21.mlp.fc2.bias FLOAT[1152] 881b4cc70f1a
visual.trunk.blocks.21.norm1.bias FLOAT[1152] e1b0c7ef4779
visual.trunk.blocks.21.norm1.weight FLOAT[1152] 3b474e688e1f
visual.trunk.blocks.21.norm2.bias FLOAT[1152] 5c1afb4d545f
visual.trunk.blocks.21.norm2.weight FLOAT[1152] 056bcd56388e
visual.trunk.blocks.22.attn.proj.bias FLOAT[1152] 65400a02b292
visual.trunk.blocks.22.attn.qkv.bias FLOAT[3456] f3927b7e3912
visual.trunk.blocks.22.mlp.fc1.bias FLOAT[4304] 3c5a1e35a467
visual.trunk.blocks.22.mlp.fc2.bias FLOAT[1152] bb1334fd40c6
visual.trunk.blocks.22.norm1.bias FLOAT[1152] 1a7a379d8485
visual.trunk.blocks.22.norm1.weight FLOAT[1152] 846ae988b707
visual.trunk.blocks.22.norm2.bias FLOAT[1152] 1210338a8eb2
visual.trunk.blocks.22.norm2.weight FLOAT[1152] 7c6e934e7193
visual.trunk.blocks.23.attn.proj.bias FLOAT[1152] 96ef5791e984
visual.trunk.blocks.23.attn.qkv.bias FLOAT[3456] 3b9b1943aaef
visual.trunk.blocks.23.mlp.fc1.bias FLOAT[4304] 3cc1e298c254
visual.trunk.blocks.23.mlp.fc2.bias FLOAT[1152] 4d1255af05d4
visual.trunk.blocks.23.norm1.bias FLOAT[1152] ecf09397bc9f
visual.trunk.blocks.23.norm1.weight FLOAT[1152] eb6e56faf47a
visual.trunk.blocks.23.norm2.bias FLOAT[1152] c7b0393c5ad9
visual.trunk.blocks.23.norm2.weight FLOAT[1152] 72faa7f310a4
visual.trunk.blocks.24.attn.proj.bias FLOAT[1152] cc2236a4cd32
visual.trunk.blocks.24.attn.qkv.bias FLOAT[3456] 66fc778722e3
visual.trunk.blocks.24.mlp.fc1.bias FLOAT[4304] b441d3dc0579
visual.trunk.blocks.24.mlp.fc2.bias FLOAT[1152] 03bce54c6d35
visual.trunk.blocks.24.norm1.bias FLOAT[1152] ee05e55f3bcc
visual.trunk.blocks.24.norm1.weight FLOAT[1152] 62e0e42320a5
visual.trunk.blocks.24.norm2.bias FLOAT[1152] 4338d19c4253
visual.trunk.blocks.24.norm2.weight FLOAT[1152] 144a64c2939f
visual.trunk.blocks.25.attn.proj.bias FLOAT[1152] e65243ffe002
visual.trunk.blocks.25.attn.qkv.bias FLOAT[3456] dbf8870b5f7d
visual.trunk.blocks.25.mlp.fc1.bias FLOAT[4304] b20295d6e9b2
visual.trunk.blocks.25.mlp.fc2.bias FLOAT[1152] 8f3585a2504a
visual.trunk.blocks.25.norm1.bias FLOAT[1152] 4b8078c025d5
visual.trunk.blocks.25.norm1.weight FLOAT[1152] c5feef6f1b75
visual.trunk.blocks.25.norm2.bias FLOAT[1152] aef773cac302
visual.trunk.blocks.25.norm2.weight FLOAT[1152] 2acbd9acf4eb
visual.trunk.blocks.26.attn.proj.bias FLOAT[1152] 820b06b0f738
visual.trunk.blocks.26.attn.qkv.bias FLOAT[3456] d313faba1e07
visual.trunk.blocks.26.mlp.fc1.bias FLOAT[4304] a321a504235f
visual.trunk.blocks.26.mlp.fc2.bias FLOAT[1152] 6cd72bdb2f9e
visual.trunk.blocks.26.norm1.bias FLOAT[1152] 0609cb60ba54
visual.trunk.blocks.26.norm1.weight FLOAT[1152] 4e52e134d1fa
visual.trunk.blocks.26.norm2.bias FLOAT[1152] 8091d06afb9c
visual.trunk.blocks.26.norm2.weight FLOAT[1152] 32dc13dad69b
visual.trunk.blocks.3.attn.proj.bias FLOAT[1152] a96bebc7d6ab
visual.trunk.blocks.3.attn.qkv.bias FLOAT[3456] 34a456089888
visual.trunk.blocks.3.mlp.fc1.bias FLOAT[4304] fc73beb5d205
visual.trunk.blocks.3.mlp.fc2.bias FLOAT[1152] 621f413f47bc
visual.trunk.blocks.3.norm1.bias FLOAT[1152] f9ceeb1ee63b
visual.trunk.blocks.3.norm1.weight FLOAT[1152] a3116b61155c
visual.trunk.blocks.3.norm2.bias FLOAT[1152] 0cd844ea26ea
visual.trunk.blocks.3.norm2.weight FLOAT[1152] 6fd3416e7d72
visual.trunk.blocks.4.attn.proj.bias FLOAT[1152] 592f06aa6abb
visual.trunk.blocks.4.attn.qkv.bias FLOAT[3456] 691e07293a86
visual.trunk.blocks.4.mlp.fc1.bias FLOAT[4304] 276b0d7d1c36
visual.trunk.blocks.4.mlp.fc2.bias FLOAT[1152] fe5f62207894
visual.trunk.blocks.4.norm1.bias FLOAT[1152] f2687b5c754d
visual.trunk.blocks.4.norm1.weight FLOAT[1152] 2d75c1188460
visual.trunk.blocks.4.norm2.bias FLOAT[1152] cc74fb7bb343
visual.trunk.blocks.4.norm2.weight FLOAT[1152] b0aacea2be29
visual.trunk.blocks.5.attn.proj.bias FLOAT[1152] bded89198750
visual.trunk.blocks.5.attn.qkv.bias FLOAT[3456] 4e2ca8a272a3
visual.trunk.blocks.5.mlp.fc1.bias FLOAT[4304] 89c3fe100459
visual.trunk.blocks.5.mlp.fc2.bias FLOAT[1152] ec026eacd0d7
visual.trunk.blocks.5.norm1.bias FLOAT[1152] ed6932a5e85a
visual.trunk.blocks.5.norm1.weight FLOAT[1152] c51ceb35acab
visual.trunk.blocks.5.norm2.bias FLOAT[1152] a1f93d10a669
visual.trunk.blocks.5.norm2.weight FLOAT[1152] 825768ac7fa8
visual.trunk.blocks.6.attn.proj.bias FLOAT[1152] 04cb64695954
visual.trunk.blocks.6.attn.qkv.bias FLOAT[3456] 012709cd0489
visual.trunk.blocks.6.mlp.fc1.bias FLOAT[4304] 7645239d83b9
visual.trunk.blocks.6.mlp.fc2.bias FLOAT[1152] 19412c22ef71
visual.trunk.blocks.6.norm1.bias FLOAT[1152] f2b3f0396340
visual.trunk.blocks.6.norm1.weight FLOAT[1152] ddaba654b79c
visual.trunk.blocks.6.norm2.bias FLOAT[1152] 4c0bd9d19499
visual.trunk.blocks.6.norm2.weight FLOAT[1152] 688f58f20433
visual.trunk.blocks.7.attn.proj.bias FLOAT[1152] f135fa475b11
visual.trunk.blocks.7.attn.qkv.bias FLOAT[3456] 0193231fc7f9
visual.trunk.blocks.7.mlp.fc1.bias FLOAT[4304] 8ce1c8ae311d
visual.trunk.blocks.7.mlp.fc2.bias FLOAT[1152] 5a60544653ad
visual.trunk.blocks.7.norm1.bias FLOAT[1152] eb71d3878100
visual.trunk.blocks.7.norm1.weight FLOAT[1152] f66f287e78b4
visual.trunk.blocks.7.norm2.bias FLOAT[1152] 8b0089507766
visual.trunk.blocks.7.norm2.weight FLOAT[1152] 55b666631ce3
visual.trunk.blocks.8.attn.proj.bias FLOAT[1152] 748e6cf46a65
visual.trunk.blocks.8.attn.qkv.bias FLOAT[3456] 93e4b5d1e879
visual.trunk.blocks.8.mlp.fc1.bias FLOAT[4304] 2b08cbac66f4
visual.trunk.blocks.8.mlp.fc2.bias FLOAT[1152] 023438d1154c
visual.trunk.blocks.8.norm1.bias FLOAT[1152] b11bc2cbe940
visual.trunk.blocks.8.norm1.weight FLOAT[1152] 70cfcfefcf01
visual.trunk.blocks.8.norm2.bias FLOAT[1152] b1197f04eaeb
visual.trunk.blocks.8.norm2.weight FLOAT[1152] f5a3de041e6f
visual.trunk.blocks.9.attn.proj.bias FLOAT[1152] dcb5aea01c94
visual.trunk.blocks.9.attn.qkv.bias FLOAT[3456] fbbd08e35104
visual.trunk.blocks.9.mlp.fc1.bias FLOAT[4304] 3a0bbb7a68cb
visual.trunk.blocks.9.mlp.fc2.bias FLOAT[1152] bcdb830f8bfc
visual.trunk.blocks.9.norm1.bias FLOAT[1152] 5784c38d6d68
visual.trunk.blocks.9.norm1.weight FLOAT[1152] c294b60232f7
visual.trunk.blocks.9.norm2.bias FLOAT[1152] 6d1160ab01d3
visual.trunk.blocks.9.norm2.weight FLOAT[1152] 295520a926b7
visual.trunk.norm.bias FLOAT[1152] f3c541310175
visual.trunk.norm.weight FLOAT[1152] 219b78805f09
visual.trunk.patch_embed.proj.bias FLOAT[1152] d15722e84f3f
visual.trunk.patch_embed.proj.weight FLOAT[1152,3,14,14] 144387d93841
visual.trunk.pos_embed FLOAT[1,729,1152] 25c36e03a73c
