mirror of
https://github.com/immich-app/ml-models.git
synced 2026-09-30 13:22:55 +08:00
Mert
df0ea33c8a
* feat: torch-free export core, shared graph/IR helpers, and CLI scaffolding * feat: fused InsightFace face exporter * feat: RKNN export path with per-SoC compilation * feat: single-partition CoreML CLIP export and runtime rewrite API * feat: refactor to passes, ep-specific fixes and optimizations * feat: check command diffing committed graph renderings, wired into CI * chore: graph and rewrite-plan renderings for the catalog * use pokedex large
1116 lines
89 KiB
Plaintext
1116 lines
89 KiB
Plaintext
<
|
|
ir_version: 10,
|
|
opset_import: ["" : 23],
|
|
producer_name: "pytorch"
|
|
>
|
|
main_graph (uint8[batch,224,224,3] image) => (float[batch,768] image_embedding)
|
|
<
|
|
float[batch,257,1024] add_1088
|
|
float[batch,257,1024] add_1109
|
|
float[batch,257,1024] add_1224
|
|
float[batch,257,1024] add_1245
|
|
float[batch,257,1024] add_136
|
|
float[batch,257,1024] add_1360
|
|
float[batch,257,1024] add_1381
|
|
float[batch,257,1024] add_1496
|
|
float[batch,257,1024] add_1517
|
|
float[batch,257,1024] add_157
|
|
float[batch,257,1024] add_1632
|
|
float[batch,257,1024] add_1653
|
|
float[batch,257,1024] add_17
|
|
float[batch,257,1024] add_1768
|
|
float[batch,257,1024] add_1789
|
|
float[batch,257,1024] add_1904
|
|
float[batch,257,1024] add_1925
|
|
float[batch,257,1024] add_2040
|
|
float[batch,257,1024] add_2061
|
|
float[batch,257,1024] add_2176
|
|
float[batch,257,1024] add_2197
|
|
float[batch,257,1024] add_2312
|
|
float[batch,257,1024] add_2333
|
|
float[batch,257,1024] add_2448
|
|
float[batch,257,1024] add_2469
|
|
float[batch,257,1024] add_2584
|
|
float[batch,257,1024] add_2605
|
|
float[batch,257,1024] add_272
|
|
float[batch,257,1024] add_2720
|
|
float[batch,257,1024] add_2741
|
|
float[batch,257,1024] add_2856
|
|
float[batch,257,1024] add_2877
|
|
float[batch,257,1024] add_293
|
|
float[batch,257,1024] add_2992
|
|
float[batch,257,1024] add_3013
|
|
float[batch,257,1024] add_3128
|
|
float[batch,257,1024] add_3149
|
|
float[batch,1,1024] add_3149_pooled
|
|
float[batch,1,1024] add_3264
|
|
float[batch,1,1024] add_3285
|
|
float[batch,257,1024] add_408
|
|
float[batch,257,1024] add_429
|
|
float[batch,257,1024] add_544
|
|
float[batch,257,1024] add_565
|
|
float[batch,257,1024] add_680
|
|
float[batch,257,1024] add_701
|
|
float[batch,257,1024] add_816
|
|
float[batch,257,1024] add_837
|
|
float[batch,257,1024] add_952
|
|
float[batch,257,1024] add_973
|
|
float[batch,1] clamp_min
|
|
float[batch,1024,16,16] conv2d
|
|
float[batch,257,4096] gelu
|
|
float[batch,257,4096] gelu_1
|
|
float[batch,257,4096] gelu_10
|
|
float[batch,257,4096] gelu_11
|
|
float[batch,257,4096] gelu_12
|
|
float[batch,257,4096] gelu_13
|
|
float[batch,257,4096] gelu_14
|
|
float[batch,257,4096] gelu_15
|
|
float[batch,257,4096] gelu_16
|
|
float[batch,257,4096] gelu_17
|
|
float[batch,257,4096] gelu_18
|
|
float[batch,257,4096] gelu_19
|
|
float[batch,257,4096] gelu_2
|
|
float[batch,257,4096] gelu_20
|
|
float[batch,257,4096] gelu_21
|
|
float[batch,257,4096] gelu_22
|
|
float[batch,1,4096] gelu_23
|
|
float[batch,257,4096] gelu_3
|
|
float[batch,257,4096] gelu_4
|
|
float[batch,257,4096] gelu_5
|
|
float[batch,257,4096] gelu_6
|
|
float[batch,257,4096] gelu_7
|
|
float[batch,257,4096] gelu_8
|
|
float[batch,257,4096] gelu_9
|
|
float[batch,3,224,224] image_chw
|
|
float[batch,224,224,3] image_f32
|
|
float[batch,257,1024] layer_norm
|
|
float[batch,257,1024] layer_norm_1
|
|
float[batch,257,1024] layer_norm_10
|
|
float[batch,257,1024] layer_norm_11
|
|
float[batch,257,1024] layer_norm_12
|
|
float[batch,257,1024] layer_norm_13
|
|
float[batch,257,1024] layer_norm_14
|
|
float[batch,257,1024] layer_norm_15
|
|
float[batch,257,1024] layer_norm_16
|
|
float[batch,257,1024] layer_norm_17
|
|
float[batch,257,1024] layer_norm_18
|
|
float[batch,257,1024] layer_norm_19
|
|
float[batch,257,1024] layer_norm_2
|
|
float[batch,257,1024] layer_norm_20
|
|
float[batch,257,1024] layer_norm_21
|
|
float[batch,257,1024] layer_norm_22
|
|
float[batch,257,1024] layer_norm_23
|
|
float[batch,257,1024] layer_norm_24
|
|
float[batch,257,1024] layer_norm_25
|
|
float[batch,257,1024] layer_norm_26
|
|
float[batch,257,1024] layer_norm_27
|
|
float[batch,257,1024] layer_norm_28
|
|
float[batch,257,1024] layer_norm_29
|
|
float[batch,257,1024] layer_norm_3
|
|
float[batch,257,1024] layer_norm_30
|
|
float[batch,257,1024] layer_norm_31
|
|
float[batch,257,1024] layer_norm_32
|
|
float[batch,257,1024] layer_norm_33
|
|
float[batch,257,1024] layer_norm_34
|
|
float[batch,257,1024] layer_norm_35
|
|
float[batch,257,1024] layer_norm_36
|
|
float[batch,257,1024] layer_norm_37
|
|
float[batch,257,1024] layer_norm_38
|
|
float[batch,257,1024] layer_norm_39
|
|
float[batch,257,1024] layer_norm_4
|
|
float[batch,257,1024] layer_norm_40
|
|
float[batch,257,1024] layer_norm_41
|
|
float[batch,257,1024] layer_norm_42
|
|
float[batch,257,1024] layer_norm_43
|
|
float[batch,257,1024] layer_norm_44
|
|
float[batch,257,1024] layer_norm_45
|
|
float[batch,257,1024] layer_norm_46
|
|
float[batch,257,1024] layer_norm_47
|
|
float[batch,1,1024] layer_norm_48
|
|
float[batch,257,1024] layer_norm_5
|
|
float[batch,257,1024] layer_norm_6
|
|
float[batch,257,1024] layer_norm_7
|
|
float[batch,257,1024] layer_norm_8
|
|
float[batch,257,1024] layer_norm_9
|
|
float[batch,1] linalg_vector_norm
|
|
float[batch,257,4096] linear_10
|
|
float[batch,257,1024] linear_11
|
|
float[batch,257,4096] linear_14
|
|
float[batch,257,1024] linear_15
|
|
float[batch,257,4096] linear_18
|
|
float[batch,257,1024] linear_19
|
|
float[batch,257,4096] linear_2
|
|
float[batch,257,4096] linear_22
|
|
float[batch,257,1024] linear_23
|
|
float[batch,257,4096] linear_26
|
|
float[batch,257,1024] linear_27
|
|
float[batch,257,1024] linear_3
|
|
float[batch,257,4096] linear_30
|
|
float[batch,257,1024] linear_31
|
|
float[batch,257,4096] linear_34
|
|
float[batch,257,1024] linear_35
|
|
float[batch,257,4096] linear_38
|
|
float[batch,257,1024] linear_39
|
|
float[batch,257,4096] linear_42
|
|
float[batch,257,1024] linear_43
|
|
float[batch,257,4096] linear_46
|
|
float[batch,257,1024] linear_47
|
|
float[batch,257,4096] linear_50
|
|
float[batch,257,1024] linear_51
|
|
float[batch,257,4096] linear_54
|
|
float[batch,257,1024] linear_55
|
|
float[batch,257,4096] linear_58
|
|
float[batch,257,1024] linear_59
|
|
float[batch,257,4096] linear_6
|
|
float[batch,257,4096] linear_62
|
|
float[batch,257,1024] linear_63
|
|
float[batch,257,4096] linear_66
|
|
float[batch,257,1024] linear_67
|
|
float[batch,257,1024] linear_7
|
|
float[batch,257,4096] linear_70
|
|
float[batch,257,1024] linear_71
|
|
float[batch,257,4096] linear_74
|
|
float[batch,257,1024] linear_75
|
|
float[batch,257,4096] linear_78
|
|
float[batch,257,1024] linear_79
|
|
float[batch,257,4096] linear_82
|
|
float[batch,257,1024] linear_83
|
|
float[batch,257,4096] linear_86
|
|
float[batch,257,1024] linear_87
|
|
float[batch,257,4096] linear_90
|
|
float[batch,257,1024] linear_91
|
|
float[batch,1,4096] linear_94
|
|
float[batch,1,1024] linear_95
|
|
float[batch,768] matmul
|
|
float[batch,257,1024] node_scaled_dot_product_attention_10_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_10_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_10_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_10_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_10_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_10_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_10_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_11_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_11_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_11_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_11_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_11_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_11_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_11_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_12_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_12_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_12_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_12_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_12_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_12_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_12_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_13_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_13_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_13_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_13_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_13_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_13_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_13_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_14_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_14_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_14_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_14_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_14_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_14_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_14_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_15_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_15_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_15_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_15_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_15_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_15_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_15_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_16_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_16_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_16_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_16_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_16_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_16_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_16_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_17_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_17_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_17_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_17_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_17_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_17_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_17_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_18_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_18_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_18_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_18_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_18_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_18_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_18_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_19_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_19_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_19_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_19_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_19_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_19_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_19_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_1_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_1_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_1_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_1_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_1_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_1_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_1_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_20_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_20_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_20_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_20_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_20_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_20_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_20_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_21_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_21_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_21_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_21_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_21_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_21_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_21_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_22_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_22_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_22_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_22_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_22_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_22_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_22_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_23_k
|
|
float[batch,1,1024] node_scaled_dot_product_attention_23_out
|
|
float[batch,1,1024] node_scaled_dot_product_attention_23_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_23_q
|
|
float[batch,1,1024] node_scaled_dot_product_attention_23_q_pooled
|
|
float[batch,257,3072] node_scaled_dot_product_attention_23_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_23_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_23_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_2_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_2_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_2_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_2_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_2_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_2_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_2_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_3_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_3_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_3_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_3_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_3_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_3_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_3_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_4_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_4_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_4_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_4_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_4_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_4_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_4_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_5_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_5_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_5_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_5_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_5_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_5_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_5_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_6_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_6_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_6_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_6_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_6_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_6_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_6_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_7_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_7_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_7_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_7_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_7_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_7_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_7_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_8_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_8_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_8_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_8_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_8_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_8_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_8_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_9_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_9_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_9_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_9_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_9_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_9_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_9_v
|
|
float[batch,257,1024] node_scaled_dot_product_attention_k
|
|
float[batch,257,1024] node_scaled_dot_product_attention_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_out_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_q
|
|
float[batch,257,3072] node_scaled_dot_product_attention_qkv
|
|
float[batch,257,3072] node_scaled_dot_product_attention_qkv_mm_out
|
|
float[batch,257,1024] node_scaled_dot_product_attention_v
|
|
float[batch,256,1024] permute
|
|
float[batch,257,1024] scaled_dot_product_attention
|
|
float[batch,257,1024] scaled_dot_product_attention_1
|
|
float[batch,257,1024] scaled_dot_product_attention_10
|
|
float[batch,257,1024] scaled_dot_product_attention_11
|
|
float[batch,257,1024] scaled_dot_product_attention_12
|
|
float[batch,257,1024] scaled_dot_product_attention_13
|
|
float[batch,257,1024] scaled_dot_product_attention_14
|
|
float[batch,257,1024] scaled_dot_product_attention_15
|
|
float[batch,257,1024] scaled_dot_product_attention_16
|
|
float[batch,257,1024] scaled_dot_product_attention_17
|
|
float[batch,257,1024] scaled_dot_product_attention_18
|
|
float[batch,257,1024] scaled_dot_product_attention_19
|
|
float[batch,257,1024] scaled_dot_product_attention_2
|
|
float[batch,257,1024] scaled_dot_product_attention_20
|
|
float[batch,257,1024] scaled_dot_product_attention_21
|
|
float[batch,257,1024] scaled_dot_product_attention_22
|
|
float[batch,1,1024] scaled_dot_product_attention_23
|
|
float[batch,257,1024] scaled_dot_product_attention_3
|
|
float[batch,257,1024] scaled_dot_product_attention_4
|
|
float[batch,257,1024] scaled_dot_product_attention_5
|
|
float[batch,257,1024] scaled_dot_product_attention_6
|
|
float[batch,257,1024] scaled_dot_product_attention_7
|
|
float[batch,257,1024] scaled_dot_product_attention_8
|
|
float[batch,257,1024] scaled_dot_product_attention_9
|
|
float[batch,1024] select_72
|
|
float[batch,257,4096] val_100
|
|
float[batch,257,1024] val_101
|
|
float[batch,257,4096] val_102
|
|
float[batch,257,1024] val_103
|
|
float[batch,257,4096] val_104
|
|
float[batch,257,1024] val_105
|
|
float[batch,257,4096] val_106
|
|
float[batch,257,1024] val_107
|
|
float[batch,257,4096] val_108
|
|
float[batch,257,1024] val_109
|
|
float[batch,257,4096] val_110
|
|
float[batch,257,1024] val_111
|
|
float[batch,257,4096] val_112
|
|
float[batch,257,1024] val_113
|
|
float[batch,257,4096] val_114
|
|
float[batch,257,1024] val_115
|
|
float[batch,257,4096] val_116
|
|
float[batch,257,1024] val_117
|
|
float[batch,257,4096] val_118
|
|
float[batch,257,1024] val_119
|
|
float[batch,257,4096] val_120
|
|
float[batch,257,1024] val_121
|
|
float[batch,257,4096] val_122
|
|
float[batch,257,1024] val_123
|
|
float[batch,257,4096] val_124
|
|
float[batch,257,1024] val_125
|
|
float[batch,1,4096] val_126
|
|
float[batch,1,1024] val_127
|
|
float[batch,1024] val_128
|
|
float[batch,257,1024] val_79
|
|
float[batch,257,4096] val_80
|
|
float[batch,257,1024] val_81
|
|
float[batch,257,4096] val_82
|
|
float[batch,257,1024] val_83
|
|
float[batch,257,4096] val_84
|
|
float[batch,257,1024] val_85
|
|
float[batch,257,4096] val_86
|
|
float[batch,257,1024] val_87
|
|
float[batch,257,4096] val_88
|
|
float[batch,257,1024] val_89
|
|
float[batch,257,4096] val_90
|
|
float[batch,257,1024] val_91
|
|
float[batch,257,4096] val_92
|
|
float[batch,257,1024] val_93
|
|
float[batch,257,4096] val_94
|
|
float[batch,257,1024] val_95
|
|
float[batch,257,4096] val_96
|
|
float[batch,257,1024] val_97
|
|
float[batch,257,4096] val_98
|
|
float[batch,257,1024] val_99
|
|
float[batch,1024,256] 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)
|
|
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.conv1.weight", node_Conv_1352_fused_bias)
|
|
view = Reshape <allowzero: int = 1> (conv2d, view_target)
|
|
[node_permute] permute = Transpose <perm: ints = [0, 2, 1]> (view)
|
|
val_79 = Pad (permute, val_3, val_4)
|
|
add_17 = Add (val_79, val_78)
|
|
[node_layer_norm] layer_norm = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_17, "visual.ln_pre.weight", "visual.ln_pre.bias")
|
|
layer_norm_1 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (layer_norm, "visual.transformer.resblocks.0.ln_1.weight", "visual.transformer.resblocks.0.ln_1.bias")
|
|
[node_scaled_dot_product_attention_qkv_mm] node_scaled_dot_product_attention_qkv_mm_out = MatMul (layer_norm_1, val_6)
|
|
[node_scaled_dot_product_attention_qkv_bias] node_scaled_dot_product_attention_qkv = Add (node_scaled_dot_product_attention_qkv_mm_out, "visual.transformer.resblocks.0.attn.in_proj_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> (node_scaled_dot_product_attention_qkv, attn3d_split_3x1024)
|
|
[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)
|
|
[node_scaled_dot_product_attention_out_mm] node_scaled_dot_product_attention_out_mm_out = MatMul (scaled_dot_product_attention, node_scaled_dot_product_attention_wo_t)
|
|
[node_scaled_dot_product_attention_out_bias] node_scaled_dot_product_attention_out = Add (node_scaled_dot_product_attention_out_mm_out, "visual.transformer.resblocks.0.attn.out_proj.bias")
|
|
add_136 = Add (layer_norm, node_scaled_dot_product_attention_out)
|
|
layer_norm_2 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_136, "visual.transformer.resblocks.0.ln_2.weight", "visual.transformer.resblocks.0.ln_2.bias")
|
|
val_80 = MatMul (layer_norm_2, val_7)
|
|
linear_2 = Add (val_80, "visual.transformer.resblocks.0.mlp.c_fc.bias")
|
|
[node_gelu] gelu = Gelu <approximate: string = "none"> (linear_2)
|
|
val_81 = MatMul (gelu, val_8)
|
|
linear_3 = Add (val_81, "visual.transformer.resblocks.0.mlp.c_proj.bias")
|
|
add_157 = Add (add_136, linear_3)
|
|
layer_norm_3 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_157, "visual.transformer.resblocks.1.ln_1.weight", "visual.transformer.resblocks.1.ln_1.bias")
|
|
[node_scaled_dot_product_attention_1_qkv_mm] node_scaled_dot_product_attention_1_qkv_mm_out = MatMul (layer_norm_3, val_9)
|
|
[node_scaled_dot_product_attention_1_qkv_bias] node_scaled_dot_product_attention_1_qkv = Add (node_scaled_dot_product_attention_1_qkv_mm_out, "visual.transformer.resblocks.1.attn.in_proj_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> (node_scaled_dot_product_attention_1_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_1_out_mm] node_scaled_dot_product_attention_1_out_mm_out = MatMul (scaled_dot_product_attention_1, node_scaled_dot_product_attention_1_wo_t)
|
|
[node_scaled_dot_product_attention_1_out_bias] node_scaled_dot_product_attention_1_out = Add (node_scaled_dot_product_attention_1_out_mm_out, "visual.transformer.resblocks.1.attn.out_proj.bias")
|
|
add_272 = Add (add_157, node_scaled_dot_product_attention_1_out)
|
|
layer_norm_4 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_272, "visual.transformer.resblocks.1.ln_2.weight", "visual.transformer.resblocks.1.ln_2.bias")
|
|
val_82 = MatMul (layer_norm_4, val_10)
|
|
linear_6 = Add (val_82, "visual.transformer.resblocks.1.mlp.c_fc.bias")
|
|
gelu_1 = Gelu <approximate: string = "none"> (linear_6)
|
|
val_83 = MatMul (gelu_1, val_11)
|
|
linear_7 = Add (val_83, "visual.transformer.resblocks.1.mlp.c_proj.bias")
|
|
add_293 = Add (add_272, linear_7)
|
|
layer_norm_5 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_293, "visual.transformer.resblocks.2.ln_1.weight", "visual.transformer.resblocks.2.ln_1.bias")
|
|
[node_scaled_dot_product_attention_2_qkv_mm] node_scaled_dot_product_attention_2_qkv_mm_out = MatMul (layer_norm_5, val_12)
|
|
[node_scaled_dot_product_attention_2_qkv_bias] node_scaled_dot_product_attention_2_qkv = Add (node_scaled_dot_product_attention_2_qkv_mm_out, "visual.transformer.resblocks.2.attn.in_proj_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> (node_scaled_dot_product_attention_2_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_2_out_mm] node_scaled_dot_product_attention_2_out_mm_out = MatMul (scaled_dot_product_attention_2, node_scaled_dot_product_attention_2_wo_t)
|
|
[node_scaled_dot_product_attention_2_out_bias] node_scaled_dot_product_attention_2_out = Add (node_scaled_dot_product_attention_2_out_mm_out, "visual.transformer.resblocks.2.attn.out_proj.bias")
|
|
add_408 = Add (add_293, node_scaled_dot_product_attention_2_out)
|
|
layer_norm_6 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_408, "visual.transformer.resblocks.2.ln_2.weight", "visual.transformer.resblocks.2.ln_2.bias")
|
|
val_84 = MatMul (layer_norm_6, val_13)
|
|
linear_10 = Add (val_84, "visual.transformer.resblocks.2.mlp.c_fc.bias")
|
|
gelu_2 = Gelu <approximate: string = "none"> (linear_10)
|
|
val_85 = MatMul (gelu_2, val_14)
|
|
linear_11 = Add (val_85, "visual.transformer.resblocks.2.mlp.c_proj.bias")
|
|
add_429 = Add (add_408, linear_11)
|
|
layer_norm_7 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_429, "visual.transformer.resblocks.3.ln_1.weight", "visual.transformer.resblocks.3.ln_1.bias")
|
|
[node_scaled_dot_product_attention_3_qkv_mm] node_scaled_dot_product_attention_3_qkv_mm_out = MatMul (layer_norm_7, val_15)
|
|
[node_scaled_dot_product_attention_3_qkv_bias] node_scaled_dot_product_attention_3_qkv = Add (node_scaled_dot_product_attention_3_qkv_mm_out, "visual.transformer.resblocks.3.attn.in_proj_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> (node_scaled_dot_product_attention_3_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_3_out_mm] node_scaled_dot_product_attention_3_out_mm_out = MatMul (scaled_dot_product_attention_3, node_scaled_dot_product_attention_3_wo_t)
|
|
[node_scaled_dot_product_attention_3_out_bias] node_scaled_dot_product_attention_3_out = Add (node_scaled_dot_product_attention_3_out_mm_out, "visual.transformer.resblocks.3.attn.out_proj.bias")
|
|
add_544 = Add (add_429, node_scaled_dot_product_attention_3_out)
|
|
layer_norm_8 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_544, "visual.transformer.resblocks.3.ln_2.weight", "visual.transformer.resblocks.3.ln_2.bias")
|
|
val_86 = MatMul (layer_norm_8, val_16)
|
|
linear_14 = Add (val_86, "visual.transformer.resblocks.3.mlp.c_fc.bias")
|
|
gelu_3 = Gelu <approximate: string = "none"> (linear_14)
|
|
val_87 = MatMul (gelu_3, val_17)
|
|
linear_15 = Add (val_87, "visual.transformer.resblocks.3.mlp.c_proj.bias")
|
|
add_565 = Add (add_544, linear_15)
|
|
layer_norm_9 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_565, "visual.transformer.resblocks.4.ln_1.weight", "visual.transformer.resblocks.4.ln_1.bias")
|
|
[node_scaled_dot_product_attention_4_qkv_mm] node_scaled_dot_product_attention_4_qkv_mm_out = MatMul (layer_norm_9, val_18)
|
|
[node_scaled_dot_product_attention_4_qkv_bias] node_scaled_dot_product_attention_4_qkv = Add (node_scaled_dot_product_attention_4_qkv_mm_out, "visual.transformer.resblocks.4.attn.in_proj_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> (node_scaled_dot_product_attention_4_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_4_out_mm] node_scaled_dot_product_attention_4_out_mm_out = MatMul (scaled_dot_product_attention_4, node_scaled_dot_product_attention_4_wo_t)
|
|
[node_scaled_dot_product_attention_4_out_bias] node_scaled_dot_product_attention_4_out = Add (node_scaled_dot_product_attention_4_out_mm_out, "visual.transformer.resblocks.4.attn.out_proj.bias")
|
|
add_680 = Add (add_565, node_scaled_dot_product_attention_4_out)
|
|
layer_norm_10 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_680, "visual.transformer.resblocks.4.ln_2.weight", "visual.transformer.resblocks.4.ln_2.bias")
|
|
val_88 = MatMul (layer_norm_10, val_19)
|
|
linear_18 = Add (val_88, "visual.transformer.resblocks.4.mlp.c_fc.bias")
|
|
gelu_4 = Gelu <approximate: string = "none"> (linear_18)
|
|
val_89 = MatMul (gelu_4, val_20)
|
|
linear_19 = Add (val_89, "visual.transformer.resblocks.4.mlp.c_proj.bias")
|
|
add_701 = Add (add_680, linear_19)
|
|
layer_norm_11 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_701, "visual.transformer.resblocks.5.ln_1.weight", "visual.transformer.resblocks.5.ln_1.bias")
|
|
[node_scaled_dot_product_attention_5_qkv_mm] node_scaled_dot_product_attention_5_qkv_mm_out = MatMul (layer_norm_11, val_21)
|
|
[node_scaled_dot_product_attention_5_qkv_bias] node_scaled_dot_product_attention_5_qkv = Add (node_scaled_dot_product_attention_5_qkv_mm_out, "visual.transformer.resblocks.5.attn.in_proj_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> (node_scaled_dot_product_attention_5_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_5_out_mm] node_scaled_dot_product_attention_5_out_mm_out = MatMul (scaled_dot_product_attention_5, node_scaled_dot_product_attention_5_wo_t)
|
|
[node_scaled_dot_product_attention_5_out_bias] node_scaled_dot_product_attention_5_out = Add (node_scaled_dot_product_attention_5_out_mm_out, "visual.transformer.resblocks.5.attn.out_proj.bias")
|
|
add_816 = Add (add_701, node_scaled_dot_product_attention_5_out)
|
|
layer_norm_12 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_816, "visual.transformer.resblocks.5.ln_2.weight", "visual.transformer.resblocks.5.ln_2.bias")
|
|
val_90 = MatMul (layer_norm_12, val_22)
|
|
linear_22 = Add (val_90, "visual.transformer.resblocks.5.mlp.c_fc.bias")
|
|
gelu_5 = Gelu <approximate: string = "none"> (linear_22)
|
|
val_91 = MatMul (gelu_5, val_23)
|
|
linear_23 = Add (val_91, "visual.transformer.resblocks.5.mlp.c_proj.bias")
|
|
add_837 = Add (add_816, linear_23)
|
|
layer_norm_13 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_837, "visual.transformer.resblocks.6.ln_1.weight", "visual.transformer.resblocks.6.ln_1.bias")
|
|
[node_scaled_dot_product_attention_6_qkv_mm] node_scaled_dot_product_attention_6_qkv_mm_out = MatMul (layer_norm_13, val_24)
|
|
[node_scaled_dot_product_attention_6_qkv_bias] node_scaled_dot_product_attention_6_qkv = Add (node_scaled_dot_product_attention_6_qkv_mm_out, "visual.transformer.resblocks.6.attn.in_proj_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> (node_scaled_dot_product_attention_6_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_6_out_mm] node_scaled_dot_product_attention_6_out_mm_out = MatMul (scaled_dot_product_attention_6, node_scaled_dot_product_attention_6_wo_t)
|
|
[node_scaled_dot_product_attention_6_out_bias] node_scaled_dot_product_attention_6_out = Add (node_scaled_dot_product_attention_6_out_mm_out, "visual.transformer.resblocks.6.attn.out_proj.bias")
|
|
add_952 = Add (add_837, node_scaled_dot_product_attention_6_out)
|
|
layer_norm_14 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_952, "visual.transformer.resblocks.6.ln_2.weight", "visual.transformer.resblocks.6.ln_2.bias")
|
|
val_92 = MatMul (layer_norm_14, val_25)
|
|
linear_26 = Add (val_92, "visual.transformer.resblocks.6.mlp.c_fc.bias")
|
|
gelu_6 = Gelu <approximate: string = "none"> (linear_26)
|
|
val_93 = MatMul (gelu_6, val_26)
|
|
linear_27 = Add (val_93, "visual.transformer.resblocks.6.mlp.c_proj.bias")
|
|
add_973 = Add (add_952, linear_27)
|
|
layer_norm_15 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_973, "visual.transformer.resblocks.7.ln_1.weight", "visual.transformer.resblocks.7.ln_1.bias")
|
|
[node_scaled_dot_product_attention_7_qkv_mm] node_scaled_dot_product_attention_7_qkv_mm_out = MatMul (layer_norm_15, val_27)
|
|
[node_scaled_dot_product_attention_7_qkv_bias] node_scaled_dot_product_attention_7_qkv = Add (node_scaled_dot_product_attention_7_qkv_mm_out, "visual.transformer.resblocks.7.attn.in_proj_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> (node_scaled_dot_product_attention_7_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_7_out_mm] node_scaled_dot_product_attention_7_out_mm_out = MatMul (scaled_dot_product_attention_7, node_scaled_dot_product_attention_7_wo_t)
|
|
[node_scaled_dot_product_attention_7_out_bias] node_scaled_dot_product_attention_7_out = Add (node_scaled_dot_product_attention_7_out_mm_out, "visual.transformer.resblocks.7.attn.out_proj.bias")
|
|
add_1088 = Add (add_973, node_scaled_dot_product_attention_7_out)
|
|
layer_norm_16 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1088, "visual.transformer.resblocks.7.ln_2.weight", "visual.transformer.resblocks.7.ln_2.bias")
|
|
val_94 = MatMul (layer_norm_16, val_28)
|
|
linear_30 = Add (val_94, "visual.transformer.resblocks.7.mlp.c_fc.bias")
|
|
gelu_7 = Gelu <approximate: string = "none"> (linear_30)
|
|
val_95 = MatMul (gelu_7, val_29)
|
|
linear_31 = Add (val_95, "visual.transformer.resblocks.7.mlp.c_proj.bias")
|
|
add_1109 = Add (add_1088, linear_31)
|
|
layer_norm_17 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1109, "visual.transformer.resblocks.8.ln_1.weight", "visual.transformer.resblocks.8.ln_1.bias")
|
|
[node_scaled_dot_product_attention_8_qkv_mm] node_scaled_dot_product_attention_8_qkv_mm_out = MatMul (layer_norm_17, val_30)
|
|
[node_scaled_dot_product_attention_8_qkv_bias] node_scaled_dot_product_attention_8_qkv = Add (node_scaled_dot_product_attention_8_qkv_mm_out, "visual.transformer.resblocks.8.attn.in_proj_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> (node_scaled_dot_product_attention_8_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_8_out_mm] node_scaled_dot_product_attention_8_out_mm_out = MatMul (scaled_dot_product_attention_8, node_scaled_dot_product_attention_8_wo_t)
|
|
[node_scaled_dot_product_attention_8_out_bias] node_scaled_dot_product_attention_8_out = Add (node_scaled_dot_product_attention_8_out_mm_out, "visual.transformer.resblocks.8.attn.out_proj.bias")
|
|
add_1224 = Add (add_1109, node_scaled_dot_product_attention_8_out)
|
|
layer_norm_18 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1224, "visual.transformer.resblocks.8.ln_2.weight", "visual.transformer.resblocks.8.ln_2.bias")
|
|
val_96 = MatMul (layer_norm_18, val_31)
|
|
linear_34 = Add (val_96, "visual.transformer.resblocks.8.mlp.c_fc.bias")
|
|
gelu_8 = Gelu <approximate: string = "none"> (linear_34)
|
|
val_97 = MatMul (gelu_8, val_32)
|
|
linear_35 = Add (val_97, "visual.transformer.resblocks.8.mlp.c_proj.bias")
|
|
add_1245 = Add (add_1224, linear_35)
|
|
layer_norm_19 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1245, "visual.transformer.resblocks.9.ln_1.weight", "visual.transformer.resblocks.9.ln_1.bias")
|
|
[node_scaled_dot_product_attention_9_qkv_mm] node_scaled_dot_product_attention_9_qkv_mm_out = MatMul (layer_norm_19, val_33)
|
|
[node_scaled_dot_product_attention_9_qkv_bias] node_scaled_dot_product_attention_9_qkv = Add (node_scaled_dot_product_attention_9_qkv_mm_out, "visual.transformer.resblocks.9.attn.in_proj_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> (node_scaled_dot_product_attention_9_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_9_out_mm] node_scaled_dot_product_attention_9_out_mm_out = MatMul (scaled_dot_product_attention_9, node_scaled_dot_product_attention_9_wo_t)
|
|
[node_scaled_dot_product_attention_9_out_bias] node_scaled_dot_product_attention_9_out = Add (node_scaled_dot_product_attention_9_out_mm_out, "visual.transformer.resblocks.9.attn.out_proj.bias")
|
|
add_1360 = Add (add_1245, node_scaled_dot_product_attention_9_out)
|
|
layer_norm_20 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1360, "visual.transformer.resblocks.9.ln_2.weight", "visual.transformer.resblocks.9.ln_2.bias")
|
|
val_98 = MatMul (layer_norm_20, val_34)
|
|
linear_38 = Add (val_98, "visual.transformer.resblocks.9.mlp.c_fc.bias")
|
|
gelu_9 = Gelu <approximate: string = "none"> (linear_38)
|
|
val_99 = MatMul (gelu_9, val_35)
|
|
linear_39 = Add (val_99, "visual.transformer.resblocks.9.mlp.c_proj.bias")
|
|
add_1381 = Add (add_1360, linear_39)
|
|
layer_norm_21 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1381, "visual.transformer.resblocks.10.ln_1.weight", "visual.transformer.resblocks.10.ln_1.bias")
|
|
[node_scaled_dot_product_attention_10_qkv_mm] node_scaled_dot_product_attention_10_qkv_mm_out = MatMul (layer_norm_21, val_36)
|
|
[node_scaled_dot_product_attention_10_qkv_bias] node_scaled_dot_product_attention_10_qkv = Add (node_scaled_dot_product_attention_10_qkv_mm_out, "visual.transformer.resblocks.10.attn.in_proj_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> (node_scaled_dot_product_attention_10_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_10_out_mm] node_scaled_dot_product_attention_10_out_mm_out = MatMul (scaled_dot_product_attention_10, node_scaled_dot_product_attention_10_wo_t)
|
|
[node_scaled_dot_product_attention_10_out_bias] node_scaled_dot_product_attention_10_out = Add (node_scaled_dot_product_attention_10_out_mm_out, "visual.transformer.resblocks.10.attn.out_proj.bias")
|
|
add_1496 = Add (add_1381, node_scaled_dot_product_attention_10_out)
|
|
layer_norm_22 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1496, "visual.transformer.resblocks.10.ln_2.weight", "visual.transformer.resblocks.10.ln_2.bias")
|
|
val_100 = MatMul (layer_norm_22, val_37)
|
|
linear_42 = Add (val_100, "visual.transformer.resblocks.10.mlp.c_fc.bias")
|
|
gelu_10 = Gelu <approximate: string = "none"> (linear_42)
|
|
val_101 = MatMul (gelu_10, val_38)
|
|
linear_43 = Add (val_101, "visual.transformer.resblocks.10.mlp.c_proj.bias")
|
|
add_1517 = Add (add_1496, linear_43)
|
|
layer_norm_23 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1517, "visual.transformer.resblocks.11.ln_1.weight", "visual.transformer.resblocks.11.ln_1.bias")
|
|
[node_scaled_dot_product_attention_11_qkv_mm] node_scaled_dot_product_attention_11_qkv_mm_out = MatMul (layer_norm_23, val_39)
|
|
[node_scaled_dot_product_attention_11_qkv_bias] node_scaled_dot_product_attention_11_qkv = Add (node_scaled_dot_product_attention_11_qkv_mm_out, "visual.transformer.resblocks.11.attn.in_proj_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> (node_scaled_dot_product_attention_11_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_11_out_mm] node_scaled_dot_product_attention_11_out_mm_out = MatMul (scaled_dot_product_attention_11, node_scaled_dot_product_attention_11_wo_t)
|
|
[node_scaled_dot_product_attention_11_out_bias] node_scaled_dot_product_attention_11_out = Add (node_scaled_dot_product_attention_11_out_mm_out, "visual.transformer.resblocks.11.attn.out_proj.bias")
|
|
add_1632 = Add (add_1517, node_scaled_dot_product_attention_11_out)
|
|
layer_norm_24 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1632, "visual.transformer.resblocks.11.ln_2.weight", "visual.transformer.resblocks.11.ln_2.bias")
|
|
val_102 = MatMul (layer_norm_24, val_40)
|
|
linear_46 = Add (val_102, "visual.transformer.resblocks.11.mlp.c_fc.bias")
|
|
gelu_11 = Gelu <approximate: string = "none"> (linear_46)
|
|
val_103 = MatMul (gelu_11, val_41)
|
|
linear_47 = Add (val_103, "visual.transformer.resblocks.11.mlp.c_proj.bias")
|
|
add_1653 = Add (add_1632, linear_47)
|
|
layer_norm_25 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1653, "visual.transformer.resblocks.12.ln_1.weight", "visual.transformer.resblocks.12.ln_1.bias")
|
|
[node_scaled_dot_product_attention_12_qkv_mm] node_scaled_dot_product_attention_12_qkv_mm_out = MatMul (layer_norm_25, val_42)
|
|
[node_scaled_dot_product_attention_12_qkv_bias] node_scaled_dot_product_attention_12_qkv = Add (node_scaled_dot_product_attention_12_qkv_mm_out, "visual.transformer.resblocks.12.attn.in_proj_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> (node_scaled_dot_product_attention_12_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_12_out_mm] node_scaled_dot_product_attention_12_out_mm_out = MatMul (scaled_dot_product_attention_12, node_scaled_dot_product_attention_12_wo_t)
|
|
[node_scaled_dot_product_attention_12_out_bias] node_scaled_dot_product_attention_12_out = Add (node_scaled_dot_product_attention_12_out_mm_out, "visual.transformer.resblocks.12.attn.out_proj.bias")
|
|
add_1768 = Add (add_1653, node_scaled_dot_product_attention_12_out)
|
|
layer_norm_26 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1768, "visual.transformer.resblocks.12.ln_2.weight", "visual.transformer.resblocks.12.ln_2.bias")
|
|
val_104 = MatMul (layer_norm_26, val_43)
|
|
linear_50 = Add (val_104, "visual.transformer.resblocks.12.mlp.c_fc.bias")
|
|
gelu_12 = Gelu <approximate: string = "none"> (linear_50)
|
|
val_105 = MatMul (gelu_12, val_44)
|
|
linear_51 = Add (val_105, "visual.transformer.resblocks.12.mlp.c_proj.bias")
|
|
add_1789 = Add (add_1768, linear_51)
|
|
layer_norm_27 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1789, "visual.transformer.resblocks.13.ln_1.weight", "visual.transformer.resblocks.13.ln_1.bias")
|
|
[node_scaled_dot_product_attention_13_qkv_mm] node_scaled_dot_product_attention_13_qkv_mm_out = MatMul (layer_norm_27, val_45)
|
|
[node_scaled_dot_product_attention_13_qkv_bias] node_scaled_dot_product_attention_13_qkv = Add (node_scaled_dot_product_attention_13_qkv_mm_out, "visual.transformer.resblocks.13.attn.in_proj_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> (node_scaled_dot_product_attention_13_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_13_out_mm] node_scaled_dot_product_attention_13_out_mm_out = MatMul (scaled_dot_product_attention_13, node_scaled_dot_product_attention_13_wo_t)
|
|
[node_scaled_dot_product_attention_13_out_bias] node_scaled_dot_product_attention_13_out = Add (node_scaled_dot_product_attention_13_out_mm_out, "visual.transformer.resblocks.13.attn.out_proj.bias")
|
|
add_1904 = Add (add_1789, node_scaled_dot_product_attention_13_out)
|
|
layer_norm_28 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1904, "visual.transformer.resblocks.13.ln_2.weight", "visual.transformer.resblocks.13.ln_2.bias")
|
|
val_106 = MatMul (layer_norm_28, val_46)
|
|
linear_54 = Add (val_106, "visual.transformer.resblocks.13.mlp.c_fc.bias")
|
|
gelu_13 = Gelu <approximate: string = "none"> (linear_54)
|
|
val_107 = MatMul (gelu_13, val_47)
|
|
linear_55 = Add (val_107, "visual.transformer.resblocks.13.mlp.c_proj.bias")
|
|
add_1925 = Add (add_1904, linear_55)
|
|
layer_norm_29 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_1925, "visual.transformer.resblocks.14.ln_1.weight", "visual.transformer.resblocks.14.ln_1.bias")
|
|
[node_scaled_dot_product_attention_14_qkv_mm] node_scaled_dot_product_attention_14_qkv_mm_out = MatMul (layer_norm_29, val_48)
|
|
[node_scaled_dot_product_attention_14_qkv_bias] node_scaled_dot_product_attention_14_qkv = Add (node_scaled_dot_product_attention_14_qkv_mm_out, "visual.transformer.resblocks.14.attn.in_proj_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> (node_scaled_dot_product_attention_14_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_14_out_mm] node_scaled_dot_product_attention_14_out_mm_out = MatMul (scaled_dot_product_attention_14, node_scaled_dot_product_attention_14_wo_t)
|
|
[node_scaled_dot_product_attention_14_out_bias] node_scaled_dot_product_attention_14_out = Add (node_scaled_dot_product_attention_14_out_mm_out, "visual.transformer.resblocks.14.attn.out_proj.bias")
|
|
add_2040 = Add (add_1925, node_scaled_dot_product_attention_14_out)
|
|
layer_norm_30 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2040, "visual.transformer.resblocks.14.ln_2.weight", "visual.transformer.resblocks.14.ln_2.bias")
|
|
val_108 = MatMul (layer_norm_30, val_49)
|
|
linear_58 = Add (val_108, "visual.transformer.resblocks.14.mlp.c_fc.bias")
|
|
gelu_14 = Gelu <approximate: string = "none"> (linear_58)
|
|
val_109 = MatMul (gelu_14, val_50)
|
|
linear_59 = Add (val_109, "visual.transformer.resblocks.14.mlp.c_proj.bias")
|
|
add_2061 = Add (add_2040, linear_59)
|
|
layer_norm_31 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2061, "visual.transformer.resblocks.15.ln_1.weight", "visual.transformer.resblocks.15.ln_1.bias")
|
|
[node_scaled_dot_product_attention_15_qkv_mm] node_scaled_dot_product_attention_15_qkv_mm_out = MatMul (layer_norm_31, val_51)
|
|
[node_scaled_dot_product_attention_15_qkv_bias] node_scaled_dot_product_attention_15_qkv = Add (node_scaled_dot_product_attention_15_qkv_mm_out, "visual.transformer.resblocks.15.attn.in_proj_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> (node_scaled_dot_product_attention_15_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_15_out_mm] node_scaled_dot_product_attention_15_out_mm_out = MatMul (scaled_dot_product_attention_15, node_scaled_dot_product_attention_15_wo_t)
|
|
[node_scaled_dot_product_attention_15_out_bias] node_scaled_dot_product_attention_15_out = Add (node_scaled_dot_product_attention_15_out_mm_out, "visual.transformer.resblocks.15.attn.out_proj.bias")
|
|
add_2176 = Add (add_2061, node_scaled_dot_product_attention_15_out)
|
|
layer_norm_32 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2176, "visual.transformer.resblocks.15.ln_2.weight", "visual.transformer.resblocks.15.ln_2.bias")
|
|
val_110 = MatMul (layer_norm_32, val_52)
|
|
linear_62 = Add (val_110, "visual.transformer.resblocks.15.mlp.c_fc.bias")
|
|
gelu_15 = Gelu <approximate: string = "none"> (linear_62)
|
|
val_111 = MatMul (gelu_15, val_53)
|
|
linear_63 = Add (val_111, "visual.transformer.resblocks.15.mlp.c_proj.bias")
|
|
add_2197 = Add (add_2176, linear_63)
|
|
layer_norm_33 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2197, "visual.transformer.resblocks.16.ln_1.weight", "visual.transformer.resblocks.16.ln_1.bias")
|
|
[node_scaled_dot_product_attention_16_qkv_mm] node_scaled_dot_product_attention_16_qkv_mm_out = MatMul (layer_norm_33, val_54)
|
|
[node_scaled_dot_product_attention_16_qkv_bias] node_scaled_dot_product_attention_16_qkv = Add (node_scaled_dot_product_attention_16_qkv_mm_out, "visual.transformer.resblocks.16.attn.in_proj_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> (node_scaled_dot_product_attention_16_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_16_out_mm] node_scaled_dot_product_attention_16_out_mm_out = MatMul (scaled_dot_product_attention_16, node_scaled_dot_product_attention_16_wo_t)
|
|
[node_scaled_dot_product_attention_16_out_bias] node_scaled_dot_product_attention_16_out = Add (node_scaled_dot_product_attention_16_out_mm_out, "visual.transformer.resblocks.16.attn.out_proj.bias")
|
|
add_2312 = Add (add_2197, node_scaled_dot_product_attention_16_out)
|
|
layer_norm_34 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2312, "visual.transformer.resblocks.16.ln_2.weight", "visual.transformer.resblocks.16.ln_2.bias")
|
|
val_112 = MatMul (layer_norm_34, val_55)
|
|
linear_66 = Add (val_112, "visual.transformer.resblocks.16.mlp.c_fc.bias")
|
|
gelu_16 = Gelu <approximate: string = "none"> (linear_66)
|
|
val_113 = MatMul (gelu_16, val_56)
|
|
linear_67 = Add (val_113, "visual.transformer.resblocks.16.mlp.c_proj.bias")
|
|
add_2333 = Add (add_2312, linear_67)
|
|
layer_norm_35 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2333, "visual.transformer.resblocks.17.ln_1.weight", "visual.transformer.resblocks.17.ln_1.bias")
|
|
[node_scaled_dot_product_attention_17_qkv_mm] node_scaled_dot_product_attention_17_qkv_mm_out = MatMul (layer_norm_35, val_57)
|
|
[node_scaled_dot_product_attention_17_qkv_bias] node_scaled_dot_product_attention_17_qkv = Add (node_scaled_dot_product_attention_17_qkv_mm_out, "visual.transformer.resblocks.17.attn.in_proj_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> (node_scaled_dot_product_attention_17_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_17_out_mm] node_scaled_dot_product_attention_17_out_mm_out = MatMul (scaled_dot_product_attention_17, node_scaled_dot_product_attention_17_wo_t)
|
|
[node_scaled_dot_product_attention_17_out_bias] node_scaled_dot_product_attention_17_out = Add (node_scaled_dot_product_attention_17_out_mm_out, "visual.transformer.resblocks.17.attn.out_proj.bias")
|
|
add_2448 = Add (add_2333, node_scaled_dot_product_attention_17_out)
|
|
layer_norm_36 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2448, "visual.transformer.resblocks.17.ln_2.weight", "visual.transformer.resblocks.17.ln_2.bias")
|
|
val_114 = MatMul (layer_norm_36, val_58)
|
|
linear_70 = Add (val_114, "visual.transformer.resblocks.17.mlp.c_fc.bias")
|
|
gelu_17 = Gelu <approximate: string = "none"> (linear_70)
|
|
val_115 = MatMul (gelu_17, val_59)
|
|
linear_71 = Add (val_115, "visual.transformer.resblocks.17.mlp.c_proj.bias")
|
|
add_2469 = Add (add_2448, linear_71)
|
|
layer_norm_37 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2469, "visual.transformer.resblocks.18.ln_1.weight", "visual.transformer.resblocks.18.ln_1.bias")
|
|
[node_scaled_dot_product_attention_18_qkv_mm] node_scaled_dot_product_attention_18_qkv_mm_out = MatMul (layer_norm_37, val_60)
|
|
[node_scaled_dot_product_attention_18_qkv_bias] node_scaled_dot_product_attention_18_qkv = Add (node_scaled_dot_product_attention_18_qkv_mm_out, "visual.transformer.resblocks.18.attn.in_proj_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> (node_scaled_dot_product_attention_18_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_18_out_mm] node_scaled_dot_product_attention_18_out_mm_out = MatMul (scaled_dot_product_attention_18, node_scaled_dot_product_attention_18_wo_t)
|
|
[node_scaled_dot_product_attention_18_out_bias] node_scaled_dot_product_attention_18_out = Add (node_scaled_dot_product_attention_18_out_mm_out, "visual.transformer.resblocks.18.attn.out_proj.bias")
|
|
add_2584 = Add (add_2469, node_scaled_dot_product_attention_18_out)
|
|
layer_norm_38 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2584, "visual.transformer.resblocks.18.ln_2.weight", "visual.transformer.resblocks.18.ln_2.bias")
|
|
val_116 = MatMul (layer_norm_38, val_61)
|
|
linear_74 = Add (val_116, "visual.transformer.resblocks.18.mlp.c_fc.bias")
|
|
gelu_18 = Gelu <approximate: string = "none"> (linear_74)
|
|
val_117 = MatMul (gelu_18, val_62)
|
|
linear_75 = Add (val_117, "visual.transformer.resblocks.18.mlp.c_proj.bias")
|
|
add_2605 = Add (add_2584, linear_75)
|
|
layer_norm_39 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2605, "visual.transformer.resblocks.19.ln_1.weight", "visual.transformer.resblocks.19.ln_1.bias")
|
|
[node_scaled_dot_product_attention_19_qkv_mm] node_scaled_dot_product_attention_19_qkv_mm_out = MatMul (layer_norm_39, val_63)
|
|
[node_scaled_dot_product_attention_19_qkv_bias] node_scaled_dot_product_attention_19_qkv = Add (node_scaled_dot_product_attention_19_qkv_mm_out, "visual.transformer.resblocks.19.attn.in_proj_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> (node_scaled_dot_product_attention_19_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_19_out_mm] node_scaled_dot_product_attention_19_out_mm_out = MatMul (scaled_dot_product_attention_19, node_scaled_dot_product_attention_19_wo_t)
|
|
[node_scaled_dot_product_attention_19_out_bias] node_scaled_dot_product_attention_19_out = Add (node_scaled_dot_product_attention_19_out_mm_out, "visual.transformer.resblocks.19.attn.out_proj.bias")
|
|
add_2720 = Add (add_2605, node_scaled_dot_product_attention_19_out)
|
|
layer_norm_40 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2720, "visual.transformer.resblocks.19.ln_2.weight", "visual.transformer.resblocks.19.ln_2.bias")
|
|
val_118 = MatMul (layer_norm_40, val_64)
|
|
linear_78 = Add (val_118, "visual.transformer.resblocks.19.mlp.c_fc.bias")
|
|
gelu_19 = Gelu <approximate: string = "none"> (linear_78)
|
|
val_119 = MatMul (gelu_19, val_65)
|
|
linear_79 = Add (val_119, "visual.transformer.resblocks.19.mlp.c_proj.bias")
|
|
add_2741 = Add (add_2720, linear_79)
|
|
layer_norm_41 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2741, "visual.transformer.resblocks.20.ln_1.weight", "visual.transformer.resblocks.20.ln_1.bias")
|
|
[node_scaled_dot_product_attention_20_qkv_mm] node_scaled_dot_product_attention_20_qkv_mm_out = MatMul (layer_norm_41, val_66)
|
|
[node_scaled_dot_product_attention_20_qkv_bias] node_scaled_dot_product_attention_20_qkv = Add (node_scaled_dot_product_attention_20_qkv_mm_out, "visual.transformer.resblocks.20.attn.in_proj_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> (node_scaled_dot_product_attention_20_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_20_out_mm] node_scaled_dot_product_attention_20_out_mm_out = MatMul (scaled_dot_product_attention_20, node_scaled_dot_product_attention_20_wo_t)
|
|
[node_scaled_dot_product_attention_20_out_bias] node_scaled_dot_product_attention_20_out = Add (node_scaled_dot_product_attention_20_out_mm_out, "visual.transformer.resblocks.20.attn.out_proj.bias")
|
|
add_2856 = Add (add_2741, node_scaled_dot_product_attention_20_out)
|
|
layer_norm_42 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2856, "visual.transformer.resblocks.20.ln_2.weight", "visual.transformer.resblocks.20.ln_2.bias")
|
|
val_120 = MatMul (layer_norm_42, val_67)
|
|
linear_82 = Add (val_120, "visual.transformer.resblocks.20.mlp.c_fc.bias")
|
|
gelu_20 = Gelu <approximate: string = "none"> (linear_82)
|
|
val_121 = MatMul (gelu_20, val_68)
|
|
linear_83 = Add (val_121, "visual.transformer.resblocks.20.mlp.c_proj.bias")
|
|
add_2877 = Add (add_2856, linear_83)
|
|
layer_norm_43 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2877, "visual.transformer.resblocks.21.ln_1.weight", "visual.transformer.resblocks.21.ln_1.bias")
|
|
[node_scaled_dot_product_attention_21_qkv_mm] node_scaled_dot_product_attention_21_qkv_mm_out = MatMul (layer_norm_43, val_69)
|
|
[node_scaled_dot_product_attention_21_qkv_bias] node_scaled_dot_product_attention_21_qkv = Add (node_scaled_dot_product_attention_21_qkv_mm_out, "visual.transformer.resblocks.21.attn.in_proj_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> (node_scaled_dot_product_attention_21_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_21_out_mm] node_scaled_dot_product_attention_21_out_mm_out = MatMul (scaled_dot_product_attention_21, node_scaled_dot_product_attention_21_wo_t)
|
|
[node_scaled_dot_product_attention_21_out_bias] node_scaled_dot_product_attention_21_out = Add (node_scaled_dot_product_attention_21_out_mm_out, "visual.transformer.resblocks.21.attn.out_proj.bias")
|
|
add_2992 = Add (add_2877, node_scaled_dot_product_attention_21_out)
|
|
layer_norm_44 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_2992, "visual.transformer.resblocks.21.ln_2.weight", "visual.transformer.resblocks.21.ln_2.bias")
|
|
val_122 = MatMul (layer_norm_44, val_70)
|
|
linear_86 = Add (val_122, "visual.transformer.resblocks.21.mlp.c_fc.bias")
|
|
gelu_21 = Gelu <approximate: string = "none"> (linear_86)
|
|
val_123 = MatMul (gelu_21, val_71)
|
|
linear_87 = Add (val_123, "visual.transformer.resblocks.21.mlp.c_proj.bias")
|
|
add_3013 = Add (add_2992, linear_87)
|
|
layer_norm_45 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_3013, "visual.transformer.resblocks.22.ln_1.weight", "visual.transformer.resblocks.22.ln_1.bias")
|
|
[node_scaled_dot_product_attention_22_qkv_mm] node_scaled_dot_product_attention_22_qkv_mm_out = MatMul (layer_norm_45, val_72)
|
|
[node_scaled_dot_product_attention_22_qkv_bias] node_scaled_dot_product_attention_22_qkv = Add (node_scaled_dot_product_attention_22_qkv_mm_out, "visual.transformer.resblocks.22.attn.in_proj_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> (node_scaled_dot_product_attention_22_qkv, attn3d_split_3x1024)
|
|
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)
|
|
[node_scaled_dot_product_attention_22_out_mm] node_scaled_dot_product_attention_22_out_mm_out = MatMul (scaled_dot_product_attention_22, node_scaled_dot_product_attention_22_wo_t)
|
|
[node_scaled_dot_product_attention_22_out_bias] node_scaled_dot_product_attention_22_out = Add (node_scaled_dot_product_attention_22_out_mm_out, "visual.transformer.resblocks.22.attn.out_proj.bias")
|
|
add_3128 = Add (add_3013, node_scaled_dot_product_attention_22_out)
|
|
layer_norm_46 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_3128, "visual.transformer.resblocks.22.ln_2.weight", "visual.transformer.resblocks.22.ln_2.bias")
|
|
val_124 = MatMul (layer_norm_46, val_73)
|
|
linear_90 = Add (val_124, "visual.transformer.resblocks.22.mlp.c_fc.bias")
|
|
gelu_22 = Gelu <approximate: string = "none"> (linear_90)
|
|
val_125 = MatMul (gelu_22, val_74)
|
|
linear_91 = Add (val_125, "visual.transformer.resblocks.22.mlp.c_proj.bias")
|
|
add_3149 = Add (add_3128, linear_91)
|
|
layer_norm_47 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_3149, "visual.transformer.resblocks.23.ln_1.weight", "visual.transformer.resblocks.23.ln_1.bias")
|
|
[node_scaled_dot_product_attention_23_qkv_mm] node_scaled_dot_product_attention_23_qkv_mm_out = MatMul (layer_norm_47, val_75)
|
|
[node_scaled_dot_product_attention_23_qkv_bias] node_scaled_dot_product_attention_23_qkv = Add (node_scaled_dot_product_attention_23_qkv_mm_out, "visual.transformer.resblocks.23.attn.in_proj_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> (node_scaled_dot_product_attention_23_qkv, attn3d_split_3x1024)
|
|
[pool_hoist_node_scaled_dot_product_attention_23_q] node_scaled_dot_product_attention_23_q_pooled = Slice (node_scaled_dot_product_attention_23_q, val_2, val_5, val_5)
|
|
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_pooled, node_scaled_dot_product_attention_23_k, node_scaled_dot_product_attention_23_v)
|
|
[node_scaled_dot_product_attention_23_out_mm] node_scaled_dot_product_attention_23_out_mm_out = MatMul (scaled_dot_product_attention_23, node_scaled_dot_product_attention_23_wo_t)
|
|
[node_scaled_dot_product_attention_23_out_bias] node_scaled_dot_product_attention_23_out = Add (node_scaled_dot_product_attention_23_out_mm_out, "visual.transformer.resblocks.23.attn.out_proj.bias")
|
|
[pool_hoist_add_3149] add_3149_pooled = Slice (add_3149, val_2, val_5, val_5)
|
|
add_3264 = Add (add_3149_pooled, node_scaled_dot_product_attention_23_out)
|
|
layer_norm_48 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (add_3264, "visual.transformer.resblocks.23.ln_2.weight", "visual.transformer.resblocks.23.ln_2.bias")
|
|
val_126 = MatMul (layer_norm_48, val_76)
|
|
linear_94 = Add (val_126, "visual.transformer.resblocks.23.mlp.c_fc.bias")
|
|
gelu_23 = Gelu <approximate: string = "none"> (linear_94)
|
|
val_127 = MatMul (gelu_23, val_77)
|
|
linear_95 = Add (val_127, "visual.transformer.resblocks.23.mlp.c_proj.bias")
|
|
add_3285 = Add (add_3264, linear_95)
|
|
val_128 = Squeeze (add_3285, val_5)
|
|
select_72 = LayerNormalization <axis: int = -1, epsilon: float = 1e-05, stash_type: int = 1> (val_128, "visual.ln_post.weight", "visual.ln_post.bias")
|
|
[node_matmul] matmul = MatMul (select_72, "visual.proj")
|
|
[node_linalg_vector_norm] linalg_vector_norm = ReduceL2 <keepdims: int = 1, noop_with_empty_axes: int = 0> (matmul, val_0)
|
|
[node_clamp_min] clamp_min = Clip (linalg_vector_norm, val_1)
|
|
[node_div] image_embedding = Div (matmul, clamp_min)
|
|
}
|
|
|
|
weights:
|
|
attn3d_split_3x1024 INT64[3] 0ec6f5651fd5
|
|
node_Conv_1352_fused_bias FLOAT[1024] c0b822893311
|
|
node_scaled_dot_product_attention_10_wo_t FLOAT[1024,1024] 69fe12f4883b
|
|
node_scaled_dot_product_attention_11_wo_t FLOAT[1024,1024] 8fcd635ca6eb
|
|
node_scaled_dot_product_attention_12_wo_t FLOAT[1024,1024] 3f15b9442bc0
|
|
node_scaled_dot_product_attention_13_wo_t FLOAT[1024,1024] c79b955d775f
|
|
node_scaled_dot_product_attention_14_wo_t FLOAT[1024,1024] d659d935f884
|
|
node_scaled_dot_product_attention_15_wo_t FLOAT[1024,1024] 07699dca6796
|
|
node_scaled_dot_product_attention_16_wo_t FLOAT[1024,1024] 71918a2654e8
|
|
node_scaled_dot_product_attention_17_wo_t FLOAT[1024,1024] f93e5b70122d
|
|
node_scaled_dot_product_attention_18_wo_t FLOAT[1024,1024] a05c5fd4221e
|
|
node_scaled_dot_product_attention_19_wo_t FLOAT[1024,1024] 91f1e2786e20
|
|
node_scaled_dot_product_attention_1_wo_t FLOAT[1024,1024] 0845bff57995
|
|
node_scaled_dot_product_attention_20_wo_t FLOAT[1024,1024] 8b238c73bb3b
|
|
node_scaled_dot_product_attention_21_wo_t FLOAT[1024,1024] 6e817f172dd9
|
|
node_scaled_dot_product_attention_22_wo_t FLOAT[1024,1024] cfc23515e85c
|
|
node_scaled_dot_product_attention_23_wo_t FLOAT[1024,1024] fc39de82b1a4
|
|
node_scaled_dot_product_attention_2_wo_t FLOAT[1024,1024] 2fea2f158cc4
|
|
node_scaled_dot_product_attention_3_wo_t FLOAT[1024,1024] c5e2af8dec64
|
|
node_scaled_dot_product_attention_4_wo_t FLOAT[1024,1024] e0002d8c9d31
|
|
node_scaled_dot_product_attention_5_wo_t FLOAT[1024,1024] 3bb1448eaa0e
|
|
node_scaled_dot_product_attention_6_wo_t FLOAT[1024,1024] 8fa2e12b0c67
|
|
node_scaled_dot_product_attention_7_wo_t FLOAT[1024,1024] 9afa0f9d86e7
|
|
node_scaled_dot_product_attention_8_wo_t FLOAT[1024,1024] 3b025d442c0b
|
|
node_scaled_dot_product_attention_9_wo_t FLOAT[1024,1024] 55e5f063452a
|
|
node_scaled_dot_product_attention_wo_t FLOAT[1024,1024] 551b3843a295
|
|
val_0 INT64[1] 12a3ae445661
|
|
val_1 FLOAT[] 6708d9be4956
|
|
val_10 FLOAT[1024,4096] 81a8b912dda4
|
|
val_11 FLOAT[4096,1024] c53a201ac1f8
|
|
val_12 FLOAT[1024,3072] e3661e251335
|
|
val_13 FLOAT[1024,4096] 2bb727710579
|
|
val_14 FLOAT[4096,1024] 3d8bbef979ed
|
|
val_15 FLOAT[1024,3072] c217671fcb27
|
|
val_16 FLOAT[1024,4096] 9fc07c1565cd
|
|
val_17 FLOAT[4096,1024] 77acb21b9920
|
|
val_18 FLOAT[1024,3072] 159e1656295a
|
|
val_19 FLOAT[1024,4096] 62f4db0ed101
|
|
val_2 INT64[1] af5570f5a181
|
|
val_20 FLOAT[4096,1024] e4114c319539
|
|
val_21 FLOAT[1024,3072] 793de010382b
|
|
val_22 FLOAT[1024,4096] 7e4f10bccf0a
|
|
val_23 FLOAT[4096,1024] 505ed9d5b1e3
|
|
val_24 FLOAT[1024,3072] 8efb226d5014
|
|
val_25 FLOAT[1024,4096] d7a180428e62
|
|
val_26 FLOAT[4096,1024] b6b4cc684e87
|
|
val_27 FLOAT[1024,3072] 9521b8b31d41
|
|
val_28 FLOAT[1024,4096] fc3190ce0c44
|
|
val_29 FLOAT[4096,1024] fec717e0d15b
|
|
val_3 INT64[6] 6b7d92eaae70
|
|
val_30 FLOAT[1024,3072] f364a4656696
|
|
val_31 FLOAT[1024,4096] e7a16513b3a7
|
|
val_32 FLOAT[4096,1024] 272b22c9a5ce
|
|
val_33 FLOAT[1024,3072] fc54c5442628
|
|
val_34 FLOAT[1024,4096] 8c1271f9373e
|
|
val_35 FLOAT[4096,1024] cb33d3d665b6
|
|
val_36 FLOAT[1024,3072] 4f9e8f196c6f
|
|
val_37 FLOAT[1024,4096] cabd2692cda8
|
|
val_38 FLOAT[4096,1024] ccdf727765e6
|
|
val_39 FLOAT[1024,3072] de3a2d981f01
|
|
val_4 FLOAT[] df3f619804a9
|
|
val_40 FLOAT[1024,4096] ba74fa89a7d9
|
|
val_41 FLOAT[4096,1024] b61d6ccb6ae0
|
|
val_42 FLOAT[1024,3072] d4b4373c14b0
|
|
val_43 FLOAT[1024,4096] 9c5913871102
|
|
val_44 FLOAT[4096,1024] e5eaafa68c4c
|
|
val_45 FLOAT[1024,3072] 0d3fa51df10e
|
|
val_46 FLOAT[1024,4096] d1e94d963dca
|
|
val_47 FLOAT[4096,1024] 179565dbc81a
|
|
val_48 FLOAT[1024,3072] 77caa39b1df3
|
|
val_49 FLOAT[1024,4096] e8d280052726
|
|
val_5 INT64[1] 7c9fa136d441
|
|
val_50 FLOAT[4096,1024] b013e14216f2
|
|
val_51 FLOAT[1024,3072] effdb43e8c90
|
|
val_52 FLOAT[1024,4096] 50e8cc0518f3
|
|
val_53 FLOAT[4096,1024] 1607033c74d4
|
|
val_54 FLOAT[1024,3072] 95c3fa955efa
|
|
val_55 FLOAT[1024,4096] 06ce3c57c78a
|
|
val_56 FLOAT[4096,1024] ad19097b5750
|
|
val_57 FLOAT[1024,3072] 9d0d805b22d7
|
|
val_58 FLOAT[1024,4096] 35f194076dc5
|
|
val_59 FLOAT[4096,1024] b814f21812f6
|
|
val_6 FLOAT[1024,3072] 8401daeabd3b
|
|
val_60 FLOAT[1024,3072] 78d7bb539981
|
|
val_61 FLOAT[1024,4096] f17fd25c3c98
|
|
val_62 FLOAT[4096,1024] 9c720a108ccc
|
|
val_63 FLOAT[1024,3072] f05bcf52607b
|
|
val_64 FLOAT[1024,4096] 4cb2f35edd5a
|
|
val_65 FLOAT[4096,1024] 4d6e3a7985da
|
|
val_66 FLOAT[1024,3072] a1ba81dfa3e7
|
|
val_67 FLOAT[1024,4096] 0a68989ddf8b
|
|
val_68 FLOAT[4096,1024] e96f7853207f
|
|
val_69 FLOAT[1024,3072] 0961c9258c1a
|
|
val_7 FLOAT[1024,4096] ce70420cfc58
|
|
val_70 FLOAT[1024,4096] 52324ec22f50
|
|
val_71 FLOAT[4096,1024] caf3726539b0
|
|
val_72 FLOAT[1024,3072] c78ff4063b46
|
|
val_73 FLOAT[1024,4096] 1b6629a1e4fa
|
|
val_74 FLOAT[4096,1024] 19b9b4a51603
|
|
val_75 FLOAT[1024,3072] db9f52355a10
|
|
val_76 FLOAT[1024,4096] 1a2cdc34a1bc
|
|
val_77 FLOAT[4096,1024] f21d94f0c83d
|
|
val_78 FLOAT[1,257,1024] 8049132696a9
|
|
val_8 FLOAT[4096,1024] 7e52d520f65b
|
|
val_9 FLOAT[1024,3072] 654e5cd12883
|
|
view_target INT64[3] 3f85cfe8397f
|
|
visual.conv1.weight FLOAT[1024,3,14,14] b98cf2247624
|
|
visual.ln_post.bias FLOAT[1024] 194f59948fad
|
|
visual.ln_post.weight FLOAT[1024] f903ed094aa2
|
|
visual.ln_pre.bias FLOAT[1024] 018fdd8e0b7d
|
|
visual.ln_pre.weight FLOAT[1024] e001e01c191c
|
|
visual.proj FLOAT[1024,768] 5679fb6b8765
|
|
visual.transformer.resblocks.0.attn.in_proj_bias FLOAT[3072] 3c460cb76b29
|
|
visual.transformer.resblocks.0.attn.out_proj.bias FLOAT[1024] 5af659f5b98c
|
|
visual.transformer.resblocks.0.ln_1.bias FLOAT[1024] b836323efaff
|
|
visual.transformer.resblocks.0.ln_1.weight FLOAT[1024] 5b5ead53e5ef
|
|
visual.transformer.resblocks.0.ln_2.bias FLOAT[1024] 0fd2b0c99bde
|
|
visual.transformer.resblocks.0.ln_2.weight FLOAT[1024] be6addb29060
|
|
visual.transformer.resblocks.0.mlp.c_fc.bias FLOAT[4096] 3ee7d8838462
|
|
visual.transformer.resblocks.0.mlp.c_proj.bias FLOAT[1024] 29891e44936c
|
|
visual.transformer.resblocks.1.attn.in_proj_bias FLOAT[3072] 3bebce01d7a3
|
|
visual.transformer.resblocks.1.attn.out_proj.bias FLOAT[1024] 802d730fed78
|
|
visual.transformer.resblocks.1.ln_1.bias FLOAT[1024] dfcdcd7fae21
|
|
visual.transformer.resblocks.1.ln_1.weight FLOAT[1024] 9401861f6d8c
|
|
visual.transformer.resblocks.1.ln_2.bias FLOAT[1024] cdce2060fde1
|
|
visual.transformer.resblocks.1.ln_2.weight FLOAT[1024] 2c1f5a4e56a7
|
|
visual.transformer.resblocks.1.mlp.c_fc.bias FLOAT[4096] 9b31e6c03b02
|
|
visual.transformer.resblocks.1.mlp.c_proj.bias FLOAT[1024] 546beea64203
|
|
visual.transformer.resblocks.10.attn.in_proj_bias FLOAT[3072] 6cce0513adfa
|
|
visual.transformer.resblocks.10.attn.out_proj.bias FLOAT[1024] 219f1eef3e74
|
|
visual.transformer.resblocks.10.ln_1.bias FLOAT[1024] 94358230dffb
|
|
visual.transformer.resblocks.10.ln_1.weight FLOAT[1024] fd222ca7ae2d
|
|
visual.transformer.resblocks.10.ln_2.bias FLOAT[1024] b12632e20529
|
|
visual.transformer.resblocks.10.ln_2.weight FLOAT[1024] c375eea0394f
|
|
visual.transformer.resblocks.10.mlp.c_fc.bias FLOAT[4096] 6743ad5f59b1
|
|
visual.transformer.resblocks.10.mlp.c_proj.bias FLOAT[1024] 400a8bb8c8c5
|
|
visual.transformer.resblocks.11.attn.in_proj_bias FLOAT[3072] 57b06492a4d0
|
|
visual.transformer.resblocks.11.attn.out_proj.bias FLOAT[1024] 3fcd301f5484
|
|
visual.transformer.resblocks.11.ln_1.bias FLOAT[1024] 94b5bfb608d6
|
|
visual.transformer.resblocks.11.ln_1.weight FLOAT[1024] 9233b36135a9
|
|
visual.transformer.resblocks.11.ln_2.bias FLOAT[1024] a87a5425ddf3
|
|
visual.transformer.resblocks.11.ln_2.weight FLOAT[1024] 47d3a6cd46f9
|
|
visual.transformer.resblocks.11.mlp.c_fc.bias FLOAT[4096] 54fdd8fab9eb
|
|
visual.transformer.resblocks.11.mlp.c_proj.bias FLOAT[1024] 267d2bdd865a
|
|
visual.transformer.resblocks.12.attn.in_proj_bias FLOAT[3072] 0fcac749d209
|
|
visual.transformer.resblocks.12.attn.out_proj.bias FLOAT[1024] 406d67ffe1be
|
|
visual.transformer.resblocks.12.ln_1.bias FLOAT[1024] b32c9beaad5b
|
|
visual.transformer.resblocks.12.ln_1.weight FLOAT[1024] eb9f71d99be5
|
|
visual.transformer.resblocks.12.ln_2.bias FLOAT[1024] 9f3aff976b3a
|
|
visual.transformer.resblocks.12.ln_2.weight FLOAT[1024] 0df0388ec203
|
|
visual.transformer.resblocks.12.mlp.c_fc.bias FLOAT[4096] dafd495eeca9
|
|
visual.transformer.resblocks.12.mlp.c_proj.bias FLOAT[1024] aa88619ebf80
|
|
visual.transformer.resblocks.13.attn.in_proj_bias FLOAT[3072] 6240d2214f1e
|
|
visual.transformer.resblocks.13.attn.out_proj.bias FLOAT[1024] fc68b1283570
|
|
visual.transformer.resblocks.13.ln_1.bias FLOAT[1024] 539402e56b84
|
|
visual.transformer.resblocks.13.ln_1.weight FLOAT[1024] 56e450f79c98
|
|
visual.transformer.resblocks.13.ln_2.bias FLOAT[1024] dd4d724962de
|
|
visual.transformer.resblocks.13.ln_2.weight FLOAT[1024] b86f84bb7fdf
|
|
visual.transformer.resblocks.13.mlp.c_fc.bias FLOAT[4096] d66f447cf0f3
|
|
visual.transformer.resblocks.13.mlp.c_proj.bias FLOAT[1024] 349983596c87
|
|
visual.transformer.resblocks.14.attn.in_proj_bias FLOAT[3072] 6d916b62a464
|
|
visual.transformer.resblocks.14.attn.out_proj.bias FLOAT[1024] 55c13d33c8a6
|
|
visual.transformer.resblocks.14.ln_1.bias FLOAT[1024] 5b6b1c8c9cee
|
|
visual.transformer.resblocks.14.ln_1.weight FLOAT[1024] a595b61dad5f
|
|
visual.transformer.resblocks.14.ln_2.bias FLOAT[1024] df0fd630ab7e
|
|
visual.transformer.resblocks.14.ln_2.weight FLOAT[1024] bf1f754ecf7a
|
|
visual.transformer.resblocks.14.mlp.c_fc.bias FLOAT[4096] 6ddd4cc51d05
|
|
visual.transformer.resblocks.14.mlp.c_proj.bias FLOAT[1024] 0a780a68d40c
|
|
visual.transformer.resblocks.15.attn.in_proj_bias FLOAT[3072] e354b3407034
|
|
visual.transformer.resblocks.15.attn.out_proj.bias FLOAT[1024] 3afffaba07b7
|
|
visual.transformer.resblocks.15.ln_1.bias FLOAT[1024] b08db6f5cd72
|
|
visual.transformer.resblocks.15.ln_1.weight FLOAT[1024] 54602be0b89b
|
|
visual.transformer.resblocks.15.ln_2.bias FLOAT[1024] 4ae39033378a
|
|
visual.transformer.resblocks.15.ln_2.weight FLOAT[1024] 65e1cbbfde0c
|
|
visual.transformer.resblocks.15.mlp.c_fc.bias FLOAT[4096] 63a47568d04e
|
|
visual.transformer.resblocks.15.mlp.c_proj.bias FLOAT[1024] bd59cdbfbb23
|
|
visual.transformer.resblocks.16.attn.in_proj_bias FLOAT[3072] a20738ca97a8
|
|
visual.transformer.resblocks.16.attn.out_proj.bias FLOAT[1024] 985a0732413c
|
|
visual.transformer.resblocks.16.ln_1.bias FLOAT[1024] 542857e87076
|
|
visual.transformer.resblocks.16.ln_1.weight FLOAT[1024] 9a21abf21ef9
|
|
visual.transformer.resblocks.16.ln_2.bias FLOAT[1024] 2f17f39f4835
|
|
visual.transformer.resblocks.16.ln_2.weight FLOAT[1024] 396c9766d0a2
|
|
visual.transformer.resblocks.16.mlp.c_fc.bias FLOAT[4096] ce5b098767d2
|
|
visual.transformer.resblocks.16.mlp.c_proj.bias FLOAT[1024] ca85cc1c2a62
|
|
visual.transformer.resblocks.17.attn.in_proj_bias FLOAT[3072] 9e70a743ddc7
|
|
visual.transformer.resblocks.17.attn.out_proj.bias FLOAT[1024] fa51dcc70094
|
|
visual.transformer.resblocks.17.ln_1.bias FLOAT[1024] 533a43227e8d
|
|
visual.transformer.resblocks.17.ln_1.weight FLOAT[1024] a7f8326e0c54
|
|
visual.transformer.resblocks.17.ln_2.bias FLOAT[1024] 46c38d3de85e
|
|
visual.transformer.resblocks.17.ln_2.weight FLOAT[1024] 7dd3310795fb
|
|
visual.transformer.resblocks.17.mlp.c_fc.bias FLOAT[4096] 0f018f428f58
|
|
visual.transformer.resblocks.17.mlp.c_proj.bias FLOAT[1024] eb132530090d
|
|
visual.transformer.resblocks.18.attn.in_proj_bias FLOAT[3072] 75f1817e6e09
|
|
visual.transformer.resblocks.18.attn.out_proj.bias FLOAT[1024] 7d374b20b02d
|
|
visual.transformer.resblocks.18.ln_1.bias FLOAT[1024] 6d3359456dbc
|
|
visual.transformer.resblocks.18.ln_1.weight FLOAT[1024] 25da82a6ea31
|
|
visual.transformer.resblocks.18.ln_2.bias FLOAT[1024] 94be4313e2c1
|
|
visual.transformer.resblocks.18.ln_2.weight FLOAT[1024] 91d8ed900e24
|
|
visual.transformer.resblocks.18.mlp.c_fc.bias FLOAT[4096] 7b50b7443cae
|
|
visual.transformer.resblocks.18.mlp.c_proj.bias FLOAT[1024] 18711b15846d
|
|
visual.transformer.resblocks.19.attn.in_proj_bias FLOAT[3072] e807aed3a823
|
|
visual.transformer.resblocks.19.attn.out_proj.bias FLOAT[1024] 241fc8942002
|
|
visual.transformer.resblocks.19.ln_1.bias FLOAT[1024] 4f8ffa6c9477
|
|
visual.transformer.resblocks.19.ln_1.weight FLOAT[1024] 5115d85cbc40
|
|
visual.transformer.resblocks.19.ln_2.bias FLOAT[1024] 745723dbc9ab
|
|
visual.transformer.resblocks.19.ln_2.weight FLOAT[1024] a055bbc1cf20
|
|
visual.transformer.resblocks.19.mlp.c_fc.bias FLOAT[4096] f0ba1bcc289b
|
|
visual.transformer.resblocks.19.mlp.c_proj.bias FLOAT[1024] e90b581d9586
|
|
visual.transformer.resblocks.2.attn.in_proj_bias FLOAT[3072] 163af8a4029e
|
|
visual.transformer.resblocks.2.attn.out_proj.bias FLOAT[1024] fc8062f7ab21
|
|
visual.transformer.resblocks.2.ln_1.bias FLOAT[1024] 40590810f06d
|
|
visual.transformer.resblocks.2.ln_1.weight FLOAT[1024] 747239a00a03
|
|
visual.transformer.resblocks.2.ln_2.bias FLOAT[1024] b9ea4ec7279a
|
|
visual.transformer.resblocks.2.ln_2.weight FLOAT[1024] 54f2be11440c
|
|
visual.transformer.resblocks.2.mlp.c_fc.bias FLOAT[4096] fb53f6212a9e
|
|
visual.transformer.resblocks.2.mlp.c_proj.bias FLOAT[1024] a5f8e127cbd5
|
|
visual.transformer.resblocks.20.attn.in_proj_bias FLOAT[3072] 076989b1a2a3
|
|
visual.transformer.resblocks.20.attn.out_proj.bias FLOAT[1024] 0c2f725db3ac
|
|
visual.transformer.resblocks.20.ln_1.bias FLOAT[1024] 16c0a0b483d4
|
|
visual.transformer.resblocks.20.ln_1.weight FLOAT[1024] 5e2e704138df
|
|
visual.transformer.resblocks.20.ln_2.bias FLOAT[1024] 703332375c5f
|
|
visual.transformer.resblocks.20.ln_2.weight FLOAT[1024] 9f8dad11d301
|
|
visual.transformer.resblocks.20.mlp.c_fc.bias FLOAT[4096] 6c67b17140d0
|
|
visual.transformer.resblocks.20.mlp.c_proj.bias FLOAT[1024] e06bb00411bc
|
|
visual.transformer.resblocks.21.attn.in_proj_bias FLOAT[3072] 006a0d5098ba
|
|
visual.transformer.resblocks.21.attn.out_proj.bias FLOAT[1024] fc9e069d4328
|
|
visual.transformer.resblocks.21.ln_1.bias FLOAT[1024] baa17fe0a8c2
|
|
visual.transformer.resblocks.21.ln_1.weight FLOAT[1024] c0b304f27568
|
|
visual.transformer.resblocks.21.ln_2.bias FLOAT[1024] 74a38c56ef16
|
|
visual.transformer.resblocks.21.ln_2.weight FLOAT[1024] fb88c30de487
|
|
visual.transformer.resblocks.21.mlp.c_fc.bias FLOAT[4096] 5c799959e5a6
|
|
visual.transformer.resblocks.21.mlp.c_proj.bias FLOAT[1024] 9dce71875fbc
|
|
visual.transformer.resblocks.22.attn.in_proj_bias FLOAT[3072] aecbdf6afc69
|
|
visual.transformer.resblocks.22.attn.out_proj.bias FLOAT[1024] 71dfcbb3d9fc
|
|
visual.transformer.resblocks.22.ln_1.bias FLOAT[1024] e86a29328f7d
|
|
visual.transformer.resblocks.22.ln_1.weight FLOAT[1024] a266eee66fe0
|
|
visual.transformer.resblocks.22.ln_2.bias FLOAT[1024] f9787a4ea5f8
|
|
visual.transformer.resblocks.22.ln_2.weight FLOAT[1024] 66de7a21326b
|
|
visual.transformer.resblocks.22.mlp.c_fc.bias FLOAT[4096] d000f4a134e1
|
|
visual.transformer.resblocks.22.mlp.c_proj.bias FLOAT[1024] 92831fb758bb
|
|
visual.transformer.resblocks.23.attn.in_proj_bias FLOAT[3072] 92f13c000de8
|
|
visual.transformer.resblocks.23.attn.out_proj.bias FLOAT[1024] 0d5d24506676
|
|
visual.transformer.resblocks.23.ln_1.bias FLOAT[1024] 360d2928470b
|
|
visual.transformer.resblocks.23.ln_1.weight FLOAT[1024] 838dc8233d10
|
|
visual.transformer.resblocks.23.ln_2.bias FLOAT[1024] 51799cbeb377
|
|
visual.transformer.resblocks.23.ln_2.weight FLOAT[1024] 853492004e43
|
|
visual.transformer.resblocks.23.mlp.c_fc.bias FLOAT[4096] bd5b4571400e
|
|
visual.transformer.resblocks.23.mlp.c_proj.bias FLOAT[1024] 342f941af2e1
|
|
visual.transformer.resblocks.3.attn.in_proj_bias FLOAT[3072] 225f4643b209
|
|
visual.transformer.resblocks.3.attn.out_proj.bias FLOAT[1024] b41c4e34c334
|
|
visual.transformer.resblocks.3.ln_1.bias FLOAT[1024] f3c46add2c50
|
|
visual.transformer.resblocks.3.ln_1.weight FLOAT[1024] a8afc2c9bac5
|
|
visual.transformer.resblocks.3.ln_2.bias FLOAT[1024] 4a4fcdd01065
|
|
visual.transformer.resblocks.3.ln_2.weight FLOAT[1024] 51bc7aef4cf6
|
|
visual.transformer.resblocks.3.mlp.c_fc.bias FLOAT[4096] 45f61e989ce9
|
|
visual.transformer.resblocks.3.mlp.c_proj.bias FLOAT[1024] 61e23993cf5d
|
|
visual.transformer.resblocks.4.attn.in_proj_bias FLOAT[3072] bf026832a713
|
|
visual.transformer.resblocks.4.attn.out_proj.bias FLOAT[1024] f0933baa9621
|
|
visual.transformer.resblocks.4.ln_1.bias FLOAT[1024] a6a52b0f0f78
|
|
visual.transformer.resblocks.4.ln_1.weight FLOAT[1024] 3792a0bee5b7
|
|
visual.transformer.resblocks.4.ln_2.bias FLOAT[1024] 6f3c4f6a5c64
|
|
visual.transformer.resblocks.4.ln_2.weight FLOAT[1024] 6d60eef25780
|
|
visual.transformer.resblocks.4.mlp.c_fc.bias FLOAT[4096] 8f489d3f341e
|
|
visual.transformer.resblocks.4.mlp.c_proj.bias FLOAT[1024] fbea6249e28c
|
|
visual.transformer.resblocks.5.attn.in_proj_bias FLOAT[3072] 512db7c0fe3f
|
|
visual.transformer.resblocks.5.attn.out_proj.bias FLOAT[1024] 4c3183a088b3
|
|
visual.transformer.resblocks.5.ln_1.bias FLOAT[1024] 68279ab14b75
|
|
visual.transformer.resblocks.5.ln_1.weight FLOAT[1024] 4b580e2194e3
|
|
visual.transformer.resblocks.5.ln_2.bias FLOAT[1024] d02c3f505fa8
|
|
visual.transformer.resblocks.5.ln_2.weight FLOAT[1024] f8353026e585
|
|
visual.transformer.resblocks.5.mlp.c_fc.bias FLOAT[4096] 94e1cde4a784
|
|
visual.transformer.resblocks.5.mlp.c_proj.bias FLOAT[1024] 9613840f711b
|
|
visual.transformer.resblocks.6.attn.in_proj_bias FLOAT[3072] b4157f1b3b57
|
|
visual.transformer.resblocks.6.attn.out_proj.bias FLOAT[1024] 1c738c1255b3
|
|
visual.transformer.resblocks.6.ln_1.bias FLOAT[1024] b40e345c48e0
|
|
visual.transformer.resblocks.6.ln_1.weight FLOAT[1024] 634a7191cf8a
|
|
visual.transformer.resblocks.6.ln_2.bias FLOAT[1024] 2678048155d4
|
|
visual.transformer.resblocks.6.ln_2.weight FLOAT[1024] 49823bcb3290
|
|
visual.transformer.resblocks.6.mlp.c_fc.bias FLOAT[4096] beae1351a7e9
|
|
visual.transformer.resblocks.6.mlp.c_proj.bias FLOAT[1024] 52f78e971b7d
|
|
visual.transformer.resblocks.7.attn.in_proj_bias FLOAT[3072] aa9209f2d75f
|
|
visual.transformer.resblocks.7.attn.out_proj.bias FLOAT[1024] 6c6b1af11dd6
|
|
visual.transformer.resblocks.7.ln_1.bias FLOAT[1024] 667535d4c1c3
|
|
visual.transformer.resblocks.7.ln_1.weight FLOAT[1024] df05f5cd6904
|
|
visual.transformer.resblocks.7.ln_2.bias FLOAT[1024] eaddc3e40f63
|
|
visual.transformer.resblocks.7.ln_2.weight FLOAT[1024] 47f8858c9179
|
|
visual.transformer.resblocks.7.mlp.c_fc.bias FLOAT[4096] 74d2861c8ce6
|
|
visual.transformer.resblocks.7.mlp.c_proj.bias FLOAT[1024] f20f4d85a8f1
|
|
visual.transformer.resblocks.8.attn.in_proj_bias FLOAT[3072] 362a19d50ceb
|
|
visual.transformer.resblocks.8.attn.out_proj.bias FLOAT[1024] fb01c0599bfe
|
|
visual.transformer.resblocks.8.ln_1.bias FLOAT[1024] 83f8107f66d4
|
|
visual.transformer.resblocks.8.ln_1.weight FLOAT[1024] 6a41f316fecf
|
|
visual.transformer.resblocks.8.ln_2.bias FLOAT[1024] 1a5cdc4bc5b8
|
|
visual.transformer.resblocks.8.ln_2.weight FLOAT[1024] 09d18370f37c
|
|
visual.transformer.resblocks.8.mlp.c_fc.bias FLOAT[4096] 2178d0a92341
|
|
visual.transformer.resblocks.8.mlp.c_proj.bias FLOAT[1024] c06285b5fd32
|
|
visual.transformer.resblocks.9.attn.in_proj_bias FLOAT[3072] c9b4705fdac3
|
|
visual.transformer.resblocks.9.attn.out_proj.bias FLOAT[1024] b677626c1635
|
|
visual.transformer.resblocks.9.ln_1.bias FLOAT[1024] 74d6a771b8ce
|
|
visual.transformer.resblocks.9.ln_1.weight FLOAT[1024] bddbc9c928c9
|
|
visual.transformer.resblocks.9.ln_2.bias FLOAT[1024] f797d3b6eb50
|
|
visual.transformer.resblocks.9.ln_2.weight FLOAT[1024] 6eaf32f39b47
|
|
visual.transformer.resblocks.9.mlp.c_fc.bias FLOAT[4096] a40ea97517b3
|
|
visual.transformer.resblocks.9.mlp.c_proj.bias FLOAT[1024] 7713a33f202c
|