aotrih commited on
Commit
bf521df
·
verified ·
1 Parent(s): 89c8c39

whisperkittools-4925e685124152a1087d88d381b6e705e2c90cea generated files: openai_whisper-tiny.en

Browse files
openai_whisper-tiny.en/AudioEncoder.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b0342c606f52154e45531318e3f39bd60932c296d2fe01b8337f24d67065db67
3
  size 16422784
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1bcda05382b782419371417ec403cad5eb7d86352bc2c02465c087ed93dde24
3
  size 16422784
openai_whisper-tiny.en/MelSpectrogram.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:07f35fcd9f3fea7554f2065fb9436a78712eda880762070a397c777bb0f7adc6
3
  size 354080
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a0bf21a34f58b276938eee139df876f35ce972c6f60a0d8c2a3644102e19dc9d
3
  size 354080
openai_whisper-tiny.en/TextDecoder.mlmodelc/analytics/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3340541dfbe9cb353087fda72c13b0bf10aa6130192b1d30381a5d9cc30c3ca4
3
  size 243
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d896bfb3402104303782763f619998dd35c72c3618e7c1ce51520c78d4599b4
3
  size 243
openai_whisper-tiny.en/TextDecoder.mlmodelc/coremldata.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1ffde5d5b1131bc4938659b32f817ed62f2d76260439cd3bd6f1ce3d96fd4c4f
3
  size 593
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ba57b843270ebc0f8d8268179b7498a1d8e313a4ec970afa3dab94525f246e4d
3
  size 593
openai_whisper-tiny.en/TextDecoder.mlmodelc/metadata.json CHANGED
@@ -102,9 +102,9 @@
102
  "hasShapeFlexibility" : "0",
103
  "isOptional" : "0",
104
  "dataType" : "Float16",
105
- "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 448)",
106
  "shortDescription" : "",
107
- "shape" : "[1, 1536, 1, 448]",
108
  "name" : "key_cache",
109
  "type" : "MultiArray"
110
  },
@@ -112,9 +112,9 @@
112
  "hasShapeFlexibility" : "0",
113
  "isOptional" : "0",
114
  "dataType" : "Float16",
115
- "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 448)",
116
  "shortDescription" : "",
117
- "shape" : "[1, 1536, 1, 448]",
118
  "name" : "value_cache",
119
  "type" : "MultiArray"
120
  },
@@ -122,9 +122,9 @@
122
  "hasShapeFlexibility" : "0",
123
  "isOptional" : "0",
124
  "dataType" : "Float16",
125
- "formattedType" : "MultiArray (Float16 1 × 448)",
126
  "shortDescription" : "",
127
- "shape" : "[1, 448]",
128
  "name" : "kv_cache_update_mask",
129
  "type" : "MultiArray"
130
  },
@@ -142,9 +142,9 @@
142
  "hasShapeFlexibility" : "0",
143
  "isOptional" : "0",
144
  "dataType" : "Float16",
145
- "formattedType" : "MultiArray (Float16 1 × 448)",
146
  "shortDescription" : "",
147
- "shape" : "[1, 448]",
148
  "name" : "decoder_key_padding_mask",
149
  "type" : "MultiArray"
150
  }
 
102
  "hasShapeFlexibility" : "0",
103
  "isOptional" : "0",
104
  "dataType" : "Float16",
105
+ "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 224)",
106
  "shortDescription" : "",
107
+ "shape" : "[1, 1536, 1, 224]",
108
  "name" : "key_cache",
109
  "type" : "MultiArray"
110
  },
 
112
  "hasShapeFlexibility" : "0",
113
  "isOptional" : "0",
114
  "dataType" : "Float16",
115
+ "formattedType" : "MultiArray (Float16 1 × 1536 × 1 × 224)",
116
  "shortDescription" : "",
117
+ "shape" : "[1, 1536, 1, 224]",
118
  "name" : "value_cache",
119
  "type" : "MultiArray"
120
  },
 
122
  "hasShapeFlexibility" : "0",
123
  "isOptional" : "0",
124
  "dataType" : "Float16",
125
+ "formattedType" : "MultiArray (Float16 1 × 224)",
126
  "shortDescription" : "",
127
+ "shape" : "[1, 224]",
128
  "name" : "kv_cache_update_mask",
129
  "type" : "MultiArray"
130
  },
 
142
  "hasShapeFlexibility" : "0",
143
  "isOptional" : "0",
144
  "dataType" : "Float16",
145
+ "formattedType" : "MultiArray (Float16 1 × 224)",
146
  "shortDescription" : "",
147
+ "shape" : "[1, 224]",
148
  "name" : "decoder_key_padding_mask",
149
  "type" : "MultiArray"
150
  }
openai_whisper-tiny.en/TextDecoder.mlmodelc/model.mil CHANGED
@@ -1,7 +1,7 @@
1
  program(1.0)
2
  [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}, {"coremltools-component-torch", "2.1.2"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.1"}})]
3
  {
4
- func main<ios17>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 448]> decoder_key_padding_mask, tensor<fp16, [1, 384, 1, 1500]> encoder_output_embeds, tensor<int32, [1]> input_ids, tensor<fp16, [1, 1536, 1, 448]> key_cache, tensor<fp16, [1, 448]> kv_cache_update_mask, tensor<fp16, [1, 1536, 1, 448]> value_cache) {
5
  tensor<int32, []> var_24_axis_0 = const()[name = tensor<string, []>("op_24_axis_0"), val = tensor<int32, []>(0)];
6
  tensor<int32, []> var_24_batch_dims_0 = const()[name = tensor<string, []>("op_24_batch_dims_0"), val = tensor<int32, []>(0)];
7
  tensor<bool, []> var_24_validate_indices_0 = const()[name = tensor<string, []>("op_24_validate_indices_0"), val = tensor<bool, []>(false)];
@@ -21,10 +21,10 @@ program(1.0)
21
  tensor<fp16, [1, 384, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_42_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")];
22
  tensor<int32, [4]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [4]>([384, 384, 384, 384])];
23
  tensor<int32, []> var_47_axis_0 = const()[name = tensor<string, []>("op_47_axis_0"), val = tensor<int32, []>(1)];
24
- tensor<fp16, [1, 384, 1, 448]> var_47_cast_fp16_0, tensor<fp16, [1, 384, 1, 448]> var_47_cast_fp16_1, tensor<fp16, [1, 384, 1, 448]> var_47_cast_fp16_2, tensor<fp16, [1, 384, 1, 448]> var_47_cast_fp16_3 = split(axis = var_47_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor<string, []>("op_47_cast_fp16")];
25
  tensor<int32, [4]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [4]>([384, 384, 384, 384])];
26
  tensor<int32, []> var_54_axis_0 = const()[name = tensor<string, []>("op_54_axis_0"), val = tensor<int32, []>(1)];
27
- tensor<fp16, [1, 384, 1, 448]> var_54_cast_fp16_0, tensor<fp16, [1, 384, 1, 448]> var_54_cast_fp16_1, tensor<fp16, [1, 384, 1, 448]> var_54_cast_fp16_2, tensor<fp16, [1, 384, 1, 448]> var_54_cast_fp16_3 = split(axis = var_54_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor<string, []>("op_54_cast_fp16")];
28
  tensor<int32, []> var_64 = const()[name = tensor<string, []>("op_64"), val = tensor<int32, []>(3)];
29
  tensor<int32, []> var_71 = const()[name = tensor<string, []>("op_71"), val = tensor<int32, []>(1)];
30
  tensor<bool, []> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<bool, []>(true)];
@@ -66,34 +66,34 @@ program(1.0)
66
  tensor<fp16, [384]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41064896)))];
67
  tensor<fp16, [1, 384, 1, 1]> current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = var_121, groups = var_71, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = var_119, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("current_value_1_cast_fp16")];
68
  tensor<int32, [1]> var_125_axes_0 = const()[name = tensor<string, []>("op_125_axes_0"), val = tensor<int32, [1]>([1])];
69
- tensor<fp16, [1, 1, 448]> var_125_cast_fp16 = expand_dims(axes = var_125_axes_0, x = kv_cache_update_mask)[name = tensor<string, []>("op_125_cast_fp16")];
70
  tensor<int32, [1]> var_126_axes_0 = const()[name = tensor<string, []>("op_126_axes_0"), val = tensor<int32, [1]>([2])];
71
- tensor<fp16, [1, 1, 1, 448]> var_126_cast_fp16 = expand_dims(axes = var_126_axes_0, x = var_125_cast_fp16)[name = tensor<string, []>("op_126_cast_fp16")];
72
- tensor<fp16, [1, 384, 1, 448]> var_128_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_128_cast_fp16")];
73
  tensor<fp16, []> var_65_to_fp16 = const()[name = tensor<string, []>("op_65_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
74
- tensor<fp16, [1, 1, 1, 448]> var_129_cast_fp16 = sub(x = var_65_to_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_129_cast_fp16")];
75
- tensor<fp16, [1, 384, 1, 448]> var_130_cast_fp16 = mul(x = var_47_cast_fp16_0, y = var_129_cast_fp16)[name = tensor<string, []>("op_130_cast_fp16")];
76
- tensor<fp16, [1, 384, 1, 448]> key_1_cast_fp16 = add(x = var_128_cast_fp16, y = var_130_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")];
77
- tensor<fp16, [1, 384, 1, 448]> var_132_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_132_cast_fp16")];
78
- tensor<fp16, [1, 384, 1, 448]> var_134_cast_fp16 = mul(x = var_54_cast_fp16_0, y = var_129_cast_fp16)[name = tensor<string, []>("op_134_cast_fp16")];
79
- tensor<fp16, [1, 384, 1, 448]> value_1_cast_fp16 = add(x = var_132_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")];
80
  tensor<int32, [4]> var_137 = const()[name = tensor<string, []>("op_137"), val = tensor<int32, [4]>([1, 6, 64, -1])];
81
  tensor<fp16, [1, 6, 64, 1]> var_138_cast_fp16 = reshape(shape = var_137, x = query_1_cast_fp16)[name = tensor<string, []>("op_138_cast_fp16")];
82
  tensor<fp16, []> var_139_to_fp16 = const()[name = tensor<string, []>("op_139_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
83
  tensor<fp16, [1, 6, 64, 1]> var_140_cast_fp16 = mul(x = var_138_cast_fp16, y = var_139_to_fp16)[name = tensor<string, []>("op_140_cast_fp16")];
84
  tensor<int32, [4]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [4]>([1, 6, 64, -1])];
85
- tensor<fp16, [1, 6, 64, 448]> var_142_cast_fp16 = reshape(shape = var_141, x = key_1_cast_fp16)[name = tensor<string, []>("op_142_cast_fp16")];
86
  tensor<bool, []> mh_w_1_transpose_x_0 = const()[name = tensor<string, []>("mh_w_1_transpose_x_0"), val = tensor<bool, []>(true)];
87
  tensor<bool, []> mh_w_1_transpose_y_0 = const()[name = tensor<string, []>("mh_w_1_transpose_y_0"), val = tensor<bool, []>(false)];
88
- tensor<fp16, [1, 6, 1, 448]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_140_cast_fp16, y = var_142_cast_fp16)[name = tensor<string, []>("mh_w_1_cast_fp16")];
89
  tensor<int32, [1]> var_146_axes_0 = const()[name = tensor<string, []>("op_146_axes_0"), val = tensor<int32, [1]>([1])];
90
- tensor<fp16, [1, 1, 448]> var_146_cast_fp16 = expand_dims(axes = var_146_axes_0, x = decoder_key_padding_mask)[name = tensor<string, []>("op_146_cast_fp16")];
91
  tensor<int32, [1]> var_147_axes_0 = const()[name = tensor<string, []>("op_147_axes_0"), val = tensor<int32, [1]>([2])];
92
- tensor<fp16, [1, 1, 1, 448]> var_147_cast_fp16 = expand_dims(axes = var_147_axes_0, x = var_146_cast_fp16)[name = tensor<string, []>("op_147_cast_fp16")];
93
- tensor<fp16, [1, 6, 1, 448]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")];
94
- tensor<fp16, [1, 6, 1, 448]> var_150_cast_fp16 = softmax(axis = var_64, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_150_cast_fp16")];
95
  tensor<int32, [4]> var_151 = const()[name = tensor<string, []>("op_151"), val = tensor<int32, [4]>([1, 6, 64, -1])];
96
- tensor<fp16, [1, 6, 64, 448]> var_152_cast_fp16 = reshape(shape = var_151, x = value_1_cast_fp16)[name = tensor<string, []>("op_152_cast_fp16")];
97
  tensor<bool, []> attn_1_transpose_x_0 = const()[name = tensor<string, []>("attn_1_transpose_x_0"), val = tensor<bool, []>(false)];
98
  tensor<bool, []> attn_1_transpose_y_0 = const()[name = tensor<string, []>("attn_1_transpose_y_0"), val = tensor<bool, []>(true)];
99
  tensor<fp16, [1, 6, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_152_cast_fp16, y = var_150_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")];
@@ -237,25 +237,25 @@ program(1.0)
237
  tensor<fp16, [384, 384, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45503104)))];
238
  tensor<fp16, [384]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45798080)))];
239
  tensor<fp16, [1, 384, 1, 1]> current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = var_335, groups = var_285, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = var_333, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor<string, []>("current_value_3_cast_fp16")];
240
- tensor<fp16, [1, 384, 1, 448]> var_342_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_342_cast_fp16")];
241
- tensor<fp16, [1, 384, 1, 448]> var_344_cast_fp16 = mul(x = var_47_cast_fp16_1, y = var_129_cast_fp16)[name = tensor<string, []>("op_344_cast_fp16")];
242
- tensor<fp16, [1, 384, 1, 448]> key_5_cast_fp16 = add(x = var_342_cast_fp16, y = var_344_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")];
243
- tensor<fp16, [1, 384, 1, 448]> var_346_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_346_cast_fp16")];
244
- tensor<fp16, [1, 384, 1, 448]> var_348_cast_fp16 = mul(x = var_54_cast_fp16_1, y = var_129_cast_fp16)[name = tensor<string, []>("op_348_cast_fp16")];
245
- tensor<fp16, [1, 384, 1, 448]> value_5_cast_fp16 = add(x = var_346_cast_fp16, y = var_348_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")];
246
  tensor<int32, [4]> var_351 = const()[name = tensor<string, []>("op_351"), val = tensor<int32, [4]>([1, 6, 64, -1])];
247
  tensor<fp16, [1, 6, 64, 1]> var_352_cast_fp16 = reshape(shape = var_351, x = query_5_cast_fp16)[name = tensor<string, []>("op_352_cast_fp16")];
248
  tensor<fp16, []> var_353_to_fp16 = const()[name = tensor<string, []>("op_353_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
249
  tensor<fp16, [1, 6, 64, 1]> var_354_cast_fp16 = mul(x = var_352_cast_fp16, y = var_353_to_fp16)[name = tensor<string, []>("op_354_cast_fp16")];
250
  tensor<int32, [4]> var_355 = const()[name = tensor<string, []>("op_355"), val = tensor<int32, [4]>([1, 6, 64, -1])];
251
- tensor<fp16, [1, 6, 64, 448]> var_356_cast_fp16 = reshape(shape = var_355, x = key_5_cast_fp16)[name = tensor<string, []>("op_356_cast_fp16")];
252
  tensor<bool, []> mh_w_7_transpose_x_0 = const()[name = tensor<string, []>("mh_w_7_transpose_x_0"), val = tensor<bool, []>(true)];
253
  tensor<bool, []> mh_w_7_transpose_y_0 = const()[name = tensor<string, []>("mh_w_7_transpose_y_0"), val = tensor<bool, []>(false)];
254
- tensor<fp16, [1, 6, 1, 448]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_354_cast_fp16, y = var_356_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")];
255
- tensor<fp16, [1, 6, 1, 448]> mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")];
256
- tensor<fp16, [1, 6, 1, 448]> var_364_cast_fp16 = softmax(axis = var_278, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_364_cast_fp16")];
257
  tensor<int32, [4]> var_365 = const()[name = tensor<string, []>("op_365"), val = tensor<int32, [4]>([1, 6, 64, -1])];
258
- tensor<fp16, [1, 6, 64, 448]> var_366_cast_fp16 = reshape(shape = var_365, x = value_5_cast_fp16)[name = tensor<string, []>("op_366_cast_fp16")];
259
  tensor<bool, []> attn_5_transpose_x_0 = const()[name = tensor<string, []>("attn_5_transpose_x_0"), val = tensor<bool, []>(false)];
260
  tensor<bool, []> attn_5_transpose_y_0 = const()[name = tensor<string, []>("attn_5_transpose_y_0"), val = tensor<bool, []>(true)];
261
  tensor<fp16, [1, 6, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_366_cast_fp16, y = var_364_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")];
@@ -399,25 +399,25 @@ program(1.0)
399
  tensor<fp16, [384, 384, 1, 1]> layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50236288)))];
400
  tensor<fp16, [384]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50531264)))];
401
  tensor<fp16, [1, 384, 1, 1]> current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = var_549, groups = var_499, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = var_547, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor<string, []>("current_value_5_cast_fp16")];
402
- tensor<fp16, [1, 384, 1, 448]> var_556_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_556_cast_fp16")];
403
- tensor<fp16, [1, 384, 1, 448]> var_558_cast_fp16 = mul(x = var_47_cast_fp16_2, y = var_129_cast_fp16)[name = tensor<string, []>("op_558_cast_fp16")];
404
- tensor<fp16, [1, 384, 1, 448]> key_9_cast_fp16 = add(x = var_556_cast_fp16, y = var_558_cast_fp16)[name = tensor<string, []>("key_9_cast_fp16")];
405
- tensor<fp16, [1, 384, 1, 448]> var_560_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_560_cast_fp16")];
406
- tensor<fp16, [1, 384, 1, 448]> var_562_cast_fp16 = mul(x = var_54_cast_fp16_2, y = var_129_cast_fp16)[name = tensor<string, []>("op_562_cast_fp16")];
407
- tensor<fp16, [1, 384, 1, 448]> value_9_cast_fp16 = add(x = var_560_cast_fp16, y = var_562_cast_fp16)[name = tensor<string, []>("value_9_cast_fp16")];
408
  tensor<int32, [4]> var_565 = const()[name = tensor<string, []>("op_565"), val = tensor<int32, [4]>([1, 6, 64, -1])];
409
  tensor<fp16, [1, 6, 64, 1]> var_566_cast_fp16 = reshape(shape = var_565, x = query_9_cast_fp16)[name = tensor<string, []>("op_566_cast_fp16")];
410
  tensor<fp16, []> var_567_to_fp16 = const()[name = tensor<string, []>("op_567_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
411
  tensor<fp16, [1, 6, 64, 1]> var_568_cast_fp16 = mul(x = var_566_cast_fp16, y = var_567_to_fp16)[name = tensor<string, []>("op_568_cast_fp16")];
412
  tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, 6, 64, -1])];
413
- tensor<fp16, [1, 6, 64, 448]> var_570_cast_fp16 = reshape(shape = var_569, x = key_9_cast_fp16)[name = tensor<string, []>("op_570_cast_fp16")];
414
  tensor<bool, []> mh_w_13_transpose_x_0 = const()[name = tensor<string, []>("mh_w_13_transpose_x_0"), val = tensor<bool, []>(true)];
415
  tensor<bool, []> mh_w_13_transpose_y_0 = const()[name = tensor<string, []>("mh_w_13_transpose_y_0"), val = tensor<bool, []>(false)];
416
- tensor<fp16, [1, 6, 1, 448]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_568_cast_fp16, y = var_570_cast_fp16)[name = tensor<string, []>("mh_w_13_cast_fp16")];
417
- tensor<fp16, [1, 6, 1, 448]> mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_15_cast_fp16")];
418
- tensor<fp16, [1, 6, 1, 448]> var_578_cast_fp16 = softmax(axis = var_492, x = mh_w_15_cast_fp16)[name = tensor<string, []>("op_578_cast_fp16")];
419
  tensor<int32, [4]> var_579 = const()[name = tensor<string, []>("op_579"), val = tensor<int32, [4]>([1, 6, 64, -1])];
420
- tensor<fp16, [1, 6, 64, 448]> var_580_cast_fp16 = reshape(shape = var_579, x = value_9_cast_fp16)[name = tensor<string, []>("op_580_cast_fp16")];
421
  tensor<bool, []> attn_9_transpose_x_0 = const()[name = tensor<string, []>("attn_9_transpose_x_0"), val = tensor<bool, []>(false)];
422
  tensor<bool, []> attn_9_transpose_y_0 = const()[name = tensor<string, []>("attn_9_transpose_y_0"), val = tensor<bool, []>(true)];
423
  tensor<fp16, [1, 6, 64, 1]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_580_cast_fp16, y = var_578_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")];
@@ -561,25 +561,25 @@ program(1.0)
561
  tensor<fp16, [384, 384, 1, 1]> layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54969472)))];
562
  tensor<fp16, [384]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55264448)))];
563
  tensor<fp16, [1, 384, 1, 1]> current_value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = var_763, groups = var_713, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = var_761, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor<string, []>("current_value_cast_fp16")];
564
- tensor<fp16, [1, 384, 1, 448]> var_770_cast_fp16 = mul(x = current_key_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_770_cast_fp16")];
565
- tensor<fp16, [1, 384, 1, 448]> var_772_cast_fp16 = mul(x = var_47_cast_fp16_3, y = var_129_cast_fp16)[name = tensor<string, []>("op_772_cast_fp16")];
566
- tensor<fp16, [1, 384, 1, 448]> key_13_cast_fp16 = add(x = var_770_cast_fp16, y = var_772_cast_fp16)[name = tensor<string, []>("key_13_cast_fp16")];
567
- tensor<fp16, [1, 384, 1, 448]> var_774_cast_fp16 = mul(x = current_value_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_774_cast_fp16")];
568
- tensor<fp16, [1, 384, 1, 448]> var_776_cast_fp16 = mul(x = var_54_cast_fp16_3, y = var_129_cast_fp16)[name = tensor<string, []>("op_776_cast_fp16")];
569
- tensor<fp16, [1, 384, 1, 448]> value_13_cast_fp16 = add(x = var_774_cast_fp16, y = var_776_cast_fp16)[name = tensor<string, []>("value_13_cast_fp16")];
570
  tensor<int32, [4]> var_779 = const()[name = tensor<string, []>("op_779"), val = tensor<int32, [4]>([1, 6, 64, -1])];
571
  tensor<fp16, [1, 6, 64, 1]> var_780_cast_fp16 = reshape(shape = var_779, x = query_13_cast_fp16)[name = tensor<string, []>("op_780_cast_fp16")];
572
  tensor<fp16, []> var_781_to_fp16 = const()[name = tensor<string, []>("op_781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
573
  tensor<fp16, [1, 6, 64, 1]> var_782_cast_fp16 = mul(x = var_780_cast_fp16, y = var_781_to_fp16)[name = tensor<string, []>("op_782_cast_fp16")];
574
  tensor<int32, [4]> var_783 = const()[name = tensor<string, []>("op_783"), val = tensor<int32, [4]>([1, 6, 64, -1])];
575
- tensor<fp16, [1, 6, 64, 448]> var_784_cast_fp16 = reshape(shape = var_783, x = key_13_cast_fp16)[name = tensor<string, []>("op_784_cast_fp16")];
576
  tensor<bool, []> mh_w_19_transpose_x_0 = const()[name = tensor<string, []>("mh_w_19_transpose_x_0"), val = tensor<bool, []>(true)];
577
  tensor<bool, []> mh_w_19_transpose_y_0 = const()[name = tensor<string, []>("mh_w_19_transpose_y_0"), val = tensor<bool, []>(false)];
578
- tensor<fp16, [1, 6, 1, 448]> mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_782_cast_fp16, y = var_784_cast_fp16)[name = tensor<string, []>("mh_w_19_cast_fp16")];
579
- tensor<fp16, [1, 6, 1, 448]> mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_21_cast_fp16")];
580
- tensor<fp16, [1, 6, 1, 448]> var_792_cast_fp16 = softmax(axis = var_706, x = mh_w_21_cast_fp16)[name = tensor<string, []>("op_792_cast_fp16")];
581
  tensor<int32, [4]> var_793 = const()[name = tensor<string, []>("op_793"), val = tensor<int32, [4]>([1, 6, 64, -1])];
582
- tensor<fp16, [1, 6, 64, 448]> var_794_cast_fp16 = reshape(shape = var_793, x = value_13_cast_fp16)[name = tensor<string, []>("op_794_cast_fp16")];
583
  tensor<bool, []> attn_13_transpose_x_0 = const()[name = tensor<string, []>("attn_13_transpose_x_0"), val = tensor<bool, []>(false)];
584
  tensor<bool, []> attn_13_transpose_y_0 = const()[name = tensor<string, []>("attn_13_transpose_y_0"), val = tensor<bool, []>(true)];
585
  tensor<fp16, [1, 6, 64, 1]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_794_cast_fp16, y = var_792_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")];
 
1
  program(1.0)
2
  [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}, {"coremltools-component-torch", "2.1.2"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.1"}})]
3
  {
4
+ func main<ios17>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 224]> decoder_key_padding_mask, tensor<fp16, [1, 384, 1, 1500]> encoder_output_embeds, tensor<int32, [1]> input_ids, tensor<fp16, [1, 1536, 1, 224]> key_cache, tensor<fp16, [1, 224]> kv_cache_update_mask, tensor<fp16, [1, 1536, 1, 224]> value_cache) {
5
  tensor<int32, []> var_24_axis_0 = const()[name = tensor<string, []>("op_24_axis_0"), val = tensor<int32, []>(0)];
6
  tensor<int32, []> var_24_batch_dims_0 = const()[name = tensor<string, []>("op_24_batch_dims_0"), val = tensor<int32, []>(0)];
7
  tensor<bool, []> var_24_validate_indices_0 = const()[name = tensor<string, []>("op_24_validate_indices_0"), val = tensor<bool, []>(false)];
 
21
  tensor<fp16, [1, 384, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_42_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")];
22
  tensor<int32, [4]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [4]>([384, 384, 384, 384])];
23
  tensor<int32, []> var_47_axis_0 = const()[name = tensor<string, []>("op_47_axis_0"), val = tensor<int32, []>(1)];
24
+ tensor<fp16, [1, 384, 1, 224]> var_47_cast_fp16_0, tensor<fp16, [1, 384, 1, 224]> var_47_cast_fp16_1, tensor<fp16, [1, 384, 1, 224]> var_47_cast_fp16_2, tensor<fp16, [1, 384, 1, 224]> var_47_cast_fp16_3 = split(axis = var_47_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor<string, []>("op_47_cast_fp16")];
25
  tensor<int32, [4]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [4]>([384, 384, 384, 384])];
26
  tensor<int32, []> var_54_axis_0 = const()[name = tensor<string, []>("op_54_axis_0"), val = tensor<int32, []>(1)];
27
+ tensor<fp16, [1, 384, 1, 224]> var_54_cast_fp16_0, tensor<fp16, [1, 384, 1, 224]> var_54_cast_fp16_1, tensor<fp16, [1, 384, 1, 224]> var_54_cast_fp16_2, tensor<fp16, [1, 384, 1, 224]> var_54_cast_fp16_3 = split(axis = var_54_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor<string, []>("op_54_cast_fp16")];
28
  tensor<int32, []> var_64 = const()[name = tensor<string, []>("op_64"), val = tensor<int32, []>(3)];
29
  tensor<int32, []> var_71 = const()[name = tensor<string, []>("op_71"), val = tensor<int32, []>(1)];
30
  tensor<bool, []> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<bool, []>(true)];
 
66
  tensor<fp16, [384]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41064896)))];
67
  tensor<fp16, [1, 384, 1, 1]> current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = var_121, groups = var_71, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = var_119, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("current_value_1_cast_fp16")];
68
  tensor<int32, [1]> var_125_axes_0 = const()[name = tensor<string, []>("op_125_axes_0"), val = tensor<int32, [1]>([1])];
69
+ tensor<fp16, [1, 1, 224]> var_125_cast_fp16 = expand_dims(axes = var_125_axes_0, x = kv_cache_update_mask)[name = tensor<string, []>("op_125_cast_fp16")];
70
  tensor<int32, [1]> var_126_axes_0 = const()[name = tensor<string, []>("op_126_axes_0"), val = tensor<int32, [1]>([2])];
71
+ tensor<fp16, [1, 1, 1, 224]> var_126_cast_fp16 = expand_dims(axes = var_126_axes_0, x = var_125_cast_fp16)[name = tensor<string, []>("op_126_cast_fp16")];
72
+ tensor<fp16, [1, 384, 1, 224]> var_128_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_128_cast_fp16")];
73
  tensor<fp16, []> var_65_to_fp16 = const()[name = tensor<string, []>("op_65_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
74
+ tensor<fp16, [1, 1, 1, 224]> var_129_cast_fp16 = sub(x = var_65_to_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_129_cast_fp16")];
75
+ tensor<fp16, [1, 384, 1, 224]> var_130_cast_fp16 = mul(x = var_47_cast_fp16_0, y = var_129_cast_fp16)[name = tensor<string, []>("op_130_cast_fp16")];
76
+ tensor<fp16, [1, 384, 1, 224]> key_1_cast_fp16 = add(x = var_128_cast_fp16, y = var_130_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")];
77
+ tensor<fp16, [1, 384, 1, 224]> var_132_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_132_cast_fp16")];
78
+ tensor<fp16, [1, 384, 1, 224]> var_134_cast_fp16 = mul(x = var_54_cast_fp16_0, y = var_129_cast_fp16)[name = tensor<string, []>("op_134_cast_fp16")];
79
+ tensor<fp16, [1, 384, 1, 224]> value_1_cast_fp16 = add(x = var_132_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")];
80
  tensor<int32, [4]> var_137 = const()[name = tensor<string, []>("op_137"), val = tensor<int32, [4]>([1, 6, 64, -1])];
81
  tensor<fp16, [1, 6, 64, 1]> var_138_cast_fp16 = reshape(shape = var_137, x = query_1_cast_fp16)[name = tensor<string, []>("op_138_cast_fp16")];
82
  tensor<fp16, []> var_139_to_fp16 = const()[name = tensor<string, []>("op_139_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
83
  tensor<fp16, [1, 6, 64, 1]> var_140_cast_fp16 = mul(x = var_138_cast_fp16, y = var_139_to_fp16)[name = tensor<string, []>("op_140_cast_fp16")];
84
  tensor<int32, [4]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [4]>([1, 6, 64, -1])];
85
+ tensor<fp16, [1, 6, 64, 224]> var_142_cast_fp16 = reshape(shape = var_141, x = key_1_cast_fp16)[name = tensor<string, []>("op_142_cast_fp16")];
86
  tensor<bool, []> mh_w_1_transpose_x_0 = const()[name = tensor<string, []>("mh_w_1_transpose_x_0"), val = tensor<bool, []>(true)];
87
  tensor<bool, []> mh_w_1_transpose_y_0 = const()[name = tensor<string, []>("mh_w_1_transpose_y_0"), val = tensor<bool, []>(false)];
88
+ tensor<fp16, [1, 6, 1, 224]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_140_cast_fp16, y = var_142_cast_fp16)[name = tensor<string, []>("mh_w_1_cast_fp16")];
89
  tensor<int32, [1]> var_146_axes_0 = const()[name = tensor<string, []>("op_146_axes_0"), val = tensor<int32, [1]>([1])];
90
+ tensor<fp16, [1, 1, 224]> var_146_cast_fp16 = expand_dims(axes = var_146_axes_0, x = decoder_key_padding_mask)[name = tensor<string, []>("op_146_cast_fp16")];
91
  tensor<int32, [1]> var_147_axes_0 = const()[name = tensor<string, []>("op_147_axes_0"), val = tensor<int32, [1]>([2])];
92
+ tensor<fp16, [1, 1, 1, 224]> var_147_cast_fp16 = expand_dims(axes = var_147_axes_0, x = var_146_cast_fp16)[name = tensor<string, []>("op_147_cast_fp16")];
93
+ tensor<fp16, [1, 6, 1, 224]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")];
94
+ tensor<fp16, [1, 6, 1, 224]> var_150_cast_fp16 = softmax(axis = var_64, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_150_cast_fp16")];
95
  tensor<int32, [4]> var_151 = const()[name = tensor<string, []>("op_151"), val = tensor<int32, [4]>([1, 6, 64, -1])];
96
+ tensor<fp16, [1, 6, 64, 224]> var_152_cast_fp16 = reshape(shape = var_151, x = value_1_cast_fp16)[name = tensor<string, []>("op_152_cast_fp16")];
97
  tensor<bool, []> attn_1_transpose_x_0 = const()[name = tensor<string, []>("attn_1_transpose_x_0"), val = tensor<bool, []>(false)];
98
  tensor<bool, []> attn_1_transpose_y_0 = const()[name = tensor<string, []>("attn_1_transpose_y_0"), val = tensor<bool, []>(true)];
99
  tensor<fp16, [1, 6, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_152_cast_fp16, y = var_150_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")];
 
237
  tensor<fp16, [384, 384, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45503104)))];
238
  tensor<fp16, [384]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45798080)))];
239
  tensor<fp16, [1, 384, 1, 1]> current_value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = var_335, groups = var_285, pad = current_value_3_pad_0, pad_type = current_value_3_pad_type_0, strides = var_333, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = tensor<string, []>("current_value_3_cast_fp16")];
240
+ tensor<fp16, [1, 384, 1, 224]> var_342_cast_fp16 = mul(x = current_key_3_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_342_cast_fp16")];
241
+ tensor<fp16, [1, 384, 1, 224]> var_344_cast_fp16 = mul(x = var_47_cast_fp16_1, y = var_129_cast_fp16)[name = tensor<string, []>("op_344_cast_fp16")];
242
+ tensor<fp16, [1, 384, 1, 224]> key_5_cast_fp16 = add(x = var_342_cast_fp16, y = var_344_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")];
243
+ tensor<fp16, [1, 384, 1, 224]> var_346_cast_fp16 = mul(x = current_value_3_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_346_cast_fp16")];
244
+ tensor<fp16, [1, 384, 1, 224]> var_348_cast_fp16 = mul(x = var_54_cast_fp16_1, y = var_129_cast_fp16)[name = tensor<string, []>("op_348_cast_fp16")];
245
+ tensor<fp16, [1, 384, 1, 224]> value_5_cast_fp16 = add(x = var_346_cast_fp16, y = var_348_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")];
246
  tensor<int32, [4]> var_351 = const()[name = tensor<string, []>("op_351"), val = tensor<int32, [4]>([1, 6, 64, -1])];
247
  tensor<fp16, [1, 6, 64, 1]> var_352_cast_fp16 = reshape(shape = var_351, x = query_5_cast_fp16)[name = tensor<string, []>("op_352_cast_fp16")];
248
  tensor<fp16, []> var_353_to_fp16 = const()[name = tensor<string, []>("op_353_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
249
  tensor<fp16, [1, 6, 64, 1]> var_354_cast_fp16 = mul(x = var_352_cast_fp16, y = var_353_to_fp16)[name = tensor<string, []>("op_354_cast_fp16")];
250
  tensor<int32, [4]> var_355 = const()[name = tensor<string, []>("op_355"), val = tensor<int32, [4]>([1, 6, 64, -1])];
251
+ tensor<fp16, [1, 6, 64, 224]> var_356_cast_fp16 = reshape(shape = var_355, x = key_5_cast_fp16)[name = tensor<string, []>("op_356_cast_fp16")];
252
  tensor<bool, []> mh_w_7_transpose_x_0 = const()[name = tensor<string, []>("mh_w_7_transpose_x_0"), val = tensor<bool, []>(true)];
253
  tensor<bool, []> mh_w_7_transpose_y_0 = const()[name = tensor<string, []>("mh_w_7_transpose_y_0"), val = tensor<bool, []>(false)];
254
+ tensor<fp16, [1, 6, 1, 224]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_354_cast_fp16, y = var_356_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")];
255
+ tensor<fp16, [1, 6, 1, 224]> mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")];
256
+ tensor<fp16, [1, 6, 1, 224]> var_364_cast_fp16 = softmax(axis = var_278, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_364_cast_fp16")];
257
  tensor<int32, [4]> var_365 = const()[name = tensor<string, []>("op_365"), val = tensor<int32, [4]>([1, 6, 64, -1])];
258
+ tensor<fp16, [1, 6, 64, 224]> var_366_cast_fp16 = reshape(shape = var_365, x = value_5_cast_fp16)[name = tensor<string, []>("op_366_cast_fp16")];
259
  tensor<bool, []> attn_5_transpose_x_0 = const()[name = tensor<string, []>("attn_5_transpose_x_0"), val = tensor<bool, []>(false)];
260
  tensor<bool, []> attn_5_transpose_y_0 = const()[name = tensor<string, []>("attn_5_transpose_y_0"), val = tensor<bool, []>(true)];
261
  tensor<fp16, [1, 6, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_366_cast_fp16, y = var_364_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")];
 
399
  tensor<fp16, [384, 384, 1, 1]> layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50236288)))];
400
  tensor<fp16, [384]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50531264)))];
401
  tensor<fp16, [1, 384, 1, 1]> current_value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = var_549, groups = var_499, pad = current_value_5_pad_0, pad_type = current_value_5_pad_type_0, strides = var_547, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_25_cast_fp16)[name = tensor<string, []>("current_value_5_cast_fp16")];
402
+ tensor<fp16, [1, 384, 1, 224]> var_556_cast_fp16 = mul(x = current_key_5_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_556_cast_fp16")];
403
+ tensor<fp16, [1, 384, 1, 224]> var_558_cast_fp16 = mul(x = var_47_cast_fp16_2, y = var_129_cast_fp16)[name = tensor<string, []>("op_558_cast_fp16")];
404
+ tensor<fp16, [1, 384, 1, 224]> key_9_cast_fp16 = add(x = var_556_cast_fp16, y = var_558_cast_fp16)[name = tensor<string, []>("key_9_cast_fp16")];
405
+ tensor<fp16, [1, 384, 1, 224]> var_560_cast_fp16 = mul(x = current_value_5_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_560_cast_fp16")];
406
+ tensor<fp16, [1, 384, 1, 224]> var_562_cast_fp16 = mul(x = var_54_cast_fp16_2, y = var_129_cast_fp16)[name = tensor<string, []>("op_562_cast_fp16")];
407
+ tensor<fp16, [1, 384, 1, 224]> value_9_cast_fp16 = add(x = var_560_cast_fp16, y = var_562_cast_fp16)[name = tensor<string, []>("value_9_cast_fp16")];
408
  tensor<int32, [4]> var_565 = const()[name = tensor<string, []>("op_565"), val = tensor<int32, [4]>([1, 6, 64, -1])];
409
  tensor<fp16, [1, 6, 64, 1]> var_566_cast_fp16 = reshape(shape = var_565, x = query_9_cast_fp16)[name = tensor<string, []>("op_566_cast_fp16")];
410
  tensor<fp16, []> var_567_to_fp16 = const()[name = tensor<string, []>("op_567_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
411
  tensor<fp16, [1, 6, 64, 1]> var_568_cast_fp16 = mul(x = var_566_cast_fp16, y = var_567_to_fp16)[name = tensor<string, []>("op_568_cast_fp16")];
412
  tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, 6, 64, -1])];
413
+ tensor<fp16, [1, 6, 64, 224]> var_570_cast_fp16 = reshape(shape = var_569, x = key_9_cast_fp16)[name = tensor<string, []>("op_570_cast_fp16")];
414
  tensor<bool, []> mh_w_13_transpose_x_0 = const()[name = tensor<string, []>("mh_w_13_transpose_x_0"), val = tensor<bool, []>(true)];
415
  tensor<bool, []> mh_w_13_transpose_y_0 = const()[name = tensor<string, []>("mh_w_13_transpose_y_0"), val = tensor<bool, []>(false)];
416
+ tensor<fp16, [1, 6, 1, 224]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_568_cast_fp16, y = var_570_cast_fp16)[name = tensor<string, []>("mh_w_13_cast_fp16")];
417
+ tensor<fp16, [1, 6, 1, 224]> mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_15_cast_fp16")];
418
+ tensor<fp16, [1, 6, 1, 224]> var_578_cast_fp16 = softmax(axis = var_492, x = mh_w_15_cast_fp16)[name = tensor<string, []>("op_578_cast_fp16")];
419
  tensor<int32, [4]> var_579 = const()[name = tensor<string, []>("op_579"), val = tensor<int32, [4]>([1, 6, 64, -1])];
420
+ tensor<fp16, [1, 6, 64, 224]> var_580_cast_fp16 = reshape(shape = var_579, x = value_9_cast_fp16)[name = tensor<string, []>("op_580_cast_fp16")];
421
  tensor<bool, []> attn_9_transpose_x_0 = const()[name = tensor<string, []>("attn_9_transpose_x_0"), val = tensor<bool, []>(false)];
422
  tensor<bool, []> attn_9_transpose_y_0 = const()[name = tensor<string, []>("attn_9_transpose_y_0"), val = tensor<bool, []>(true)];
423
  tensor<fp16, [1, 6, 64, 1]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_580_cast_fp16, y = var_578_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")];
 
561
  tensor<fp16, [384, 384, 1, 1]> layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54969472)))];
562
  tensor<fp16, [384]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55264448)))];
563
  tensor<fp16, [1, 384, 1, 1]> current_value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = var_763, groups = var_713, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = var_761, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_37_cast_fp16)[name = tensor<string, []>("current_value_cast_fp16")];
564
+ tensor<fp16, [1, 384, 1, 224]> var_770_cast_fp16 = mul(x = current_key_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_770_cast_fp16")];
565
+ tensor<fp16, [1, 384, 1, 224]> var_772_cast_fp16 = mul(x = var_47_cast_fp16_3, y = var_129_cast_fp16)[name = tensor<string, []>("op_772_cast_fp16")];
566
+ tensor<fp16, [1, 384, 1, 224]> key_13_cast_fp16 = add(x = var_770_cast_fp16, y = var_772_cast_fp16)[name = tensor<string, []>("key_13_cast_fp16")];
567
+ tensor<fp16, [1, 384, 1, 224]> var_774_cast_fp16 = mul(x = current_value_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("op_774_cast_fp16")];
568
+ tensor<fp16, [1, 384, 1, 224]> var_776_cast_fp16 = mul(x = var_54_cast_fp16_3, y = var_129_cast_fp16)[name = tensor<string, []>("op_776_cast_fp16")];
569
+ tensor<fp16, [1, 384, 1, 224]> value_13_cast_fp16 = add(x = var_774_cast_fp16, y = var_776_cast_fp16)[name = tensor<string, []>("value_13_cast_fp16")];
570
  tensor<int32, [4]> var_779 = const()[name = tensor<string, []>("op_779"), val = tensor<int32, [4]>([1, 6, 64, -1])];
571
  tensor<fp16, [1, 6, 64, 1]> var_780_cast_fp16 = reshape(shape = var_779, x = query_13_cast_fp16)[name = tensor<string, []>("op_780_cast_fp16")];
572
  tensor<fp16, []> var_781_to_fp16 = const()[name = tensor<string, []>("op_781_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
573
  tensor<fp16, [1, 6, 64, 1]> var_782_cast_fp16 = mul(x = var_780_cast_fp16, y = var_781_to_fp16)[name = tensor<string, []>("op_782_cast_fp16")];
574
  tensor<int32, [4]> var_783 = const()[name = tensor<string, []>("op_783"), val = tensor<int32, [4]>([1, 6, 64, -1])];
575
+ tensor<fp16, [1, 6, 64, 224]> var_784_cast_fp16 = reshape(shape = var_783, x = key_13_cast_fp16)[name = tensor<string, []>("op_784_cast_fp16")];
576
  tensor<bool, []> mh_w_19_transpose_x_0 = const()[name = tensor<string, []>("mh_w_19_transpose_x_0"), val = tensor<bool, []>(true)];
577
  tensor<bool, []> mh_w_19_transpose_y_0 = const()[name = tensor<string, []>("mh_w_19_transpose_y_0"), val = tensor<bool, []>(false)];
578
+ tensor<fp16, [1, 6, 1, 224]> mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_782_cast_fp16, y = var_784_cast_fp16)[name = tensor<string, []>("mh_w_19_cast_fp16")];
579
+ tensor<fp16, [1, 6, 1, 224]> mh_w_21_cast_fp16 = add(x = mh_w_19_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("mh_w_21_cast_fp16")];
580
+ tensor<fp16, [1, 6, 1, 224]> var_792_cast_fp16 = softmax(axis = var_706, x = mh_w_21_cast_fp16)[name = tensor<string, []>("op_792_cast_fp16")];
581
  tensor<int32, [4]> var_793 = const()[name = tensor<string, []>("op_793"), val = tensor<int32, [4]>([1, 6, 64, -1])];
582
+ tensor<fp16, [1, 6, 64, 224]> var_794_cast_fp16 = reshape(shape = var_793, x = value_13_cast_fp16)[name = tensor<string, []>("op_794_cast_fp16")];
583
  tensor<bool, []> attn_13_transpose_x_0 = const()[name = tensor<string, []>("attn_13_transpose_x_0"), val = tensor<bool, []>(false)];
584
  tensor<bool, []> attn_13_transpose_y_0 = const()[name = tensor<string, []>("attn_13_transpose_y_0"), val = tensor<bool, []>(true)];
585
  tensor<fp16, [1, 6, 64, 1]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_794_cast_fp16, y = var_792_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")];
openai_whisper-tiny.en/TextDecoder.mlmodelc/weights/weight.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9a0eab78e8bcc54112f5d79d59018fb8de8b6327f19350fbc0833f1c55533be8
3
  size 59215664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92bfb389e3107e1dd84afa81bb1ecb416332b042736b313d673bf5c008e8b36d
3
  size 59215664