| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}})] | |
| { | |
| func main<ios16>(tensor<fp32, [1, 77]> input_ids) { | |
| tensor<int32, []> var_5 = const()[name = tensor<string, []>("op_5"), val = tensor<int32, []>(-1)]; | |
| tensor<bool, []> var_6 = const()[name = tensor<string, []>("op_6"), val = tensor<bool, []>(false)]; | |
| tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("int32")]; | |
| tensor<int32, []> inputs_embeds_axis_0 = const()[name = tensor<string, []>("inputs_embeds_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> inputs_embeds_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [49408, 768]> text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor<fp16, [49408, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<int32, [1, 77]> cast_1 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor<string, []>("cast_2")]; | |
| tensor<fp16, [1, 77, 768]> inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_1, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> position_embeddings_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [44352]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75890816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935232))), name = tensor<string, []>("position_embeddings_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 768]> input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16_palettized)[name = tensor<string, []>("input_3_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935424)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75937024)))]; | |
| tensor<fp16, []> var_15_to_fp16 = const()[name = tensor<string, []>("op_15_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75938624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381056))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381248)))]; | |
| tensor<fp16, [1, 77, 768]> linear_0_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; | |
| tensor<fp16, []> var_107_to_fp16 = const()[name = tensor<string, []>("op_107_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_107_to_fp16)[name = tensor<string, []>("tensor_5_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76382848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825280))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825472)))]; | |
| tensor<fp16, [1, 77, 768]> linear_1_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")]; | |
| tensor<int32, [4]> var_112 = const()[name = tensor<string, []>("op_112"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_113_cast_fp16 = reshape(shape = var_112, x = linear_1_cast_fp16)[name = tensor<string, []>("op_113_cast_fp16")]; | |
| tensor<int32, [4]> var_114_perm_0 = const()[name = tensor<string, []>("op_114_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76827072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269504))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269696)))]; | |
| tensor<fp16, [1, 77, 768]> linear_2_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")]; | |
| tensor<int32, [4]> var_119 = const()[name = tensor<string, []>("op_119"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_120_cast_fp16 = reshape(shape = var_119, x = linear_2_cast_fp16)[name = tensor<string, []>("op_120_cast_fp16")]; | |
| tensor<int32, [4]> var_121_perm_0 = const()[name = tensor<string, []>("op_121_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_128 = const()[name = tensor<string, []>("op_128"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_129_cast_fp16 = reshape(shape = var_128, x = tensor_5_cast_fp16)[name = tensor<string, []>("op_129_cast_fp16")]; | |
| tensor<int32, [4]> var_130_perm_0 = const()[name = tensor<string, []>("op_130_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_132 = const()[name = tensor<string, []>("op_132"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_130_cast_fp16 = transpose(perm = var_130_perm_0, x = var_129_cast_fp16)[name = tensor<string, []>("transpose_47")]; | |
| tensor<fp16, [12, 77, 64]> query_states_1_cast_fp16 = reshape(shape = var_132, x = var_130_cast_fp16)[name = tensor<string, []>("query_states_1_cast_fp16")]; | |
| tensor<int32, [3]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_114_cast_fp16 = transpose(perm = var_114_perm_0, x = var_113_cast_fp16)[name = tensor<string, []>("transpose_46")]; | |
| tensor<fp16, [12, 77, 64]> key_states_3_cast_fp16 = reshape(shape = var_134, x = var_114_cast_fp16)[name = tensor<string, []>("key_states_3_cast_fp16")]; | |
| tensor<int32, [3]> var_136 = const()[name = tensor<string, []>("op_136"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_121_cast_fp16 = transpose(perm = var_121_perm_0, x = var_120_cast_fp16)[name = tensor<string, []>("transpose_45")]; | |
| tensor<fp16, [12, 77, 64]> value_states_3_cast_fp16 = reshape(shape = var_136, x = var_121_cast_fp16)[name = tensor<string, []>("value_states_3_cast_fp16")]; | |
| tensor<bool, []> attn_weights_1_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_1_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_1_cast_fp16 = matmul(transpose_x = attn_weights_1_transpose_x_1, transpose_y = attn_weights_1_transpose_y_1, x = query_states_1_cast_fp16, y = key_states_3_cast_fp16)[name = tensor<string, []>("attn_weights_1_cast_fp16")]; | |
| tensor<int32, [4]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_142_cast_fp16 = reshape(shape = var_141, x = attn_weights_1_cast_fp16)[name = tensor<string, []>("op_142_cast_fp16")]; | |
| tensor<fp16, [1, 1, 77, 77]> op_56_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4447]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77271296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77275840))), name = tensor<string, []>("op_56_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_3_cast_fp16 = add(x = var_142_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_3_cast_fp16")]; | |
| tensor<int32, [3]> var_147 = const()[name = tensor<string, []>("op_147"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_5_cast_fp16 = reshape(shape = var_147, x = attn_weights_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_7_cast_fp16 = softmax(axis = var_5, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; | |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast_fp16, y = value_states_3_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")]; | |
| tensor<int32, [4]> var_152 = const()[name = tensor<string, []>("op_152"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_3_cast_fp16 = reshape(shape = var_152, x = attn_output_1_cast_fp16)[name = tensor<string, []>("attn_output_3_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_5_perm_0 = const()[name = tensor<string, []>("attn_output_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_155 = const()[name = tensor<string, []>("op_155"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_5_cast_fp16 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast_fp16)[name = tensor<string, []>("transpose_44")]; | |
| tensor<fp16, [1, 77, 768]> input_9_cast_fp16 = reshape(shape = var_155, x = attn_output_5_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77276032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718656)))]; | |
| tensor<fp16, [1, 77, 768]> linear_3_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_11_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; | |
| tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77720256)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77721856)))]; | |
| tensor<fp16, [1, 77, 768]> input_13_cast_fp16 = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77723456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79492992))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79493184)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_4_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; | |
| tensor<fp16, []> var_170_to_fp16 = const()[name = tensor<string, []>("op_170_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_171_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_170_to_fp16)[name = tensor<string, []>("op_171_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_172_cast_fp16 = sigmoid(x = var_171_cast_fp16)[name = tensor<string, []>("op_172_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_17_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_172_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79499392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81268928))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81269120)))]; | |
| tensor<fp16, [1, 77, 768]> linear_5_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_19_cast_fp16 = add(x = input_11_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81270720)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81272320)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81273920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81716352))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81716544)))]; | |
| tensor<fp16, [1, 77, 768]> linear_6_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")]; | |
| tensor<fp16, []> var_197_to_fp16 = const()[name = tensor<string, []>("op_197_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_11_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_197_to_fp16)[name = tensor<string, []>("tensor_11_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81718144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82160576))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82160768)))]; | |
| tensor<fp16, [1, 77, 768]> linear_7_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")]; | |
| tensor<int32, [4]> var_202 = const()[name = tensor<string, []>("op_202"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_203_cast_fp16 = reshape(shape = var_202, x = linear_7_cast_fp16)[name = tensor<string, []>("op_203_cast_fp16")]; | |
| tensor<int32, [4]> var_204_perm_0 = const()[name = tensor<string, []>("op_204_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82162368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82604800))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82604992)))]; | |
| tensor<fp16, [1, 77, 768]> linear_8_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> var_209 = const()[name = tensor<string, []>("op_209"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_210_cast_fp16 = reshape(shape = var_209, x = linear_8_cast_fp16)[name = tensor<string, []>("op_210_cast_fp16")]; | |
| tensor<int32, [4]> var_211_perm_0 = const()[name = tensor<string, []>("op_211_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_218 = const()[name = tensor<string, []>("op_218"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_219_cast_fp16 = reshape(shape = var_218, x = tensor_11_cast_fp16)[name = tensor<string, []>("op_219_cast_fp16")]; | |
| tensor<int32, [4]> var_220_perm_0 = const()[name = tensor<string, []>("op_220_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_222 = const()[name = tensor<string, []>("op_222"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_220_cast_fp16 = transpose(perm = var_220_perm_0, x = var_219_cast_fp16)[name = tensor<string, []>("transpose_43")]; | |
| tensor<fp16, [12, 77, 64]> query_states_3_cast_fp16 = reshape(shape = var_222, x = var_220_cast_fp16)[name = tensor<string, []>("query_states_3_cast_fp16")]; | |
| tensor<int32, [3]> var_224 = const()[name = tensor<string, []>("op_224"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_204_cast_fp16 = transpose(perm = var_204_perm_0, x = var_203_cast_fp16)[name = tensor<string, []>("transpose_42")]; | |
| tensor<fp16, [12, 77, 64]> key_states_7_cast_fp16 = reshape(shape = var_224, x = var_204_cast_fp16)[name = tensor<string, []>("key_states_7_cast_fp16")]; | |
| tensor<int32, [3]> var_226 = const()[name = tensor<string, []>("op_226"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_211_cast_fp16 = transpose(perm = var_211_perm_0, x = var_210_cast_fp16)[name = tensor<string, []>("transpose_41")]; | |
| tensor<fp16, [12, 77, 64]> value_states_7_cast_fp16 = reshape(shape = var_226, x = var_211_cast_fp16)[name = tensor<string, []>("value_states_7_cast_fp16")]; | |
| tensor<bool, []> attn_weights_7_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_7_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_7_cast_fp16 = matmul(transpose_x = attn_weights_7_transpose_x_1, transpose_y = attn_weights_7_transpose_y_1, x = query_states_3_cast_fp16, y = key_states_7_cast_fp16)[name = tensor<string, []>("attn_weights_7_cast_fp16")]; | |
| tensor<int32, [4]> var_231 = const()[name = tensor<string, []>("op_231"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_232_cast_fp16 = reshape(shape = var_231, x = attn_weights_7_cast_fp16)[name = tensor<string, []>("op_232_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_9_cast_fp16 = add(x = var_232_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_9_cast_fp16")]; | |
| tensor<int32, [3]> var_237 = const()[name = tensor<string, []>("op_237"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_21_cast_fp16 = reshape(shape = var_237, x = attn_weights_9_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_23_cast_fp16 = softmax(axis = var_5, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; | |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast_fp16, y = value_states_7_cast_fp16)[name = tensor<string, []>("attn_output_7_cast_fp16")]; | |
| tensor<int32, [4]> var_242 = const()[name = tensor<string, []>("op_242"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_9_cast_fp16 = reshape(shape = var_242, x = attn_output_7_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_11_perm_0 = const()[name = tensor<string, []>("attn_output_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_245 = const()[name = tensor<string, []>("op_245"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_11_cast_fp16 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_40")]; | |
| tensor<fp16, [1, 77, 768]> input_25_cast_fp16 = reshape(shape = var_245, x = attn_output_11_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82606592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83049024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83049216)))]; | |
| tensor<fp16, [1, 77, 768]> linear_9_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; | |
| tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83050816)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83052416)))]; | |
| tensor<fp16, [1, 77, 768]> input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83054016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84823552))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84823744)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_10_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; | |
| tensor<fp16, []> var_260_to_fp16 = const()[name = tensor<string, []>("op_260_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_261_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_260_to_fp16)[name = tensor<string, []>("op_261_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_262_cast_fp16 = sigmoid(x = var_261_cast_fp16)[name = tensor<string, []>("op_262_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_33_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_262_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84829952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86599488))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86599680)))]; | |
| tensor<fp16, [1, 77, 768]> linear_11_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_35_cast_fp16 = add(x = input_27_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86601280)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86602880)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86604480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87046912))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87047104)))]; | |
| tensor<fp16, [1, 77, 768]> linear_12_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")]; | |
| tensor<fp16, []> var_287_to_fp16 = const()[name = tensor<string, []>("op_287_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_17_cast_fp16 = mul(x = linear_12_cast_fp16, y = var_287_to_fp16)[name = tensor<string, []>("tensor_17_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87048704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87491136))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87491328)))]; | |
| tensor<fp16, [1, 77, 768]> linear_13_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")]; | |
| tensor<int32, [4]> var_292 = const()[name = tensor<string, []>("op_292"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_293_cast_fp16 = reshape(shape = var_292, x = linear_13_cast_fp16)[name = tensor<string, []>("op_293_cast_fp16")]; | |
| tensor<int32, [4]> var_294_perm_0 = const()[name = tensor<string, []>("op_294_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87492928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87935360))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87935552)))]; | |
| tensor<fp16, [1, 77, 768]> linear_14_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")]; | |
| tensor<int32, [4]> var_299 = const()[name = tensor<string, []>("op_299"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_300_cast_fp16 = reshape(shape = var_299, x = linear_14_cast_fp16)[name = tensor<string, []>("op_300_cast_fp16")]; | |
| tensor<int32, [4]> var_301_perm_0 = const()[name = tensor<string, []>("op_301_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_308 = const()[name = tensor<string, []>("op_308"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_309_cast_fp16 = reshape(shape = var_308, x = tensor_17_cast_fp16)[name = tensor<string, []>("op_309_cast_fp16")]; | |
| tensor<int32, [4]> var_310_perm_0 = const()[name = tensor<string, []>("op_310_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_312 = const()[name = tensor<string, []>("op_312"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_310_cast_fp16 = transpose(perm = var_310_perm_0, x = var_309_cast_fp16)[name = tensor<string, []>("transpose_39")]; | |
| tensor<fp16, [12, 77, 64]> query_states_5_cast_fp16 = reshape(shape = var_312, x = var_310_cast_fp16)[name = tensor<string, []>("query_states_5_cast_fp16")]; | |
| tensor<int32, [3]> var_314 = const()[name = tensor<string, []>("op_314"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_294_cast_fp16 = transpose(perm = var_294_perm_0, x = var_293_cast_fp16)[name = tensor<string, []>("transpose_38")]; | |
| tensor<fp16, [12, 77, 64]> key_states_11_cast_fp16 = reshape(shape = var_314, x = var_294_cast_fp16)[name = tensor<string, []>("key_states_11_cast_fp16")]; | |
| tensor<int32, [3]> var_316 = const()[name = tensor<string, []>("op_316"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_301_cast_fp16 = transpose(perm = var_301_perm_0, x = var_300_cast_fp16)[name = tensor<string, []>("transpose_37")]; | |
| tensor<fp16, [12, 77, 64]> value_states_11_cast_fp16 = reshape(shape = var_316, x = var_301_cast_fp16)[name = tensor<string, []>("value_states_11_cast_fp16")]; | |
| tensor<bool, []> attn_weights_13_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_13_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_13_cast_fp16 = matmul(transpose_x = attn_weights_13_transpose_x_1, transpose_y = attn_weights_13_transpose_y_1, x = query_states_5_cast_fp16, y = key_states_11_cast_fp16)[name = tensor<string, []>("attn_weights_13_cast_fp16")]; | |
| tensor<int32, [4]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_322_cast_fp16 = reshape(shape = var_321, x = attn_weights_13_cast_fp16)[name = tensor<string, []>("op_322_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_15_cast_fp16 = add(x = var_322_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_15_cast_fp16")]; | |
| tensor<int32, [3]> var_327 = const()[name = tensor<string, []>("op_327"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_37_cast_fp16 = reshape(shape = var_327, x = attn_weights_15_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_39_cast_fp16 = softmax(axis = var_5, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; | |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast_fp16, y = value_states_11_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")]; | |
| tensor<int32, [4]> var_332 = const()[name = tensor<string, []>("op_332"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_15_cast_fp16 = reshape(shape = var_332, x = attn_output_13_cast_fp16)[name = tensor<string, []>("attn_output_15_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_17_perm_0 = const()[name = tensor<string, []>("attn_output_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_335 = const()[name = tensor<string, []>("op_335"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_17_cast_fp16 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast_fp16)[name = tensor<string, []>("transpose_36")]; | |
| tensor<fp16, [1, 77, 768]> input_41_cast_fp16 = reshape(shape = var_335, x = attn_output_17_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87937152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88379584))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88379776)))]; | |
| tensor<fp16, [1, 77, 768]> linear_15_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_43_cast_fp16 = add(x = input_35_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; | |
| tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88381376)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88382976)))]; | |
| tensor<fp16, [1, 77, 768]> input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88384576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90154112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90154304)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_16_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; | |
| tensor<fp16, []> var_350_to_fp16 = const()[name = tensor<string, []>("op_350_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_351_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_350_to_fp16)[name = tensor<string, []>("op_351_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_352_cast_fp16 = sigmoid(x = var_351_cast_fp16)[name = tensor<string, []>("op_352_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_49_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_352_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90160512))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91930048))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91930240)))]; | |
| tensor<fp16, [1, 77, 768]> linear_17_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91931840)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91933440)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91935040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92377472))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92377664)))]; | |
| tensor<fp16, [1, 77, 768]> linear_18_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")]; | |
| tensor<fp16, []> var_377_to_fp16 = const()[name = tensor<string, []>("op_377_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_23_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_377_to_fp16)[name = tensor<string, []>("tensor_23_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92379264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92821696))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92821888)))]; | |
| tensor<fp16, [1, 77, 768]> linear_19_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")]; | |
| tensor<int32, [4]> var_382 = const()[name = tensor<string, []>("op_382"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_383_cast_fp16 = reshape(shape = var_382, x = linear_19_cast_fp16)[name = tensor<string, []>("op_383_cast_fp16")]; | |
| tensor<int32, [4]> var_384_perm_0 = const()[name = tensor<string, []>("op_384_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92823488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93265920))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93266112)))]; | |
| tensor<fp16, [1, 77, 768]> linear_20_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> var_389 = const()[name = tensor<string, []>("op_389"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_390_cast_fp16 = reshape(shape = var_389, x = linear_20_cast_fp16)[name = tensor<string, []>("op_390_cast_fp16")]; | |
| tensor<int32, [4]> var_391_perm_0 = const()[name = tensor<string, []>("op_391_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_398 = const()[name = tensor<string, []>("op_398"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_399_cast_fp16 = reshape(shape = var_398, x = tensor_23_cast_fp16)[name = tensor<string, []>("op_399_cast_fp16")]; | |
| tensor<int32, [4]> var_400_perm_0 = const()[name = tensor<string, []>("op_400_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_402 = const()[name = tensor<string, []>("op_402"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_400_cast_fp16 = transpose(perm = var_400_perm_0, x = var_399_cast_fp16)[name = tensor<string, []>("transpose_35")]; | |
| tensor<fp16, [12, 77, 64]> query_states_7_cast_fp16 = reshape(shape = var_402, x = var_400_cast_fp16)[name = tensor<string, []>("query_states_7_cast_fp16")]; | |
| tensor<int32, [3]> var_404 = const()[name = tensor<string, []>("op_404"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_384_cast_fp16 = transpose(perm = var_384_perm_0, x = var_383_cast_fp16)[name = tensor<string, []>("transpose_34")]; | |
| tensor<fp16, [12, 77, 64]> key_states_15_cast_fp16 = reshape(shape = var_404, x = var_384_cast_fp16)[name = tensor<string, []>("key_states_15_cast_fp16")]; | |
| tensor<int32, [3]> var_406 = const()[name = tensor<string, []>("op_406"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_391_cast_fp16 = transpose(perm = var_391_perm_0, x = var_390_cast_fp16)[name = tensor<string, []>("transpose_33")]; | |
| tensor<fp16, [12, 77, 64]> value_states_15_cast_fp16 = reshape(shape = var_406, x = var_391_cast_fp16)[name = tensor<string, []>("value_states_15_cast_fp16")]; | |
| tensor<bool, []> attn_weights_19_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_19_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_19_cast_fp16 = matmul(transpose_x = attn_weights_19_transpose_x_1, transpose_y = attn_weights_19_transpose_y_1, x = query_states_7_cast_fp16, y = key_states_15_cast_fp16)[name = tensor<string, []>("attn_weights_19_cast_fp16")]; | |
| tensor<int32, [4]> var_411 = const()[name = tensor<string, []>("op_411"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_412_cast_fp16 = reshape(shape = var_411, x = attn_weights_19_cast_fp16)[name = tensor<string, []>("op_412_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_21_cast_fp16 = add(x = var_412_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_21_cast_fp16")]; | |
| tensor<int32, [3]> var_417 = const()[name = tensor<string, []>("op_417"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_53_cast_fp16 = reshape(shape = var_417, x = attn_weights_21_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_55_cast_fp16 = softmax(axis = var_5, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; | |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast_fp16, y = value_states_15_cast_fp16)[name = tensor<string, []>("attn_output_19_cast_fp16")]; | |
| tensor<int32, [4]> var_422 = const()[name = tensor<string, []>("op_422"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_21_cast_fp16 = reshape(shape = var_422, x = attn_output_19_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_23_perm_0 = const()[name = tensor<string, []>("attn_output_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_425 = const()[name = tensor<string, []>("op_425"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_23_cast_fp16 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_32")]; | |
| tensor<fp16, [1, 77, 768]> input_57_cast_fp16 = reshape(shape = var_425, x = attn_output_23_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93267712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93710144))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93710336)))]; | |
| tensor<fp16, [1, 77, 768]> linear_21_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_59_cast_fp16 = add(x = input_51_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; | |
| tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93711936)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93713536)))]; | |
| tensor<fp16, [1, 77, 768]> input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93715136))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95484672))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95484864)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_22_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; | |
| tensor<fp16, []> var_440_to_fp16 = const()[name = tensor<string, []>("op_440_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_441_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_440_to_fp16)[name = tensor<string, []>("op_441_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_442_cast_fp16 = sigmoid(x = var_441_cast_fp16)[name = tensor<string, []>("op_442_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_65_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_442_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95491072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97260608))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97260800)))]; | |
| tensor<fp16, [1, 77, 768]> linear_23_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_67_cast_fp16 = add(x = input_59_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97262400)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97264000)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97265600))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97708032))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97708224)))]; | |
| tensor<fp16, [1, 77, 768]> linear_24_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")]; | |
| tensor<fp16, []> var_467_to_fp16 = const()[name = tensor<string, []>("op_467_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_29_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_467_to_fp16)[name = tensor<string, []>("tensor_29_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97709824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98152256))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98152448)))]; | |
| tensor<fp16, [1, 77, 768]> linear_25_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")]; | |
| tensor<int32, [4]> var_472 = const()[name = tensor<string, []>("op_472"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_473_cast_fp16 = reshape(shape = var_472, x = linear_25_cast_fp16)[name = tensor<string, []>("op_473_cast_fp16")]; | |
| tensor<int32, [4]> var_474_perm_0 = const()[name = tensor<string, []>("op_474_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98154048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98596480))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98596672)))]; | |
| tensor<fp16, [1, 77, 768]> linear_26_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")]; | |
| tensor<int32, [4]> var_479 = const()[name = tensor<string, []>("op_479"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_480_cast_fp16 = reshape(shape = var_479, x = linear_26_cast_fp16)[name = tensor<string, []>("op_480_cast_fp16")]; | |
| tensor<int32, [4]> var_481_perm_0 = const()[name = tensor<string, []>("op_481_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_488 = const()[name = tensor<string, []>("op_488"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_489_cast_fp16 = reshape(shape = var_488, x = tensor_29_cast_fp16)[name = tensor<string, []>("op_489_cast_fp16")]; | |
| tensor<int32, [4]> var_490_perm_0 = const()[name = tensor<string, []>("op_490_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_492 = const()[name = tensor<string, []>("op_492"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_490_cast_fp16 = transpose(perm = var_490_perm_0, x = var_489_cast_fp16)[name = tensor<string, []>("transpose_31")]; | |
| tensor<fp16, [12, 77, 64]> query_states_9_cast_fp16 = reshape(shape = var_492, x = var_490_cast_fp16)[name = tensor<string, []>("query_states_9_cast_fp16")]; | |
| tensor<int32, [3]> var_494 = const()[name = tensor<string, []>("op_494"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_474_cast_fp16 = transpose(perm = var_474_perm_0, x = var_473_cast_fp16)[name = tensor<string, []>("transpose_30")]; | |
| tensor<fp16, [12, 77, 64]> key_states_19_cast_fp16 = reshape(shape = var_494, x = var_474_cast_fp16)[name = tensor<string, []>("key_states_19_cast_fp16")]; | |
| tensor<int32, [3]> var_496 = const()[name = tensor<string, []>("op_496"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_481_cast_fp16 = transpose(perm = var_481_perm_0, x = var_480_cast_fp16)[name = tensor<string, []>("transpose_29")]; | |
| tensor<fp16, [12, 77, 64]> value_states_19_cast_fp16 = reshape(shape = var_496, x = var_481_cast_fp16)[name = tensor<string, []>("value_states_19_cast_fp16")]; | |
| tensor<bool, []> attn_weights_25_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_25_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_25_cast_fp16 = matmul(transpose_x = attn_weights_25_transpose_x_1, transpose_y = attn_weights_25_transpose_y_1, x = query_states_9_cast_fp16, y = key_states_19_cast_fp16)[name = tensor<string, []>("attn_weights_25_cast_fp16")]; | |
| tensor<int32, [4]> var_501 = const()[name = tensor<string, []>("op_501"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_502_cast_fp16 = reshape(shape = var_501, x = attn_weights_25_cast_fp16)[name = tensor<string, []>("op_502_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_27_cast_fp16 = add(x = var_502_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_27_cast_fp16")]; | |
| tensor<int32, [3]> var_507 = const()[name = tensor<string, []>("op_507"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_69_cast_fp16 = reshape(shape = var_507, x = attn_weights_27_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_71_cast_fp16 = softmax(axis = var_5, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; | |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast_fp16, y = value_states_19_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")]; | |
| tensor<int32, [4]> var_512 = const()[name = tensor<string, []>("op_512"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_27_cast_fp16 = reshape(shape = var_512, x = attn_output_25_cast_fp16)[name = tensor<string, []>("attn_output_27_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_29_perm_0 = const()[name = tensor<string, []>("attn_output_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_515 = const()[name = tensor<string, []>("op_515"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_29_cast_fp16 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast_fp16)[name = tensor<string, []>("transpose_28")]; | |
| tensor<fp16, [1, 77, 768]> input_73_cast_fp16 = reshape(shape = var_515, x = attn_output_29_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98598272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99040704))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99040896)))]; | |
| tensor<fp16, [1, 77, 768]> linear_27_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; | |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99042496)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99044096)))]; | |
| tensor<fp16, [1, 77, 768]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99045696))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100815232))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100815424)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_28_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")]; | |
| tensor<fp16, []> var_530_to_fp16 = const()[name = tensor<string, []>("op_530_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_531_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_530_to_fp16)[name = tensor<string, []>("op_531_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_532_cast_fp16 = sigmoid(x = var_531_cast_fp16)[name = tensor<string, []>("op_532_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_81_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_532_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100821632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102591168))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102591360)))]; | |
| tensor<fp16, [1, 77, 768]> linear_29_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_83_cast_fp16 = add(x = input_75_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102592960)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102594560)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("hidden_states_31_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102596160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103038592))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103038784)))]; | |
| tensor<fp16, [1, 77, 768]> linear_30_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")]; | |
| tensor<fp16, []> var_557_to_fp16 = const()[name = tensor<string, []>("op_557_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_35_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_557_to_fp16)[name = tensor<string, []>("tensor_35_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103040384))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103482816))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103483008)))]; | |
| tensor<fp16, [1, 77, 768]> linear_31_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")]; | |
| tensor<int32, [4]> var_562 = const()[name = tensor<string, []>("op_562"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_563_cast_fp16 = reshape(shape = var_562, x = linear_31_cast_fp16)[name = tensor<string, []>("op_563_cast_fp16")]; | |
| tensor<int32, [4]> var_564_perm_0 = const()[name = tensor<string, []>("op_564_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103484608))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103927040))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103927232)))]; | |
| tensor<fp16, [1, 77, 768]> linear_32_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")]; | |
| tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_570_cast_fp16 = reshape(shape = var_569, x = linear_32_cast_fp16)[name = tensor<string, []>("op_570_cast_fp16")]; | |
| tensor<int32, [4]> var_571_perm_0 = const()[name = tensor<string, []>("op_571_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_578 = const()[name = tensor<string, []>("op_578"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_579_cast_fp16 = reshape(shape = var_578, x = tensor_35_cast_fp16)[name = tensor<string, []>("op_579_cast_fp16")]; | |
| tensor<int32, [4]> var_580_perm_0 = const()[name = tensor<string, []>("op_580_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_582 = const()[name = tensor<string, []>("op_582"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_580_cast_fp16 = transpose(perm = var_580_perm_0, x = var_579_cast_fp16)[name = tensor<string, []>("transpose_27")]; | |
| tensor<fp16, [12, 77, 64]> query_states_11_cast_fp16 = reshape(shape = var_582, x = var_580_cast_fp16)[name = tensor<string, []>("query_states_11_cast_fp16")]; | |
| tensor<int32, [3]> var_584 = const()[name = tensor<string, []>("op_584"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_564_cast_fp16 = transpose(perm = var_564_perm_0, x = var_563_cast_fp16)[name = tensor<string, []>("transpose_26")]; | |
| tensor<fp16, [12, 77, 64]> key_states_23_cast_fp16 = reshape(shape = var_584, x = var_564_cast_fp16)[name = tensor<string, []>("key_states_23_cast_fp16")]; | |
| tensor<int32, [3]> var_586 = const()[name = tensor<string, []>("op_586"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_571_cast_fp16 = transpose(perm = var_571_perm_0, x = var_570_cast_fp16)[name = tensor<string, []>("transpose_25")]; | |
| tensor<fp16, [12, 77, 64]> value_states_23_cast_fp16 = reshape(shape = var_586, x = var_571_cast_fp16)[name = tensor<string, []>("value_states_23_cast_fp16")]; | |
| tensor<bool, []> attn_weights_31_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_31_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_31_cast_fp16 = matmul(transpose_x = attn_weights_31_transpose_x_1, transpose_y = attn_weights_31_transpose_y_1, x = query_states_11_cast_fp16, y = key_states_23_cast_fp16)[name = tensor<string, []>("attn_weights_31_cast_fp16")]; | |
| tensor<int32, [4]> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_592_cast_fp16 = reshape(shape = var_591, x = attn_weights_31_cast_fp16)[name = tensor<string, []>("op_592_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_33_cast_fp16 = add(x = var_592_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_33_cast_fp16")]; | |
| tensor<int32, [3]> var_597 = const()[name = tensor<string, []>("op_597"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_85_cast_fp16 = reshape(shape = var_597, x = attn_weights_33_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_87_cast_fp16 = softmax(axis = var_5, x = input_85_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; | |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast_fp16, y = value_states_23_cast_fp16)[name = tensor<string, []>("attn_output_31_cast_fp16")]; | |
| tensor<int32, [4]> var_602 = const()[name = tensor<string, []>("op_602"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_33_cast_fp16 = reshape(shape = var_602, x = attn_output_31_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_35_perm_0 = const()[name = tensor<string, []>("attn_output_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_605 = const()[name = tensor<string, []>("op_605"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_35_cast_fp16 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_24")]; | |
| tensor<fp16, [1, 77, 768]> input_89_cast_fp16 = reshape(shape = var_605, x = attn_output_35_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103928832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104371264))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104371456)))]; | |
| tensor<fp16, [1, 77, 768]> linear_33_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; | |
| tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104373056)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104374656)))]; | |
| tensor<fp16, [1, 77, 768]> input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104376256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106145792))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106145984)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_34_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")]; | |
| tensor<fp16, []> var_620_to_fp16 = const()[name = tensor<string, []>("op_620_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_621_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_620_to_fp16)[name = tensor<string, []>("op_621_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_622_cast_fp16 = sigmoid(x = var_621_cast_fp16)[name = tensor<string, []>("op_622_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_97_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_622_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106152192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107921728))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107921920)))]; | |
| tensor<fp16, [1, 77, 768]> linear_35_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107923520)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107925120)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("hidden_states_37_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107926720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108369152))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108369344)))]; | |
| tensor<fp16, [1, 77, 768]> linear_36_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")]; | |
| tensor<fp16, []> var_647_to_fp16 = const()[name = tensor<string, []>("op_647_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_41_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_647_to_fp16)[name = tensor<string, []>("tensor_41_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108370944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108813376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108813568)))]; | |
| tensor<fp16, [1, 77, 768]> linear_37_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")]; | |
| tensor<int32, [4]> var_652 = const()[name = tensor<string, []>("op_652"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_653_cast_fp16 = reshape(shape = var_652, x = linear_37_cast_fp16)[name = tensor<string, []>("op_653_cast_fp16")]; | |
| tensor<int32, [4]> var_654_perm_0 = const()[name = tensor<string, []>("op_654_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108815168))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109257600))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109257792)))]; | |
| tensor<fp16, [1, 77, 768]> linear_38_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")]; | |
| tensor<int32, [4]> var_659 = const()[name = tensor<string, []>("op_659"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_660_cast_fp16 = reshape(shape = var_659, x = linear_38_cast_fp16)[name = tensor<string, []>("op_660_cast_fp16")]; | |
| tensor<int32, [4]> var_661_perm_0 = const()[name = tensor<string, []>("op_661_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_668 = const()[name = tensor<string, []>("op_668"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_669_cast_fp16 = reshape(shape = var_668, x = tensor_41_cast_fp16)[name = tensor<string, []>("op_669_cast_fp16")]; | |
| tensor<int32, [4]> var_670_perm_0 = const()[name = tensor<string, []>("op_670_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_672 = const()[name = tensor<string, []>("op_672"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_670_cast_fp16 = transpose(perm = var_670_perm_0, x = var_669_cast_fp16)[name = tensor<string, []>("transpose_23")]; | |
| tensor<fp16, [12, 77, 64]> query_states_13_cast_fp16 = reshape(shape = var_672, x = var_670_cast_fp16)[name = tensor<string, []>("query_states_13_cast_fp16")]; | |
| tensor<int32, [3]> var_674 = const()[name = tensor<string, []>("op_674"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_654_cast_fp16 = transpose(perm = var_654_perm_0, x = var_653_cast_fp16)[name = tensor<string, []>("transpose_22")]; | |
| tensor<fp16, [12, 77, 64]> key_states_27_cast_fp16 = reshape(shape = var_674, x = var_654_cast_fp16)[name = tensor<string, []>("key_states_27_cast_fp16")]; | |
| tensor<int32, [3]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_661_cast_fp16 = transpose(perm = var_661_perm_0, x = var_660_cast_fp16)[name = tensor<string, []>("transpose_21")]; | |
| tensor<fp16, [12, 77, 64]> value_states_27_cast_fp16 = reshape(shape = var_676, x = var_661_cast_fp16)[name = tensor<string, []>("value_states_27_cast_fp16")]; | |
| tensor<bool, []> attn_weights_37_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_37_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_37_cast_fp16 = matmul(transpose_x = attn_weights_37_transpose_x_1, transpose_y = attn_weights_37_transpose_y_1, x = query_states_13_cast_fp16, y = key_states_27_cast_fp16)[name = tensor<string, []>("attn_weights_37_cast_fp16")]; | |
| tensor<int32, [4]> var_681 = const()[name = tensor<string, []>("op_681"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_682_cast_fp16 = reshape(shape = var_681, x = attn_weights_37_cast_fp16)[name = tensor<string, []>("op_682_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_39_cast_fp16 = add(x = var_682_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_39_cast_fp16")]; | |
| tensor<int32, [3]> var_687 = const()[name = tensor<string, []>("op_687"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_101_cast_fp16 = reshape(shape = var_687, x = attn_weights_39_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_103_cast_fp16 = softmax(axis = var_5, x = input_101_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; | |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast_fp16, y = value_states_27_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")]; | |
| tensor<int32, [4]> var_692 = const()[name = tensor<string, []>("op_692"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_39_cast_fp16 = reshape(shape = var_692, x = attn_output_37_cast_fp16)[name = tensor<string, []>("attn_output_39_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_41_perm_0 = const()[name = tensor<string, []>("attn_output_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_695 = const()[name = tensor<string, []>("op_695"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_41_cast_fp16 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast_fp16)[name = tensor<string, []>("transpose_20")]; | |
| tensor<fp16, [1, 77, 768]> input_105_cast_fp16 = reshape(shape = var_695, x = attn_output_41_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109259392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109701824))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109702016)))]; | |
| tensor<fp16, [1, 77, 768]> linear_39_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_107_cast_fp16 = add(x = input_99_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; | |
| tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109703616)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109705216)))]; | |
| tensor<fp16, [1, 77, 768]> input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109706816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111476352))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111476544)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_40_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")]; | |
| tensor<fp16, []> var_710_to_fp16 = const()[name = tensor<string, []>("op_710_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_711_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_710_to_fp16)[name = tensor<string, []>("op_711_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_712_cast_fp16 = sigmoid(x = var_711_cast_fp16)[name = tensor<string, []>("op_712_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_113_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_712_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111482752))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113252288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113252480)))]; | |
| tensor<fp16, [1, 77, 768]> linear_41_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_115_cast_fp16 = add(x = input_107_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113254080)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113255680)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113257280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113699712))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113699904)))]; | |
| tensor<fp16, [1, 77, 768]> linear_42_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")]; | |
| tensor<fp16, []> var_737_to_fp16 = const()[name = tensor<string, []>("op_737_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_47_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_737_to_fp16)[name = tensor<string, []>("tensor_47_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113701504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114143936))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114144128)))]; | |
| tensor<fp16, [1, 77, 768]> linear_43_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")]; | |
| tensor<int32, [4]> var_742 = const()[name = tensor<string, []>("op_742"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_743_cast_fp16 = reshape(shape = var_742, x = linear_43_cast_fp16)[name = tensor<string, []>("op_743_cast_fp16")]; | |
| tensor<int32, [4]> var_744_perm_0 = const()[name = tensor<string, []>("op_744_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114145728))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114588160))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114588352)))]; | |
| tensor<fp16, [1, 77, 768]> linear_44_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")]; | |
| tensor<int32, [4]> var_749 = const()[name = tensor<string, []>("op_749"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_750_cast_fp16 = reshape(shape = var_749, x = linear_44_cast_fp16)[name = tensor<string, []>("op_750_cast_fp16")]; | |
| tensor<int32, [4]> var_751_perm_0 = const()[name = tensor<string, []>("op_751_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_758 = const()[name = tensor<string, []>("op_758"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_759_cast_fp16 = reshape(shape = var_758, x = tensor_47_cast_fp16)[name = tensor<string, []>("op_759_cast_fp16")]; | |
| tensor<int32, [4]> var_760_perm_0 = const()[name = tensor<string, []>("op_760_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_762 = const()[name = tensor<string, []>("op_762"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_760_cast_fp16 = transpose(perm = var_760_perm_0, x = var_759_cast_fp16)[name = tensor<string, []>("transpose_19")]; | |
| tensor<fp16, [12, 77, 64]> query_states_15_cast_fp16 = reshape(shape = var_762, x = var_760_cast_fp16)[name = tensor<string, []>("query_states_15_cast_fp16")]; | |
| tensor<int32, [3]> var_764 = const()[name = tensor<string, []>("op_764"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_744_cast_fp16 = transpose(perm = var_744_perm_0, x = var_743_cast_fp16)[name = tensor<string, []>("transpose_18")]; | |
| tensor<fp16, [12, 77, 64]> key_states_31_cast_fp16 = reshape(shape = var_764, x = var_744_cast_fp16)[name = tensor<string, []>("key_states_31_cast_fp16")]; | |
| tensor<int32, [3]> var_766 = const()[name = tensor<string, []>("op_766"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_751_cast_fp16 = transpose(perm = var_751_perm_0, x = var_750_cast_fp16)[name = tensor<string, []>("transpose_17")]; | |
| tensor<fp16, [12, 77, 64]> value_states_31_cast_fp16 = reshape(shape = var_766, x = var_751_cast_fp16)[name = tensor<string, []>("value_states_31_cast_fp16")]; | |
| tensor<bool, []> attn_weights_43_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_43_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_43_cast_fp16 = matmul(transpose_x = attn_weights_43_transpose_x_1, transpose_y = attn_weights_43_transpose_y_1, x = query_states_15_cast_fp16, y = key_states_31_cast_fp16)[name = tensor<string, []>("attn_weights_43_cast_fp16")]; | |
| tensor<int32, [4]> var_771 = const()[name = tensor<string, []>("op_771"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_772_cast_fp16 = reshape(shape = var_771, x = attn_weights_43_cast_fp16)[name = tensor<string, []>("op_772_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_45_cast_fp16 = add(x = var_772_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_45_cast_fp16")]; | |
| tensor<int32, [3]> var_777 = const()[name = tensor<string, []>("op_777"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_117_cast_fp16 = reshape(shape = var_777, x = attn_weights_45_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_119_cast_fp16 = softmax(axis = var_5, x = input_117_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")]; | |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast_fp16, y = value_states_31_cast_fp16)[name = tensor<string, []>("attn_output_43_cast_fp16")]; | |
| tensor<int32, [4]> var_782 = const()[name = tensor<string, []>("op_782"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_45_cast_fp16 = reshape(shape = var_782, x = attn_output_43_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_47_perm_0 = const()[name = tensor<string, []>("attn_output_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_785 = const()[name = tensor<string, []>("op_785"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_47_cast_fp16 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_16")]; | |
| tensor<fp16, [1, 77, 768]> input_121_cast_fp16 = reshape(shape = var_785, x = attn_output_47_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114589952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115032384))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115032576)))]; | |
| tensor<fp16, [1, 77, 768]> linear_45_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")]; | |
| tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115034176)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115035776)))]; | |
| tensor<fp16, [1, 77, 768]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115037376))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116806912))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116807104)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_46_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")]; | |
| tensor<fp16, []> var_800_to_fp16 = const()[name = tensor<string, []>("op_800_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_801_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_800_to_fp16)[name = tensor<string, []>("op_801_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_802_cast_fp16 = sigmoid(x = var_801_cast_fp16)[name = tensor<string, []>("op_802_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_129_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_802_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116813312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118582848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118583040)))]; | |
| tensor<fp16, [1, 77, 768]> linear_47_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_131_cast_fp16 = add(x = input_123_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118584640)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118586240)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast_fp16)[name = tensor<string, []>("hidden_states_49_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118587840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119030272))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119030464)))]; | |
| tensor<fp16, [1, 77, 768]> linear_48_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")]; | |
| tensor<fp16, []> var_827_to_fp16 = const()[name = tensor<string, []>("op_827_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_53_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_827_to_fp16)[name = tensor<string, []>("tensor_53_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119032064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119474496))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119474688)))]; | |
| tensor<fp16, [1, 77, 768]> linear_49_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")]; | |
| tensor<int32, [4]> var_832 = const()[name = tensor<string, []>("op_832"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_833_cast_fp16 = reshape(shape = var_832, x = linear_49_cast_fp16)[name = tensor<string, []>("op_833_cast_fp16")]; | |
| tensor<int32, [4]> var_834_perm_0 = const()[name = tensor<string, []>("op_834_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119476288))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119918720))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119918912)))]; | |
| tensor<fp16, [1, 77, 768]> linear_50_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")]; | |
| tensor<int32, [4]> var_839 = const()[name = tensor<string, []>("op_839"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_840_cast_fp16 = reshape(shape = var_839, x = linear_50_cast_fp16)[name = tensor<string, []>("op_840_cast_fp16")]; | |
| tensor<int32, [4]> var_841_perm_0 = const()[name = tensor<string, []>("op_841_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_848 = const()[name = tensor<string, []>("op_848"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_849_cast_fp16 = reshape(shape = var_848, x = tensor_53_cast_fp16)[name = tensor<string, []>("op_849_cast_fp16")]; | |
| tensor<int32, [4]> var_850_perm_0 = const()[name = tensor<string, []>("op_850_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_852 = const()[name = tensor<string, []>("op_852"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_850_cast_fp16 = transpose(perm = var_850_perm_0, x = var_849_cast_fp16)[name = tensor<string, []>("transpose_15")]; | |
| tensor<fp16, [12, 77, 64]> query_states_17_cast_fp16 = reshape(shape = var_852, x = var_850_cast_fp16)[name = tensor<string, []>("query_states_17_cast_fp16")]; | |
| tensor<int32, [3]> var_854 = const()[name = tensor<string, []>("op_854"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_834_cast_fp16 = transpose(perm = var_834_perm_0, x = var_833_cast_fp16)[name = tensor<string, []>("transpose_14")]; | |
| tensor<fp16, [12, 77, 64]> key_states_35_cast_fp16 = reshape(shape = var_854, x = var_834_cast_fp16)[name = tensor<string, []>("key_states_35_cast_fp16")]; | |
| tensor<int32, [3]> var_856 = const()[name = tensor<string, []>("op_856"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_841_cast_fp16 = transpose(perm = var_841_perm_0, x = var_840_cast_fp16)[name = tensor<string, []>("transpose_13")]; | |
| tensor<fp16, [12, 77, 64]> value_states_35_cast_fp16 = reshape(shape = var_856, x = var_841_cast_fp16)[name = tensor<string, []>("value_states_35_cast_fp16")]; | |
| tensor<bool, []> attn_weights_49_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_49_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_49_cast_fp16 = matmul(transpose_x = attn_weights_49_transpose_x_1, transpose_y = attn_weights_49_transpose_y_1, x = query_states_17_cast_fp16, y = key_states_35_cast_fp16)[name = tensor<string, []>("attn_weights_49_cast_fp16")]; | |
| tensor<int32, [4]> var_861 = const()[name = tensor<string, []>("op_861"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_862_cast_fp16 = reshape(shape = var_861, x = attn_weights_49_cast_fp16)[name = tensor<string, []>("op_862_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_51_cast_fp16 = add(x = var_862_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_51_cast_fp16")]; | |
| tensor<int32, [3]> var_867 = const()[name = tensor<string, []>("op_867"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_133_cast_fp16 = reshape(shape = var_867, x = attn_weights_51_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_135_cast_fp16 = softmax(axis = var_5, x = input_133_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")]; | |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast_fp16, y = value_states_35_cast_fp16)[name = tensor<string, []>("attn_output_49_cast_fp16")]; | |
| tensor<int32, [4]> var_872 = const()[name = tensor<string, []>("op_872"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_51_cast_fp16 = reshape(shape = var_872, x = attn_output_49_cast_fp16)[name = tensor<string, []>("attn_output_51_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_53_perm_0 = const()[name = tensor<string, []>("attn_output_53_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_875 = const()[name = tensor<string, []>("op_875"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_53_cast_fp16 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast_fp16)[name = tensor<string, []>("transpose_12")]; | |
| tensor<fp16, [1, 77, 768]> input_137_cast_fp16 = reshape(shape = var_875, x = attn_output_53_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119920512))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120362944))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120363136)))]; | |
| tensor<fp16, [1, 77, 768]> linear_51_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_139_cast_fp16 = add(x = input_131_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")]; | |
| tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120364736)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120366336)))]; | |
| tensor<fp16, [1, 77, 768]> input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120367936))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122137472))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122137664)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_52_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")]; | |
| tensor<fp16, []> var_890_to_fp16 = const()[name = tensor<string, []>("op_890_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_891_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_890_to_fp16)[name = tensor<string, []>("op_891_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_892_cast_fp16 = sigmoid(x = var_891_cast_fp16)[name = tensor<string, []>("op_892_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_145_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_892_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122143872))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123913408))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123913600)))]; | |
| tensor<fp16, [1, 77, 768]> linear_53_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123915200)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123916800)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("hidden_states_55_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123918400))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124360832))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124361024)))]; | |
| tensor<fp16, [1, 77, 768]> linear_54_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")]; | |
| tensor<fp16, []> var_917_to_fp16 = const()[name = tensor<string, []>("op_917_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_59_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_917_to_fp16)[name = tensor<string, []>("tensor_59_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124362624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124805056))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124805248)))]; | |
| tensor<fp16, [1, 77, 768]> linear_55_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")]; | |
| tensor<int32, [4]> var_922 = const()[name = tensor<string, []>("op_922"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_923_cast_fp16 = reshape(shape = var_922, x = linear_55_cast_fp16)[name = tensor<string, []>("op_923_cast_fp16")]; | |
| tensor<int32, [4]> var_924_perm_0 = const()[name = tensor<string, []>("op_924_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124806848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125249280))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125249472)))]; | |
| tensor<fp16, [1, 77, 768]> linear_56_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")]; | |
| tensor<int32, [4]> var_929 = const()[name = tensor<string, []>("op_929"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_930_cast_fp16 = reshape(shape = var_929, x = linear_56_cast_fp16)[name = tensor<string, []>("op_930_cast_fp16")]; | |
| tensor<int32, [4]> var_931_perm_0 = const()[name = tensor<string, []>("op_931_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_938 = const()[name = tensor<string, []>("op_938"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_939_cast_fp16 = reshape(shape = var_938, x = tensor_59_cast_fp16)[name = tensor<string, []>("op_939_cast_fp16")]; | |
| tensor<int32, [4]> var_940_perm_0 = const()[name = tensor<string, []>("op_940_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_942 = const()[name = tensor<string, []>("op_942"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_940_cast_fp16 = transpose(perm = var_940_perm_0, x = var_939_cast_fp16)[name = tensor<string, []>("transpose_11")]; | |
| tensor<fp16, [12, 77, 64]> query_states_19_cast_fp16 = reshape(shape = var_942, x = var_940_cast_fp16)[name = tensor<string, []>("query_states_19_cast_fp16")]; | |
| tensor<int32, [3]> var_944 = const()[name = tensor<string, []>("op_944"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_924_cast_fp16 = transpose(perm = var_924_perm_0, x = var_923_cast_fp16)[name = tensor<string, []>("transpose_10")]; | |
| tensor<fp16, [12, 77, 64]> key_states_39_cast_fp16 = reshape(shape = var_944, x = var_924_cast_fp16)[name = tensor<string, []>("key_states_39_cast_fp16")]; | |
| tensor<int32, [3]> var_946 = const()[name = tensor<string, []>("op_946"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_931_cast_fp16 = transpose(perm = var_931_perm_0, x = var_930_cast_fp16)[name = tensor<string, []>("transpose_9")]; | |
| tensor<fp16, [12, 77, 64]> value_states_39_cast_fp16 = reshape(shape = var_946, x = var_931_cast_fp16)[name = tensor<string, []>("value_states_39_cast_fp16")]; | |
| tensor<bool, []> attn_weights_55_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_55_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_55_cast_fp16 = matmul(transpose_x = attn_weights_55_transpose_x_1, transpose_y = attn_weights_55_transpose_y_1, x = query_states_19_cast_fp16, y = key_states_39_cast_fp16)[name = tensor<string, []>("attn_weights_55_cast_fp16")]; | |
| tensor<int32, [4]> var_951 = const()[name = tensor<string, []>("op_951"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_952_cast_fp16 = reshape(shape = var_951, x = attn_weights_55_cast_fp16)[name = tensor<string, []>("op_952_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_57_cast_fp16 = add(x = var_952_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_57_cast_fp16")]; | |
| tensor<int32, [3]> var_957 = const()[name = tensor<string, []>("op_957"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_149_cast_fp16 = reshape(shape = var_957, x = attn_weights_57_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_151_cast_fp16 = softmax(axis = var_5, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")]; | |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_55_cast_fp16 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast_fp16, y = value_states_39_cast_fp16)[name = tensor<string, []>("attn_output_55_cast_fp16")]; | |
| tensor<int32, [4]> var_962 = const()[name = tensor<string, []>("op_962"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_57_cast_fp16 = reshape(shape = var_962, x = attn_output_55_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_59_perm_0 = const()[name = tensor<string, []>("attn_output_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_965 = const()[name = tensor<string, []>("op_965"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_59_cast_fp16 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor<string, []>("transpose_8")]; | |
| tensor<fp16, [1, 77, 768]> input_153_cast_fp16 = reshape(shape = var_965, x = attn_output_59_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125251072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125693504))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125693696)))]; | |
| tensor<fp16, [1, 77, 768]> linear_57_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_155_cast_fp16 = add(x = input_147_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")]; | |
| tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125695296)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125696896)))]; | |
| tensor<fp16, [1, 77, 768]> input_157_cast_fp16 = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125698496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127468032))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127468224)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_58_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")]; | |
| tensor<fp16, []> var_980_to_fp16 = const()[name = tensor<string, []>("op_980_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_981_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_980_to_fp16)[name = tensor<string, []>("op_981_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_982_cast_fp16 = sigmoid(x = var_981_cast_fp16)[name = tensor<string, []>("op_982_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_161_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_982_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127474432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129243968))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129244160)))]; | |
| tensor<fp16, [1, 77, 768]> linear_59_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_163_cast_fp16 = add(x = input_155_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")]; | |
| tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129245760)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129247360)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("hidden_states_61_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129248960))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129691392))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129691584)))]; | |
| tensor<fp16, [1, 77, 768]> linear_60_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")]; | |
| tensor<fp16, []> var_1007_to_fp16 = const()[name = tensor<string, []>("op_1007_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_65_cast_fp16 = mul(x = linear_60_cast_fp16, y = var_1007_to_fp16)[name = tensor<string, []>("tensor_65_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129693184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130135616))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130135808)))]; | |
| tensor<fp16, [1, 77, 768]> linear_61_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1012 = const()[name = tensor<string, []>("op_1012"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1013_cast_fp16 = reshape(shape = var_1012, x = linear_61_cast_fp16)[name = tensor<string, []>("op_1013_cast_fp16")]; | |
| tensor<int32, [4]> var_1014_perm_0 = const()[name = tensor<string, []>("op_1014_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130137408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130579840))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130580032)))]; | |
| tensor<fp16, [1, 77, 768]> linear_62_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")]; | |
| tensor<int32, [4]> var_1019 = const()[name = tensor<string, []>("op_1019"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1020_cast_fp16 = reshape(shape = var_1019, x = linear_62_cast_fp16)[name = tensor<string, []>("op_1020_cast_fp16")]; | |
| tensor<int32, [4]> var_1021_perm_0 = const()[name = tensor<string, []>("op_1021_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1028 = const()[name = tensor<string, []>("op_1028"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1029_cast_fp16 = reshape(shape = var_1028, x = tensor_65_cast_fp16)[name = tensor<string, []>("op_1029_cast_fp16")]; | |
| tensor<int32, [4]> var_1030_perm_0 = const()[name = tensor<string, []>("op_1030_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1032 = const()[name = tensor<string, []>("op_1032"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_1030_cast_fp16 = transpose(perm = var_1030_perm_0, x = var_1029_cast_fp16)[name = tensor<string, []>("transpose_7")]; | |
| tensor<fp16, [12, 77, 64]> query_states_21_cast_fp16 = reshape(shape = var_1032, x = var_1030_cast_fp16)[name = tensor<string, []>("query_states_21_cast_fp16")]; | |
| tensor<int32, [3]> var_1034 = const()[name = tensor<string, []>("op_1034"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_1014_cast_fp16 = transpose(perm = var_1014_perm_0, x = var_1013_cast_fp16)[name = tensor<string, []>("transpose_6")]; | |
| tensor<fp16, [12, 77, 64]> key_states_43_cast_fp16 = reshape(shape = var_1034, x = var_1014_cast_fp16)[name = tensor<string, []>("key_states_43_cast_fp16")]; | |
| tensor<int32, [3]> var_1036 = const()[name = tensor<string, []>("op_1036"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_1021_cast_fp16 = transpose(perm = var_1021_perm_0, x = var_1020_cast_fp16)[name = tensor<string, []>("transpose_5")]; | |
| tensor<fp16, [12, 77, 64]> value_states_43_cast_fp16 = reshape(shape = var_1036, x = var_1021_cast_fp16)[name = tensor<string, []>("value_states_43_cast_fp16")]; | |
| tensor<bool, []> attn_weights_61_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_61_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_61_cast_fp16 = matmul(transpose_x = attn_weights_61_transpose_x_1, transpose_y = attn_weights_61_transpose_y_1, x = query_states_21_cast_fp16, y = key_states_43_cast_fp16)[name = tensor<string, []>("attn_weights_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1041 = const()[name = tensor<string, []>("op_1041"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1042_cast_fp16 = reshape(shape = var_1041, x = attn_weights_61_cast_fp16)[name = tensor<string, []>("op_1042_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_63_cast_fp16 = add(x = var_1042_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_63_cast_fp16")]; | |
| tensor<int32, [3]> var_1047 = const()[name = tensor<string, []>("op_1047"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_165_cast_fp16 = reshape(shape = var_1047, x = attn_weights_63_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_167_cast_fp16 = softmax(axis = var_5, x = input_165_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")]; | |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast_fp16, y = value_states_43_cast_fp16)[name = tensor<string, []>("attn_output_61_cast_fp16")]; | |
| tensor<int32, [4]> var_1052 = const()[name = tensor<string, []>("op_1052"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_63_cast_fp16 = reshape(shape = var_1052, x = attn_output_61_cast_fp16)[name = tensor<string, []>("attn_output_63_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_65_perm_0 = const()[name = tensor<string, []>("attn_output_65_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1055 = const()[name = tensor<string, []>("op_1055"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_65_cast_fp16 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast_fp16)[name = tensor<string, []>("transpose_4")]; | |
| tensor<fp16, [1, 77, 768]> input_169_cast_fp16 = reshape(shape = var_1055, x = attn_output_65_cast_fp16)[name = tensor<string, []>("input_169_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130581632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131024064))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131024256)))]; | |
| tensor<fp16, [1, 77, 768]> linear_63_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")]; | |
| tensor<int32, [1]> input_173_axes_0 = const()[name = tensor<string, []>("input_173_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131025856)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131027456)))]; | |
| tensor<fp16, [1, 77, 768]> input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131029056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132798592))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132798784)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_64_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")]; | |
| tensor<fp16, []> var_1070_to_fp16 = const()[name = tensor<string, []>("op_1070_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1071_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1070_to_fp16)[name = tensor<string, []>("op_1071_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_1072_cast_fp16 = sigmoid(x = var_1071_cast_fp16)[name = tensor<string, []>("op_1072_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_177_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_1072_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132804992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134574528))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134574720)))]; | |
| tensor<fp16, [1, 77, 768]> linear_65_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_179_cast_fp16 = add(x = input_171_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")]; | |
| tensor<string, []> input_179_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("input_179_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134576320)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134577920)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast_fp16)[name = tensor<string, []>("hidden_states_67_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134579520))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135021952))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135022144)))]; | |
| tensor<fp16, [1, 77, 768]> linear_66_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")]; | |
| tensor<fp16, []> var_1097_to_fp16 = const()[name = tensor<string, []>("op_1097_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_cast_fp16 = mul(x = linear_66_cast_fp16, y = var_1097_to_fp16)[name = tensor<string, []>("tensor_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135023744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135466176))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135466368)))]; | |
| tensor<fp16, [1, 77, 768]> linear_67_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1102 = const()[name = tensor<string, []>("op_1102"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1103_cast_fp16 = reshape(shape = var_1102, x = linear_67_cast_fp16)[name = tensor<string, []>("op_1103_cast_fp16")]; | |
| tensor<int32, [4]> var_1104_perm_0 = const()[name = tensor<string, []>("op_1104_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135467968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135910400))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135910592)))]; | |
| tensor<fp16, [1, 77, 768]> linear_68_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")]; | |
| tensor<int32, [4]> var_1109 = const()[name = tensor<string, []>("op_1109"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1110_cast_fp16 = reshape(shape = var_1109, x = linear_68_cast_fp16)[name = tensor<string, []>("op_1110_cast_fp16")]; | |
| tensor<int32, [4]> var_1111_perm_0 = const()[name = tensor<string, []>("op_1111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1118 = const()[name = tensor<string, []>("op_1118"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1119_cast_fp16 = reshape(shape = var_1118, x = tensor_cast_fp16)[name = tensor<string, []>("op_1119_cast_fp16")]; | |
| tensor<int32, [4]> var_1120_perm_0 = const()[name = tensor<string, []>("op_1120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1122 = const()[name = tensor<string, []>("op_1122"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_1120_cast_fp16 = transpose(perm = var_1120_perm_0, x = var_1119_cast_fp16)[name = tensor<string, []>("transpose_3")]; | |
| tensor<fp16, [12, 77, 64]> query_states_cast_fp16 = reshape(shape = var_1122, x = var_1120_cast_fp16)[name = tensor<string, []>("query_states_cast_fp16")]; | |
| tensor<int32, [3]> var_1124 = const()[name = tensor<string, []>("op_1124"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_1104_cast_fp16 = transpose(perm = var_1104_perm_0, x = var_1103_cast_fp16)[name = tensor<string, []>("transpose_2")]; | |
| tensor<fp16, [12, 77, 64]> key_states_cast_fp16 = reshape(shape = var_1124, x = var_1104_cast_fp16)[name = tensor<string, []>("key_states_cast_fp16")]; | |
| tensor<int32, [3]> var_1126 = const()[name = tensor<string, []>("op_1126"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> var_1111_cast_fp16 = transpose(perm = var_1111_perm_0, x = var_1110_cast_fp16)[name = tensor<string, []>("transpose_1")]; | |
| tensor<fp16, [12, 77, 64]> value_states_cast_fp16 = reshape(shape = var_1126, x = var_1111_cast_fp16)[name = tensor<string, []>("value_states_cast_fp16")]; | |
| tensor<bool, []> attn_weights_67_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_x_1"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_67_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_y_1"), val = tensor<bool, []>(true)]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_67_cast_fp16 = matmul(transpose_x = attn_weights_67_transpose_x_1, transpose_y = attn_weights_67_transpose_y_1, x = query_states_cast_fp16, y = key_states_cast_fp16)[name = tensor<string, []>("attn_weights_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1131 = const()[name = tensor<string, []>("op_1131"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1132_cast_fp16 = reshape(shape = var_1131, x = attn_weights_67_cast_fp16)[name = tensor<string, []>("op_1132_cast_fp16")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_69_cast_fp16 = add(x = var_1132_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_69_cast_fp16")]; | |
| tensor<int32, [3]> var_1137 = const()[name = tensor<string, []>("op_1137"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_181_cast_fp16 = reshape(shape = var_1137, x = attn_weights_69_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")]; | |
| tensor<fp16, [12, 77, 77]> input_183_cast_fp16 = softmax(axis = var_5, x = input_181_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")]; | |
| tensor<bool, []> attn_output_67_transpose_x_0 = const()[name = tensor<string, []>("attn_output_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_67_transpose_y_0 = const()[name = tensor<string, []>("attn_output_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_67_cast_fp16 = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast_fp16, y = value_states_cast_fp16)[name = tensor<string, []>("attn_output_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1142 = const()[name = tensor<string, []>("op_1142"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_69_cast_fp16 = reshape(shape = var_1142, x = attn_output_67_cast_fp16)[name = tensor<string, []>("attn_output_69_cast_fp16")]; | |
| tensor<int32, [4]> attn_output_perm_0 = const()[name = tensor<string, []>("attn_output_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1145 = const()[name = tensor<string, []>("op_1145"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> attn_output_cast_fp16 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast_fp16)[name = tensor<string, []>("transpose_0")]; | |
| tensor<fp16, [1, 77, 768]> input_185_cast_fp16 = reshape(shape = var_1145, x = attn_output_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135912192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136354624))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136354816)))]; | |
| tensor<fp16, [1, 77, 768]> linear_69_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_187_cast_fp16 = add(x = input_179_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")]; | |
| tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136356416)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136358016)))]; | |
| tensor<fp16, [1, 77, 768]> input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136359616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138129152))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138129344)))]; | |
| tensor<fp16, [1, 77, 3072]> linear_70_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")]; | |
| tensor<fp16, []> var_1160_to_fp16 = const()[name = tensor<string, []>("op_1160_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1161_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1160_to_fp16)[name = tensor<string, []>("op_1161_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> var_1162_cast_fp16 = sigmoid(x = var_1161_cast_fp16)[name = tensor<string, []>("op_1162_cast_fp16")]; | |
| tensor<fp16, [1, 77, 3072]> input_193_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_1162_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138135552))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139905088))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 3072])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139905280)))]; | |
| tensor<fp16, [1, 77, 768]> linear_71_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")]; | |
| tensor<fp16, [1, 77, 768]> input_cast_fp16 = add(x = input_187_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; | |
| tensor<int32, [1]> last_hidden_state_axes_0 = const()[name = tensor<string, []>("last_hidden_state_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139906880)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139908480)))]; | |
| tensor<fp16, [1, 77, 768]> last_hidden_state_cast_fp16 = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("last_hidden_state_cast_fp16")]; | |
| tensor<int32, [1]> var_1176 = const()[name = tensor<string, []>("op_1176"), val = tensor<int32, [1]>([0])]; | |
| tensor<int32, [1]> var_1178 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_1)[name = tensor<string, []>("op_1178")]; | |
| tensor<int32, []> stack_0_axis_0 = const()[name = tensor<string, []>("stack_0_axis_0"), val = tensor<int32, []>(1)]; | |
| tensor<int32, [1, 2]> stack_0 = stack(axis = stack_0_axis_0, values = (var_1176, var_1178))[name = tensor<string, []>("stack_0")]; | |
| tensor<int32, []> var_1180_transpose_batch_dims_0 = const()[name = tensor<string, []>("op_1180_transpose_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [1, 768]> var_1180_transpose_cast_fp16 = gather_nd(batch_dims = var_1180_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("op_1180_transpose_cast_fp16")]; | |
| tensor<string, []> var_1180_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_1180_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<fp32, [1, 77, 768]> hidden_embeds = cast(dtype = input_179_cast_fp16_to_fp32_dtype_0, x = input_179_cast_fp16)[name = tensor<string, []>("cast_0")]; | |
| tensor<fp32, [1, 768]> pooled_outputs = cast(dtype = var_1180_cast_fp16_to_fp32_dtype_0, x = var_1180_transpose_cast_fp16)[name = tensor<string, []>("cast_1")]; | |
| } -> (hidden_embeds, pooled_outputs); | |
| } |