| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}})] |
| { |
| func main<ios16>(tensor<fp32, [1, 77]> input_ids) { |
| 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, [59136]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75890816))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75950016))), 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, []>(75950592)))]; |
| 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, []>(75952192)))]; |
| 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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75953792))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76543680))), 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, []>(76544256)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76545856))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77135744))), 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, []>(77136320)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77137920))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77727808))), 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, []>(77728384)))]; |
| 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_112 = const()[name = tensor<string, []>("op_112"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_113_cast_fp16 = reshape(shape = var_112, x = linear_0_cast_fp16)[name = tensor<string, []>("op_113_cast_fp16")]; |
| tensor<int32, [4]> var_115 = const()[name = tensor<string, []>("op_115"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_116_cast_fp16 = reshape(shape = var_115, x = linear_1_cast_fp16)[name = tensor<string, []>("op_116_cast_fp16")]; |
| tensor<int32, [4]> var_118 = const()[name = tensor<string, []>("op_118"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_119_cast_fp16 = reshape(shape = var_118, x = linear_2_cast_fp16)[name = tensor<string, []>("op_119_cast_fp16")]; |
| tensor<int32, [4]> value_1_perm_0 = const()[name = tensor<string, []>("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, []> var_17_to_fp16 = const()[name = tensor<string, []>("op_17_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 12, 64]> mul_0_cast_fp16 = mul(x = var_113_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_0_cast_fp16")]; |
| tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_48_perm_0 = const()[name = tensor<string, []>("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_49_perm_0 = const()[name = tensor<string, []>("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_49 = transpose(perm = transpose_49_perm_0, x = var_116_cast_fp16)[name = tensor<string, []>("transpose_118")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_48 = transpose(perm = transpose_48_perm_0, x = mul_0_cast_fp16)[name = tensor<string, []>("transpose_119")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_48, y = transpose_49)[name = tensor<string, []>("matmul_0_cast_fp16")]; |
| tensor<fp16, [1, 1, 77, 77]> op_56_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5929]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77729984))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77736000))), name = tensor<string, []>("op_56_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 77, 77])]; |
| tensor<fp16, [1, 12, 77, 77]> add_0_cast_fp16 = add(x = matmul_0_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_0_cast_fp16")]; |
| tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = add_0_cast_fp16)[name = tensor<string, []>("softmax_0_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, [1, 12, 77, 64]> value_1_cast_fp16 = transpose(perm = value_1_perm_0, x = var_119_cast_fp16)[name = tensor<string, []>("transpose_117")]; |
| tensor<fp16, [1, 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 = softmax_0_cast_fp16, y = value_1_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")]; |
| tensor<int32, [4]> var_127_perm_0 = const()[name = tensor<string, []>("op_127_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_129 = const()[name = tensor<string, []>("op_129"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_127_cast_fp16 = transpose(perm = var_127_perm_0, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_116")]; |
| tensor<fp16, [1, 77, 768]> var_130_cast_fp16 = reshape(shape = var_129, x = var_127_cast_fp16)[name = tensor<string, []>("op_130_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77736576))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78326464))), 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, []>(78327040)))]; |
| 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 = var_130_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_7_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| tensor<int32, [1]> input_9_axes_0 = const()[name = tensor<string, []>("input_9_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, []>(78328640)))]; |
| 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, []>(78330240)))]; |
| tensor<fp16, [1, 77, 768]> input_9_cast_fp16 = layer_norm(axes = input_9_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_7_cast_fp16)[name = tensor<string, []>("input_9_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78331840))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80691200))), 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, []>(80691776)))]; |
| 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_9_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; |
| tensor<fp16, []> var_145_to_fp16 = const()[name = tensor<string, []>("op_145_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_146_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_145_to_fp16)[name = tensor<string, []>("op_146_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_147_cast_fp16 = sigmoid(x = var_146_cast_fp16)[name = tensor<string, []>("op_147_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_13_cast_fp16 = mul(x = linear_4_cast_fp16, y = var_147_cast_fp16)[name = tensor<string, []>("input_13_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80697984))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83057344))), 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, []>(83057920)))]; |
| 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_13_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_15_cast_fp16 = add(x = input_7_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_15_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, []>(83059520)))]; |
| 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, []>(83061120)))]; |
| 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_15_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83062720))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83652608))), 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, []>(83653184)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83654784))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84244672))), 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, []>(84245248)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84246848))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84836736))), 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, []>(84837312)))]; |
| 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_177 = const()[name = tensor<string, []>("op_177"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_178_cast_fp16 = reshape(shape = var_177, x = linear_6_cast_fp16)[name = tensor<string, []>("op_178_cast_fp16")]; |
| tensor<int32, [4]> var_180 = const()[name = tensor<string, []>("op_180"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_181_cast_fp16 = reshape(shape = var_180, x = linear_7_cast_fp16)[name = tensor<string, []>("op_181_cast_fp16")]; |
| tensor<int32, [4]> var_183 = const()[name = tensor<string, []>("op_183"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_184_cast_fp16 = reshape(shape = var_183, x = linear_8_cast_fp16)[name = tensor<string, []>("op_184_cast_fp16")]; |
| tensor<int32, [4]> value_3_perm_0 = const()[name = tensor<string, []>("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_1_cast_fp16 = mul(x = var_178_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_1_cast_fp16")]; |
| tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_50_perm_0 = const()[name = tensor<string, []>("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_51_perm_0 = const()[name = tensor<string, []>("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_51 = transpose(perm = transpose_51_perm_0, x = var_181_cast_fp16)[name = tensor<string, []>("transpose_114")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_50 = transpose(perm = transpose_50_perm_0, x = mul_1_cast_fp16)[name = tensor<string, []>("transpose_115")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_50, y = transpose_51)[name = tensor<string, []>("matmul_1_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_1_cast_fp16 = add(x = matmul_1_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_1_cast_fp16")]; |
| tensor<int32, []> softmax_1_axis_0 = const()[name = tensor<string, []>("softmax_1_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = add_1_cast_fp16)[name = tensor<string, []>("softmax_1_cast_fp16")]; |
| tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = var_184_cast_fp16)[name = tensor<string, []>("transpose_113")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_1_cast_fp16, y = value_3_cast_fp16)[name = tensor<string, []>("attn_output_5_cast_fp16")]; |
| tensor<int32, [4]> var_192_perm_0 = const()[name = tensor<string, []>("op_192_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_194 = const()[name = tensor<string, []>("op_194"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_192_cast_fp16 = transpose(perm = var_192_perm_0, x = attn_output_5_cast_fp16)[name = tensor<string, []>("transpose_112")]; |
| tensor<fp16, [1, 77, 768]> var_195_cast_fp16 = reshape(shape = var_194, x = var_192_cast_fp16)[name = tensor<string, []>("op_195_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84838912))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85428800))), 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, []>(85429376)))]; |
| 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 = var_195_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_19_cast_fp16 = add(x = input_15_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; |
| tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_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, []>(85430976)))]; |
| 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, []>(85432576)))]; |
| tensor<fp16, [1, 77, 768]> input_21_cast_fp16 = layer_norm(axes = input_21_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_19_cast_fp16)[name = tensor<string, []>("input_21_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85434176))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87793536))), 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, []>(87794112)))]; |
| 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_21_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; |
| tensor<fp16, []> var_210_to_fp16 = const()[name = tensor<string, []>("op_210_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_211_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_210_to_fp16)[name = tensor<string, []>("op_211_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_212_cast_fp16 = sigmoid(x = var_211_cast_fp16)[name = tensor<string, []>("op_212_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_25_cast_fp16 = mul(x = linear_10_cast_fp16, y = var_212_cast_fp16)[name = tensor<string, []>("input_25_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87800320))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90159680))), 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, []>(90160256)))]; |
| 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_25_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_27_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, []>(90161856)))]; |
| 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, []>(90163456)))]; |
| 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_27_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90165056))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90754944))), 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, []>(90755520)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90757120))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91347008))), 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, []>(91347584)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91349184))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91939072))), 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, []>(91939648)))]; |
| 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_242 = const()[name = tensor<string, []>("op_242"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_243_cast_fp16 = reshape(shape = var_242, x = linear_12_cast_fp16)[name = tensor<string, []>("op_243_cast_fp16")]; |
| tensor<int32, [4]> var_245 = const()[name = tensor<string, []>("op_245"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_246_cast_fp16 = reshape(shape = var_245, x = linear_13_cast_fp16)[name = tensor<string, []>("op_246_cast_fp16")]; |
| tensor<int32, [4]> var_248 = const()[name = tensor<string, []>("op_248"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_249_cast_fp16 = reshape(shape = var_248, x = linear_14_cast_fp16)[name = tensor<string, []>("op_249_cast_fp16")]; |
| tensor<int32, [4]> value_5_perm_0 = const()[name = tensor<string, []>("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_2_cast_fp16 = mul(x = var_243_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_2_cast_fp16")]; |
| tensor<bool, []> matmul_2_transpose_y_0 = const()[name = tensor<string, []>("matmul_2_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_2_transpose_x_0 = const()[name = tensor<string, []>("matmul_2_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_52_perm_0 = const()[name = tensor<string, []>("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_53_perm_0 = const()[name = tensor<string, []>("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_53 = transpose(perm = transpose_53_perm_0, x = var_246_cast_fp16)[name = tensor<string, []>("transpose_110")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_52 = transpose(perm = transpose_52_perm_0, x = mul_2_cast_fp16)[name = tensor<string, []>("transpose_111")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_52, y = transpose_53)[name = tensor<string, []>("matmul_2_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_2_cast_fp16 = add(x = matmul_2_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_2_cast_fp16")]; |
| tensor<int32, []> softmax_2_axis_0 = const()[name = tensor<string, []>("softmax_2_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = add_2_cast_fp16)[name = tensor<string, []>("softmax_2_cast_fp16")]; |
| tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = var_249_cast_fp16)[name = tensor<string, []>("transpose_109")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_2_cast_fp16, y = value_5_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")]; |
| tensor<int32, [4]> var_257_perm_0 = const()[name = tensor<string, []>("op_257_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_259 = const()[name = tensor<string, []>("op_259"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_257_cast_fp16 = transpose(perm = var_257_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_108")]; |
| tensor<fp16, [1, 77, 768]> var_260_cast_fp16 = reshape(shape = var_259, x = var_257_cast_fp16)[name = tensor<string, []>("op_260_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91941248))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92531136))), 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, []>(92531712)))]; |
| 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 = var_260_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_31_cast_fp16 = add(x = input_27_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")]; |
| tensor<int32, [1]> input_33_axes_0 = const()[name = tensor<string, []>("input_33_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, []>(92533312)))]; |
| 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, []>(92534912)))]; |
| tensor<fp16, [1, 77, 768]> input_33_cast_fp16 = layer_norm(axes = input_33_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_31_cast_fp16)[name = tensor<string, []>("input_33_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92536512))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94895872))), 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, []>(94896448)))]; |
| 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_33_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; |
| tensor<fp16, []> var_275_to_fp16 = const()[name = tensor<string, []>("op_275_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_276_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_275_to_fp16)[name = tensor<string, []>("op_276_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_277_cast_fp16 = sigmoid(x = var_276_cast_fp16)[name = tensor<string, []>("op_277_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_37_cast_fp16 = mul(x = linear_16_cast_fp16, y = var_277_cast_fp16)[name = tensor<string, []>("input_37_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94902656))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97262016))), 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, []>(97262592)))]; |
| 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_37_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_39_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, []>(97264192)))]; |
| 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, []>(97265792)))]; |
| 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_39_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97267392))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97857280))), 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, []>(97857856)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97859456))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98449344))), 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, []>(98449920)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98451520))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99041408))), 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, []>(99041984)))]; |
| 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_307 = const()[name = tensor<string, []>("op_307"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_308_cast_fp16 = reshape(shape = var_307, x = linear_18_cast_fp16)[name = tensor<string, []>("op_308_cast_fp16")]; |
| tensor<int32, [4]> var_310 = const()[name = tensor<string, []>("op_310"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_311_cast_fp16 = reshape(shape = var_310, x = linear_19_cast_fp16)[name = tensor<string, []>("op_311_cast_fp16")]; |
| tensor<int32, [4]> var_313 = const()[name = tensor<string, []>("op_313"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_314_cast_fp16 = reshape(shape = var_313, x = linear_20_cast_fp16)[name = tensor<string, []>("op_314_cast_fp16")]; |
| tensor<int32, [4]> value_7_perm_0 = const()[name = tensor<string, []>("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_3_cast_fp16 = mul(x = var_308_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_3_cast_fp16")]; |
| tensor<bool, []> matmul_3_transpose_y_0 = const()[name = tensor<string, []>("matmul_3_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_3_transpose_x_0 = const()[name = tensor<string, []>("matmul_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_54_perm_0 = const()[name = tensor<string, []>("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_55_perm_0 = const()[name = tensor<string, []>("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_55 = transpose(perm = transpose_55_perm_0, x = var_311_cast_fp16)[name = tensor<string, []>("transpose_106")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_54 = transpose(perm = transpose_54_perm_0, x = mul_3_cast_fp16)[name = tensor<string, []>("transpose_107")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = transpose_54, y = transpose_55)[name = tensor<string, []>("matmul_3_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_3_cast_fp16 = add(x = matmul_3_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_3_cast_fp16")]; |
| tensor<int32, []> softmax_3_axis_0 = const()[name = tensor<string, []>("softmax_3_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = add_3_cast_fp16)[name = tensor<string, []>("softmax_3_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, [1, 12, 77, 64]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = var_314_cast_fp16)[name = tensor<string, []>("transpose_105")]; |
| tensor<fp16, [1, 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 = softmax_3_cast_fp16, y = value_7_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")]; |
| tensor<int32, [4]> var_322_perm_0 = const()[name = tensor<string, []>("op_322_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_324 = const()[name = tensor<string, []>("op_324"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_322_cast_fp16 = transpose(perm = var_322_perm_0, x = attn_output_13_cast_fp16)[name = tensor<string, []>("transpose_104")]; |
| tensor<fp16, [1, 77, 768]> var_325_cast_fp16 = reshape(shape = var_324, x = var_322_cast_fp16)[name = tensor<string, []>("op_325_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99043584))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99633472))), 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, []>(99634048)))]; |
| 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 = var_325_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_43_cast_fp16 = add(x = input_39_cast_fp16, y = linear_21_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_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, []>(99635648)))]; |
| 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, []>(99637248)))]; |
| tensor<fp16, [1, 77, 768]> input_45_cast_fp16 = layer_norm(axes = input_45_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_43_cast_fp16)[name = tensor<string, []>("input_45_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99638848))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101998208))), 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, []>(101998784)))]; |
| 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_45_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; |
| tensor<fp16, []> var_340_to_fp16 = const()[name = tensor<string, []>("op_340_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_341_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_340_to_fp16)[name = tensor<string, []>("op_341_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_342_cast_fp16 = sigmoid(x = var_341_cast_fp16)[name = tensor<string, []>("op_342_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_49_cast_fp16 = mul(x = linear_22_cast_fp16, y = var_342_cast_fp16)[name = tensor<string, []>("input_49_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102004992))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104364352))), 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, []>(104364928)))]; |
| 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_49_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_51_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, []>(104366528)))]; |
| 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, []>(104368128)))]; |
| 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_51_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104369728))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104959616))), 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, []>(104960192)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104961792))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105551680))), 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, []>(105552256)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105553856))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106143744))), 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, []>(106144320)))]; |
| 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_372 = const()[name = tensor<string, []>("op_372"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_373_cast_fp16 = reshape(shape = var_372, x = linear_24_cast_fp16)[name = tensor<string, []>("op_373_cast_fp16")]; |
| tensor<int32, [4]> var_375 = const()[name = tensor<string, []>("op_375"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_376_cast_fp16 = reshape(shape = var_375, x = linear_25_cast_fp16)[name = tensor<string, []>("op_376_cast_fp16")]; |
| tensor<int32, [4]> var_378 = const()[name = tensor<string, []>("op_378"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_379_cast_fp16 = reshape(shape = var_378, x = linear_26_cast_fp16)[name = tensor<string, []>("op_379_cast_fp16")]; |
| tensor<int32, [4]> value_9_perm_0 = const()[name = tensor<string, []>("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_4_cast_fp16 = mul(x = var_373_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_4_cast_fp16")]; |
| tensor<bool, []> matmul_4_transpose_y_0 = const()[name = tensor<string, []>("matmul_4_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_4_transpose_x_0 = const()[name = tensor<string, []>("matmul_4_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_56_perm_0 = const()[name = tensor<string, []>("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_57_perm_0 = const()[name = tensor<string, []>("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_57 = transpose(perm = transpose_57_perm_0, x = var_376_cast_fp16)[name = tensor<string, []>("transpose_102")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_56 = transpose(perm = transpose_56_perm_0, x = mul_4_cast_fp16)[name = tensor<string, []>("transpose_103")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_56, y = transpose_57)[name = tensor<string, []>("matmul_4_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_4_cast_fp16 = add(x = matmul_4_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_4_cast_fp16")]; |
| tensor<int32, []> softmax_4_axis_0 = const()[name = tensor<string, []>("softmax_4_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = add_4_cast_fp16)[name = tensor<string, []>("softmax_4_cast_fp16")]; |
| tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = var_379_cast_fp16)[name = tensor<string, []>("transpose_101")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = softmax_4_cast_fp16, y = value_9_cast_fp16)[name = tensor<string, []>("attn_output_17_cast_fp16")]; |
| tensor<int32, [4]> var_387_perm_0 = const()[name = tensor<string, []>("op_387_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_389 = const()[name = tensor<string, []>("op_389"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_387_cast_fp16 = transpose(perm = var_387_perm_0, x = attn_output_17_cast_fp16)[name = tensor<string, []>("transpose_100")]; |
| tensor<fp16, [1, 77, 768]> var_390_cast_fp16 = reshape(shape = var_389, x = var_387_cast_fp16)[name = tensor<string, []>("op_390_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106145920))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106735808))), 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, []>(106736384)))]; |
| 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 = var_390_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_55_cast_fp16 = add(x = input_51_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; |
| tensor<int32, [1]> input_57_axes_0 = const()[name = tensor<string, []>("input_57_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, []>(106737984)))]; |
| 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, []>(106739584)))]; |
| tensor<fp16, [1, 77, 768]> input_57_cast_fp16 = layer_norm(axes = input_57_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_55_cast_fp16)[name = tensor<string, []>("input_57_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106741184))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109100544))), 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, []>(109101120)))]; |
| 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_57_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")]; |
| tensor<fp16, []> var_405_to_fp16 = const()[name = tensor<string, []>("op_405_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_406_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_405_to_fp16)[name = tensor<string, []>("op_406_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_407_cast_fp16 = sigmoid(x = var_406_cast_fp16)[name = tensor<string, []>("op_407_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_61_cast_fp16 = mul(x = linear_28_cast_fp16, y = var_407_cast_fp16)[name = tensor<string, []>("input_61_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109107328))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111466688))), 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, []>(111467264)))]; |
| 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_61_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_63_cast_fp16 = add(x = input_55_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_63_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, []>(111468864)))]; |
| 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, []>(111470464)))]; |
| 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_63_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111472064))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112061952))), 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, []>(112062528)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112064128))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112654016))), 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, []>(112654592)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112656192))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113246080))), 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, []>(113246656)))]; |
| 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_437 = const()[name = tensor<string, []>("op_437"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_438_cast_fp16 = reshape(shape = var_437, x = linear_30_cast_fp16)[name = tensor<string, []>("op_438_cast_fp16")]; |
| tensor<int32, [4]> var_440 = const()[name = tensor<string, []>("op_440"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_441_cast_fp16 = reshape(shape = var_440, x = linear_31_cast_fp16)[name = tensor<string, []>("op_441_cast_fp16")]; |
| tensor<int32, [4]> var_443 = const()[name = tensor<string, []>("op_443"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_444_cast_fp16 = reshape(shape = var_443, x = linear_32_cast_fp16)[name = tensor<string, []>("op_444_cast_fp16")]; |
| tensor<int32, [4]> value_11_perm_0 = const()[name = tensor<string, []>("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_5_cast_fp16 = mul(x = var_438_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_5_cast_fp16")]; |
| tensor<bool, []> matmul_5_transpose_y_0 = const()[name = tensor<string, []>("matmul_5_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_5_transpose_x_0 = const()[name = tensor<string, []>("matmul_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_58_perm_0 = const()[name = tensor<string, []>("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_59_perm_0 = const()[name = tensor<string, []>("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_59 = transpose(perm = transpose_59_perm_0, x = var_441_cast_fp16)[name = tensor<string, []>("transpose_98")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_58 = transpose(perm = transpose_58_perm_0, x = mul_5_cast_fp16)[name = tensor<string, []>("transpose_99")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = transpose_58, y = transpose_59)[name = tensor<string, []>("matmul_5_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_5_cast_fp16 = add(x = matmul_5_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_5_cast_fp16")]; |
| tensor<int32, []> softmax_5_axis_0 = const()[name = tensor<string, []>("softmax_5_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_5_cast_fp16 = softmax(axis = softmax_5_axis_0, x = add_5_cast_fp16)[name = tensor<string, []>("softmax_5_cast_fp16")]; |
| tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = var_444_cast_fp16)[name = tensor<string, []>("transpose_97")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = softmax_5_cast_fp16, y = value_11_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")]; |
| tensor<int32, [4]> var_452_perm_0 = const()[name = tensor<string, []>("op_452_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_454 = const()[name = tensor<string, []>("op_454"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_452_cast_fp16 = transpose(perm = var_452_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_96")]; |
| tensor<fp16, [1, 77, 768]> var_455_cast_fp16 = reshape(shape = var_454, x = var_452_cast_fp16)[name = tensor<string, []>("op_455_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113248256))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113838144))), 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, []>(113838720)))]; |
| 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 = var_455_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_67_cast_fp16 = add(x = input_63_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; |
| tensor<int32, [1]> input_69_axes_0 = const()[name = tensor<string, []>("input_69_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, []>(113840320)))]; |
| 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, []>(113841920)))]; |
| tensor<fp16, [1, 77, 768]> input_69_cast_fp16 = layer_norm(axes = input_69_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_67_cast_fp16)[name = tensor<string, []>("input_69_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113843520))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116202880))), 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, []>(116203456)))]; |
| 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_69_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")]; |
| tensor<fp16, []> var_470_to_fp16 = const()[name = tensor<string, []>("op_470_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_471_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_470_to_fp16)[name = tensor<string, []>("op_471_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_472_cast_fp16 = sigmoid(x = var_471_cast_fp16)[name = tensor<string, []>("op_472_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_73_cast_fp16 = mul(x = linear_34_cast_fp16, y = var_472_cast_fp16)[name = tensor<string, []>("input_73_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116209664))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118569024))), 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, []>(118569600)))]; |
| 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_73_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_75_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, []>(118571200)))]; |
| 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, []>(118572800)))]; |
| 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_75_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118574400))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119164288))), 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, []>(119164864)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119166464))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119756352))), 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, []>(119756928)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119758528))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120348416))), 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, []>(120348992)))]; |
| 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_502 = const()[name = tensor<string, []>("op_502"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_503_cast_fp16 = reshape(shape = var_502, x = linear_36_cast_fp16)[name = tensor<string, []>("op_503_cast_fp16")]; |
| tensor<int32, [4]> var_505 = const()[name = tensor<string, []>("op_505"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_506_cast_fp16 = reshape(shape = var_505, x = linear_37_cast_fp16)[name = tensor<string, []>("op_506_cast_fp16")]; |
| tensor<int32, [4]> var_508 = const()[name = tensor<string, []>("op_508"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_509_cast_fp16 = reshape(shape = var_508, x = linear_38_cast_fp16)[name = tensor<string, []>("op_509_cast_fp16")]; |
| tensor<int32, [4]> value_13_perm_0 = const()[name = tensor<string, []>("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_6_cast_fp16 = mul(x = var_503_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_6_cast_fp16")]; |
| tensor<bool, []> matmul_6_transpose_y_0 = const()[name = tensor<string, []>("matmul_6_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_6_transpose_x_0 = const()[name = tensor<string, []>("matmul_6_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_60_perm_0 = const()[name = tensor<string, []>("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_61_perm_0 = const()[name = tensor<string, []>("transpose_61_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_61 = transpose(perm = transpose_61_perm_0, x = var_506_cast_fp16)[name = tensor<string, []>("transpose_94")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_60 = transpose(perm = transpose_60_perm_0, x = mul_6_cast_fp16)[name = tensor<string, []>("transpose_95")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_60, y = transpose_61)[name = tensor<string, []>("matmul_6_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_6_cast_fp16 = add(x = matmul_6_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_6_cast_fp16")]; |
| tensor<int32, []> softmax_6_axis_0 = const()[name = tensor<string, []>("softmax_6_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_6_cast_fp16 = softmax(axis = softmax_6_axis_0, x = add_6_cast_fp16)[name = tensor<string, []>("softmax_6_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, [1, 12, 77, 64]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = var_509_cast_fp16)[name = tensor<string, []>("transpose_93")]; |
| tensor<fp16, [1, 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 = softmax_6_cast_fp16, y = value_13_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")]; |
| tensor<int32, [4]> var_517_perm_0 = const()[name = tensor<string, []>("op_517_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_519 = const()[name = tensor<string, []>("op_519"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_517_cast_fp16 = transpose(perm = var_517_perm_0, x = attn_output_25_cast_fp16)[name = tensor<string, []>("transpose_92")]; |
| tensor<fp16, [1, 77, 768]> var_520_cast_fp16 = reshape(shape = var_519, x = var_517_cast_fp16)[name = tensor<string, []>("op_520_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120350592))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120940480))), 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, []>(120941056)))]; |
| 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 = var_520_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_79_cast_fp16 = add(x = input_75_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")]; |
| tensor<int32, [1]> input_81_axes_0 = const()[name = tensor<string, []>("input_81_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, []>(120942656)))]; |
| 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, []>(120944256)))]; |
| tensor<fp16, [1, 77, 768]> input_81_cast_fp16 = layer_norm(axes = input_81_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_79_cast_fp16)[name = tensor<string, []>("input_81_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120945856))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123305216))), 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, []>(123305792)))]; |
| 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_81_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")]; |
| tensor<fp16, []> var_535_to_fp16 = const()[name = tensor<string, []>("op_535_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_536_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_535_to_fp16)[name = tensor<string, []>("op_536_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_537_cast_fp16 = sigmoid(x = var_536_cast_fp16)[name = tensor<string, []>("op_537_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_85_cast_fp16 = mul(x = linear_40_cast_fp16, y = var_537_cast_fp16)[name = tensor<string, []>("input_85_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123312000))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125671360))), 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, []>(125671936)))]; |
| 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_85_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_87_cast_fp16 = add(x = input_79_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_87_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, []>(125673536)))]; |
| 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, []>(125675136)))]; |
| 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_87_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125676736))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126266624))), 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, []>(126267200)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126268800))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126858688))), 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, []>(126859264)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126860864))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127450752))), 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, []>(127451328)))]; |
| 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_567 = const()[name = tensor<string, []>("op_567"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_568_cast_fp16 = reshape(shape = var_567, x = linear_42_cast_fp16)[name = tensor<string, []>("op_568_cast_fp16")]; |
| tensor<int32, [4]> var_570 = const()[name = tensor<string, []>("op_570"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_571_cast_fp16 = reshape(shape = var_570, x = linear_43_cast_fp16)[name = tensor<string, []>("op_571_cast_fp16")]; |
| tensor<int32, [4]> var_573 = const()[name = tensor<string, []>("op_573"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_574_cast_fp16 = reshape(shape = var_573, x = linear_44_cast_fp16)[name = tensor<string, []>("op_574_cast_fp16")]; |
| tensor<int32, [4]> value_15_perm_0 = const()[name = tensor<string, []>("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_7_cast_fp16 = mul(x = var_568_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_7_cast_fp16")]; |
| tensor<bool, []> matmul_7_transpose_y_0 = const()[name = tensor<string, []>("matmul_7_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_7_transpose_x_0 = const()[name = tensor<string, []>("matmul_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_62_perm_0 = const()[name = tensor<string, []>("transpose_62_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_63_perm_0 = const()[name = tensor<string, []>("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_63 = transpose(perm = transpose_63_perm_0, x = var_571_cast_fp16)[name = tensor<string, []>("transpose_90")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_62 = transpose(perm = transpose_62_perm_0, x = mul_7_cast_fp16)[name = tensor<string, []>("transpose_91")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = transpose_62, y = transpose_63)[name = tensor<string, []>("matmul_7_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_7_cast_fp16 = add(x = matmul_7_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_7_cast_fp16")]; |
| tensor<int32, []> softmax_7_axis_0 = const()[name = tensor<string, []>("softmax_7_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_7_cast_fp16 = softmax(axis = softmax_7_axis_0, x = add_7_cast_fp16)[name = tensor<string, []>("softmax_7_cast_fp16")]; |
| tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = var_574_cast_fp16)[name = tensor<string, []>("transpose_89")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = softmax_7_cast_fp16, y = value_15_cast_fp16)[name = tensor<string, []>("attn_output_29_cast_fp16")]; |
| tensor<int32, [4]> var_582_perm_0 = const()[name = tensor<string, []>("op_582_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_584 = const()[name = tensor<string, []>("op_584"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_582_cast_fp16 = transpose(perm = var_582_perm_0, x = attn_output_29_cast_fp16)[name = tensor<string, []>("transpose_88")]; |
| tensor<fp16, [1, 77, 768]> var_585_cast_fp16 = reshape(shape = var_584, x = var_582_cast_fp16)[name = tensor<string, []>("op_585_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127452928))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128042816))), 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, []>(128043392)))]; |
| 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 = var_585_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_91_cast_fp16 = add(x = input_87_cast_fp16, y = linear_45_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_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, []>(128044992)))]; |
| 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, []>(128046592)))]; |
| tensor<fp16, [1, 77, 768]> input_93_cast_fp16 = layer_norm(axes = input_93_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_91_cast_fp16)[name = tensor<string, []>("input_93_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128048192))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130407552))), 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, []>(130408128)))]; |
| 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_93_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")]; |
| tensor<fp16, []> var_600_to_fp16 = const()[name = tensor<string, []>("op_600_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_601_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_600_to_fp16)[name = tensor<string, []>("op_601_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_602_cast_fp16 = sigmoid(x = var_601_cast_fp16)[name = tensor<string, []>("op_602_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_97_cast_fp16 = mul(x = linear_46_cast_fp16, y = var_602_cast_fp16)[name = tensor<string, []>("input_97_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130414336))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132773696))), 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, []>(132774272)))]; |
| 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_97_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_99_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, []>(132775872)))]; |
| 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, []>(132777472)))]; |
| 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_99_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132779072))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133368960))), 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, []>(133369536)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133371136))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133961024))), 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, []>(133961600)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133963200))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134553088))), 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, []>(134553664)))]; |
| 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_632 = const()[name = tensor<string, []>("op_632"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_633_cast_fp16 = reshape(shape = var_632, x = linear_48_cast_fp16)[name = tensor<string, []>("op_633_cast_fp16")]; |
| tensor<int32, [4]> var_635 = const()[name = tensor<string, []>("op_635"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_636_cast_fp16 = reshape(shape = var_635, x = linear_49_cast_fp16)[name = tensor<string, []>("op_636_cast_fp16")]; |
| tensor<int32, [4]> var_638 = const()[name = tensor<string, []>("op_638"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_639_cast_fp16 = reshape(shape = var_638, x = linear_50_cast_fp16)[name = tensor<string, []>("op_639_cast_fp16")]; |
| tensor<int32, [4]> value_17_perm_0 = const()[name = tensor<string, []>("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_8_cast_fp16 = mul(x = var_633_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_8_cast_fp16")]; |
| tensor<bool, []> matmul_8_transpose_y_0 = const()[name = tensor<string, []>("matmul_8_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_8_transpose_x_0 = const()[name = tensor<string, []>("matmul_8_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_64_perm_0 = const()[name = tensor<string, []>("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_65_perm_0 = const()[name = tensor<string, []>("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_65 = transpose(perm = transpose_65_perm_0, x = var_636_cast_fp16)[name = tensor<string, []>("transpose_86")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_64 = transpose(perm = transpose_64_perm_0, x = mul_8_cast_fp16)[name = tensor<string, []>("transpose_87")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_64, y = transpose_65)[name = tensor<string, []>("matmul_8_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_8_cast_fp16 = add(x = matmul_8_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_8_cast_fp16")]; |
| tensor<int32, []> softmax_8_axis_0 = const()[name = tensor<string, []>("softmax_8_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_8_cast_fp16 = softmax(axis = softmax_8_axis_0, x = add_8_cast_fp16)[name = tensor<string, []>("softmax_8_cast_fp16")]; |
| tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = var_639_cast_fp16)[name = tensor<string, []>("transpose_85")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_33_cast_fp16 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = softmax_8_cast_fp16, y = value_17_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")]; |
| tensor<int32, [4]> var_647_perm_0 = const()[name = tensor<string, []>("op_647_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_649 = const()[name = tensor<string, []>("op_649"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_647_cast_fp16 = transpose(perm = var_647_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_84")]; |
| tensor<fp16, [1, 77, 768]> var_650_cast_fp16 = reshape(shape = var_649, x = var_647_cast_fp16)[name = tensor<string, []>("op_650_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134555264))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135145152))), 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, []>(135145728)))]; |
| 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 = var_650_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_103_cast_fp16 = add(x = input_99_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; |
| tensor<int32, [1]> input_105_axes_0 = const()[name = tensor<string, []>("input_105_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, []>(135147328)))]; |
| 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, []>(135148928)))]; |
| tensor<fp16, [1, 77, 768]> input_105_cast_fp16 = layer_norm(axes = input_105_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_103_cast_fp16)[name = tensor<string, []>("input_105_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135150528))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137509888))), 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, []>(137510464)))]; |
| 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_105_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")]; |
| tensor<fp16, []> var_665_to_fp16 = const()[name = tensor<string, []>("op_665_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_666_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_665_to_fp16)[name = tensor<string, []>("op_666_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_667_cast_fp16 = sigmoid(x = var_666_cast_fp16)[name = tensor<string, []>("op_667_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_109_cast_fp16 = mul(x = linear_52_cast_fp16, y = var_667_cast_fp16)[name = tensor<string, []>("input_109_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137516672))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139876032))), 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, []>(139876608)))]; |
| 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_109_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_111_cast_fp16 = add(x = input_103_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_111_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, []>(139878208)))]; |
| 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, []>(139879808)))]; |
| 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_111_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139881408))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140471296))), 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, []>(140471872)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140473472))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141063360))), 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, []>(141063936)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141065536))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141655424))), 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, []>(141656000)))]; |
| 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_697 = const()[name = tensor<string, []>("op_697"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_698_cast_fp16 = reshape(shape = var_697, x = linear_54_cast_fp16)[name = tensor<string, []>("op_698_cast_fp16")]; |
| tensor<int32, [4]> var_700 = const()[name = tensor<string, []>("op_700"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_701_cast_fp16 = reshape(shape = var_700, x = linear_55_cast_fp16)[name = tensor<string, []>("op_701_cast_fp16")]; |
| tensor<int32, [4]> var_703 = const()[name = tensor<string, []>("op_703"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_704_cast_fp16 = reshape(shape = var_703, x = linear_56_cast_fp16)[name = tensor<string, []>("op_704_cast_fp16")]; |
| tensor<int32, [4]> value_19_perm_0 = const()[name = tensor<string, []>("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_9_cast_fp16 = mul(x = var_698_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_9_cast_fp16")]; |
| tensor<bool, []> matmul_9_transpose_y_0 = const()[name = tensor<string, []>("matmul_9_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_9_transpose_x_0 = const()[name = tensor<string, []>("matmul_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_66_perm_0 = const()[name = tensor<string, []>("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_67_perm_0 = const()[name = tensor<string, []>("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_67 = transpose(perm = transpose_67_perm_0, x = var_701_cast_fp16)[name = tensor<string, []>("transpose_82")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_66 = transpose(perm = transpose_66_perm_0, x = mul_9_cast_fp16)[name = tensor<string, []>("transpose_83")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = transpose_66, y = transpose_67)[name = tensor<string, []>("matmul_9_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_9_cast_fp16 = add(x = matmul_9_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_9_cast_fp16")]; |
| tensor<int32, []> softmax_9_axis_0 = const()[name = tensor<string, []>("softmax_9_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_9_cast_fp16 = softmax(axis = softmax_9_axis_0, x = add_9_cast_fp16)[name = tensor<string, []>("softmax_9_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, [1, 12, 77, 64]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = var_704_cast_fp16)[name = tensor<string, []>("transpose_81")]; |
| tensor<fp16, [1, 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 = softmax_9_cast_fp16, y = value_19_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")]; |
| tensor<int32, [4]> var_712_perm_0 = const()[name = tensor<string, []>("op_712_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_714 = const()[name = tensor<string, []>("op_714"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_712_cast_fp16 = transpose(perm = var_712_perm_0, x = attn_output_37_cast_fp16)[name = tensor<string, []>("transpose_80")]; |
| tensor<fp16, [1, 77, 768]> var_715_cast_fp16 = reshape(shape = var_714, x = var_712_cast_fp16)[name = tensor<string, []>("op_715_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141657600))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142247488))), 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, []>(142248064)))]; |
| 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 = var_715_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_115_cast_fp16 = add(x = input_111_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; |
| tensor<int32, [1]> input_117_axes_0 = const()[name = tensor<string, []>("input_117_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, []>(142249664)))]; |
| 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, []>(142251264)))]; |
| tensor<fp16, [1, 77, 768]> input_117_cast_fp16 = layer_norm(axes = input_117_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_115_cast_fp16)[name = tensor<string, []>("input_117_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142252864))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144612224))), 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, []>(144612800)))]; |
| 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_117_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")]; |
| tensor<fp16, []> var_730_to_fp16 = const()[name = tensor<string, []>("op_730_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_731_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_730_to_fp16)[name = tensor<string, []>("op_731_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_732_cast_fp16 = sigmoid(x = var_731_cast_fp16)[name = tensor<string, []>("op_732_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_121_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_732_cast_fp16)[name = tensor<string, []>("input_121_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144619008))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146978368))), 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, []>(146978944)))]; |
| 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_121_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_123_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, []>(146980544)))]; |
| 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, []>(146982144)))]; |
| 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_123_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146983744))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147573632))), 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, []>(147574208)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147575808))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148165696))), 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, []>(148166272)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148167872))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148757760))), 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, []>(148758336)))]; |
| 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_762 = const()[name = tensor<string, []>("op_762"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_763_cast_fp16 = reshape(shape = var_762, x = linear_60_cast_fp16)[name = tensor<string, []>("op_763_cast_fp16")]; |
| tensor<int32, [4]> var_765 = const()[name = tensor<string, []>("op_765"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_766_cast_fp16 = reshape(shape = var_765, x = linear_61_cast_fp16)[name = tensor<string, []>("op_766_cast_fp16")]; |
| tensor<int32, [4]> var_768 = const()[name = tensor<string, []>("op_768"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_769_cast_fp16 = reshape(shape = var_768, x = linear_62_cast_fp16)[name = tensor<string, []>("op_769_cast_fp16")]; |
| tensor<int32, [4]> value_21_perm_0 = const()[name = tensor<string, []>("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_10_cast_fp16 = mul(x = var_763_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_10_cast_fp16")]; |
| tensor<bool, []> matmul_10_transpose_y_0 = const()[name = tensor<string, []>("matmul_10_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_10_transpose_x_0 = const()[name = tensor<string, []>("matmul_10_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_68_perm_0 = const()[name = tensor<string, []>("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_69_perm_0 = const()[name = tensor<string, []>("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_69 = transpose(perm = transpose_69_perm_0, x = var_766_cast_fp16)[name = tensor<string, []>("transpose_78")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_68 = transpose(perm = transpose_68_perm_0, x = mul_10_cast_fp16)[name = tensor<string, []>("transpose_79")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_68, y = transpose_69)[name = tensor<string, []>("matmul_10_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_10_cast_fp16 = add(x = matmul_10_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_10_cast_fp16")]; |
| tensor<int32, []> softmax_10_axis_0 = const()[name = tensor<string, []>("softmax_10_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_10_cast_fp16 = softmax(axis = softmax_10_axis_0, x = add_10_cast_fp16)[name = tensor<string, []>("softmax_10_cast_fp16")]; |
| tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = var_769_cast_fp16)[name = tensor<string, []>("transpose_77")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_41_cast_fp16 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = softmax_10_cast_fp16, y = value_21_cast_fp16)[name = tensor<string, []>("attn_output_41_cast_fp16")]; |
| tensor<int32, [4]> var_777_perm_0 = const()[name = tensor<string, []>("op_777_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_779 = const()[name = tensor<string, []>("op_779"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_777_cast_fp16 = transpose(perm = var_777_perm_0, x = attn_output_41_cast_fp16)[name = tensor<string, []>("transpose_76")]; |
| tensor<fp16, [1, 77, 768]> var_780_cast_fp16 = reshape(shape = var_779, x = var_777_cast_fp16)[name = tensor<string, []>("op_780_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148759936))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149349824))), 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, []>(149350400)))]; |
| 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 = var_780_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_127_cast_fp16 = add(x = input_123_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_127_cast_fp16")]; |
| tensor<int32, [1]> input_129_axes_0 = const()[name = tensor<string, []>("input_129_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, []>(149352000)))]; |
| 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, []>(149353600)))]; |
| tensor<fp16, [1, 77, 768]> input_129_cast_fp16 = layer_norm(axes = input_129_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_127_cast_fp16)[name = tensor<string, []>("input_129_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149355200))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151714560))), 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, []>(151715136)))]; |
| 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_129_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")]; |
| tensor<fp16, []> var_795_to_fp16 = const()[name = tensor<string, []>("op_795_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_796_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_795_to_fp16)[name = tensor<string, []>("op_796_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_797_cast_fp16 = sigmoid(x = var_796_cast_fp16)[name = tensor<string, []>("op_797_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_133_cast_fp16 = mul(x = linear_64_cast_fp16, y = var_797_cast_fp16)[name = tensor<string, []>("input_133_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151721344))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154080704))), 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, []>(154081280)))]; |
| 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_133_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_135_cast_fp16 = add(x = input_127_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")]; |
| 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, []>(154082880)))]; |
| 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, []>(154084480)))]; |
| 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_135_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154086080))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154675968))), 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, []>(154676544)))]; |
| 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, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154678144))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155268032))), 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, []>(155268608)))]; |
| 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<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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155270208))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155860096))), 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, []>(155860672)))]; |
| 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_827 = const()[name = tensor<string, []>("op_827"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_828_cast_fp16 = reshape(shape = var_827, x = linear_66_cast_fp16)[name = tensor<string, []>("op_828_cast_fp16")]; |
| tensor<int32, [4]> var_830 = const()[name = tensor<string, []>("op_830"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_831_cast_fp16 = reshape(shape = var_830, x = linear_67_cast_fp16)[name = tensor<string, []>("op_831_cast_fp16")]; |
| tensor<int32, [4]> var_833 = const()[name = tensor<string, []>("op_833"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 12, 64]> var_834_cast_fp16 = reshape(shape = var_833, x = linear_68_cast_fp16)[name = tensor<string, []>("op_834_cast_fp16")]; |
| tensor<int32, [4]> value_perm_0 = const()[name = tensor<string, []>("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 12, 64]> mul_11_cast_fp16 = mul(x = var_828_cast_fp16, y = var_17_to_fp16)[name = tensor<string, []>("mul_11_cast_fp16")]; |
| tensor<bool, []> matmul_11_transpose_y_0 = const()[name = tensor<string, []>("matmul_11_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_11_transpose_x_0 = const()[name = tensor<string, []>("matmul_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_70_perm_0 = const()[name = tensor<string, []>("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_71_perm_0 = const()[name = tensor<string, []>("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_71 = transpose(perm = transpose_71_perm_0, x = var_831_cast_fp16)[name = tensor<string, []>("transpose_74")]; |
| tensor<fp16, [1, 12, 77, 64]> transpose_70 = transpose(perm = transpose_70_perm_0, x = mul_11_cast_fp16)[name = tensor<string, []>("transpose_75")]; |
| tensor<fp16, [1, 12, 77, 77]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = transpose_70, y = transpose_71)[name = tensor<string, []>("matmul_11_cast_fp16")]; |
| tensor<fp16, [1, 12, 77, 77]> add_11_cast_fp16 = add(x = matmul_11_cast_fp16, y = op_56_to_fp16_palettized)[name = tensor<string, []>("add_11_cast_fp16")]; |
| tensor<int32, []> softmax_11_axis_0 = const()[name = tensor<string, []>("softmax_11_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp16, [1, 12, 77, 77]> softmax_11_cast_fp16 = softmax(axis = softmax_11_axis_0, x = add_11_cast_fp16)[name = tensor<string, []>("softmax_11_cast_fp16")]; |
| tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 12, 77, 64]> value_cast_fp16 = transpose(perm = value_perm_0, x = var_834_cast_fp16)[name = tensor<string, []>("transpose_73")]; |
| tensor<fp16, [1, 12, 77, 64]> attn_output_45_cast_fp16 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = softmax_11_cast_fp16, y = value_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")]; |
| tensor<int32, [4]> var_842_perm_0 = const()[name = tensor<string, []>("op_842_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_844 = const()[name = tensor<string, []>("op_844"), val = tensor<int32, [3]>([1, 77, 768])]; |
| tensor<fp16, [1, 77, 12, 64]> var_842_cast_fp16 = transpose(perm = var_842_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_72")]; |
| tensor<fp16, [1, 77, 768]> var_845_cast_fp16 = reshape(shape = var_844, x = var_842_cast_fp16)[name = tensor<string, []>("op_845_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, [589824]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155862272))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156452160))), 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, []>(156452736)))]; |
| 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 = var_845_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_139_cast_fp16 = add(x = input_135_cast_fp16, y = linear_69_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_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, []>(156454336)))]; |
| 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, []>(156455936)))]; |
| tensor<fp16, [1, 77, 768]> input_141_cast_fp16 = layer_norm(axes = input_141_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_139_cast_fp16)[name = tensor<string, []>("input_141_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156457536))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158816896))), 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, []>(158817472)))]; |
| 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_141_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")]; |
| tensor<fp16, []> var_860_to_fp16 = const()[name = tensor<string, []>("op_860_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; |
| tensor<fp16, [1, 77, 3072]> var_861_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_860_to_fp16)[name = tensor<string, []>("op_861_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> var_862_cast_fp16 = sigmoid(x = var_861_cast_fp16)[name = tensor<string, []>("op_862_cast_fp16")]; |
| tensor<fp16, [1, 77, 3072]> input_145_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_862_cast_fp16)[name = tensor<string, []>("input_145_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, [2359296]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158823680))), lut = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161183040))), 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, []>(161183616)))]; |
| 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_145_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")]; |
| tensor<fp16, [1, 77, 768]> input_cast_fp16 = add(x = input_139_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, []>(161185216)))]; |
| 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, []>(161186816)))]; |
| 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<string, []> last_hidden_state_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("last_hidden_state_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<int32, [1]> var_873 = const()[name = tensor<string, []>("op_873"), val = tensor<int32, [1]>([0])]; |
| tensor<int32, []> var_875_axis_0 = const()[name = tensor<string, []>("op_875_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<bool, []> var_875_keep_dims_0 = const()[name = tensor<string, []>("op_875_keep_dims_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [1]> var_875 = reduce_argmax(axis = var_875_axis_0, keep_dims = var_875_keep_dims_0, x = cast_1)[name = tensor<string, []>("op_875")]; |
| 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_873, var_875))[name = tensor<string, []>("stack_0")]; |
| tensor<int32, []> var_877_transpose_batch_dims_0 = const()[name = tensor<string, []>("op_877_transpose_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp16, [1, 768]> var_877_transpose_cast_fp16 = gather_nd(batch_dims = var_877_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("op_877_transpose_cast_fp16")]; |
| tensor<string, []> var_877_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_877_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, [1, 77, 768]> last_hidden_state = cast(dtype = last_hidden_state_cast_fp16_to_fp32_dtype_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("cast_0")]; |
| tensor<fp32, [1, 768]> pooled_outputs = cast(dtype = var_877_cast_fp16_to_fp32_dtype_0, x = var_877_transpose_cast_fp16)[name = tensor<string, []>("cast_1")]; |
| } -> (last_hidden_state, pooled_outputs); |
| } |
