| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}})] | |
| { | |
| func main<ios16>(tensor<fp32, [1, 77]> input_ids) { | |
| tensor<int32, []> var_5 = const()[name = tensor<string, []>("op_5"), val = tensor<int32, []>(-1)]; | |
| tensor<bool, []> var_6 = const()[name = tensor<string, []>("op_6"), val = tensor<bool, []>(false)]; | |
| tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("int32")]; | |
| tensor<int32, []> inputs_embeds_axis_0 = const()[name = tensor<string, []>("inputs_embeds_axis_0"), val = tensor<int32, []>(0)]; | |
| tensor<int32, []> inputs_embeds_batch_dims_0 = const()[name = tensor<string, []>("inputs_embeds_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [49408, 768]> text_encoder_text_model_embeddings_token_embedding_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_embeddings_token_embedding_weight_to_fp16"), val = tensor<fp16, [49408, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<int32, [1, 77]> cast_2 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor<string, []>("cast_2")]; | |
| tensor<fp16, [1, 77, 768]> inputs_embeds_cast = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_2, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_cast")]; | |
| tensor<fp16, [1, 77, 768]> position_embeddings_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [44352]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75890816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935232))), name = tensor<string, []>("position_embeddings_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 768]> input_3_cast = add(x = inputs_embeds_cast, y = position_embeddings_to_fp16_palettized)[name = tensor<string, []>("input_3_cast")]; | |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75935424)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75937024)))]; | |
| tensor<fp16, []> var_15_to_fp16 = const()[name = tensor<string, []>("op_15_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_1_cast = 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)[name = tensor<string, []>("hidden_states_1_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75938624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381056))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76381248)))]; | |
| tensor<fp16, [1, 77, 768]> var_106_cast = 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)[name = tensor<string, []>("op_106_cast")]; | |
| tensor<fp16, []> var_107_to_fp16 = const()[name = tensor<string, []>("op_107_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_5_cast = mul(x = var_106_cast, y = var_107_to_fp16)[name = tensor<string, []>("tensor_5_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76382848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825280))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76825472)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_1_cast = 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)[name = tensor<string, []>("tensor_1_cast")]; | |
| tensor<int32, [4]> var_112 = const()[name = tensor<string, []>("op_112"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_113_cast = reshape(shape = var_112, x = tensor_1_cast)[name = tensor<string, []>("op_113_cast")]; | |
| tensor<int32, [4]> var_114_perm_0 = const()[name = tensor<string, []>("op_114_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76827072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269504))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77269696)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_3_cast = 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)[name = tensor<string, []>("tensor_3_cast")]; | |
| tensor<int32, [4]> var_119 = const()[name = tensor<string, []>("op_119"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_120_cast = reshape(shape = var_119, x = tensor_3_cast)[name = tensor<string, []>("op_120_cast")]; | |
| tensor<int32, [4]> var_121_perm_0 = const()[name = tensor<string, []>("op_121_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_128 = const()[name = tensor<string, []>("op_128"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_129_cast = reshape(shape = var_128, x = tensor_5_cast)[name = tensor<string, []>("op_129_cast")]; | |
| tensor<int32, [4]> var_130_perm_0 = const()[name = tensor<string, []>("op_130_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_132 = const()[name = tensor<string, []>("op_132"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_59 = transpose(perm = var_130_perm_0, x = var_129_cast)[name = tensor<string, []>("transpose_59")]; | |
| tensor<fp16, [12, 77, 64]> query_states_1_cast = reshape(shape = var_132, x = transpose_59)[name = tensor<string, []>("query_states_1_cast")]; | |
| tensor<int32, [3]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_58 = transpose(perm = var_114_perm_0, x = var_113_cast)[name = tensor<string, []>("transpose_58")]; | |
| tensor<fp16, [12, 77, 64]> key_states_3_cast = reshape(shape = var_134, x = transpose_58)[name = tensor<string, []>("key_states_3_cast")]; | |
| tensor<int32, [3]> var_136 = const()[name = tensor<string, []>("op_136"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_57 = transpose(perm = var_121_perm_0, x = var_120_cast)[name = tensor<string, []>("transpose_57")]; | |
| tensor<fp16, [12, 77, 64]> value_states_3_cast = reshape(shape = var_136, x = transpose_57)[name = tensor<string, []>("value_states_3_cast")]; | |
| tensor<int32, [3]> var_139_perm_0 = const()[name = tensor<string, []>("op_139_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_1_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_1_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_56 = transpose(perm = var_139_perm_0, x = key_states_3_cast)[name = tensor<string, []>("transpose_56")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_1_cast = matmul(transpose_x = attn_weights_1_transpose_x_0, transpose_y = attn_weights_1_transpose_y_0, x = query_states_1_cast, y = transpose_56)[name = tensor<string, []>("attn_weights_1_cast")]; | |
| tensor<int32, [4]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_142_cast = reshape(shape = var_141, x = attn_weights_1_cast)[name = tensor<string, []>("op_142_cast")]; | |
| tensor<fp16, [1, 1, 77, 77]> op_56_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4447]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77271296))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77275840))), name = tensor<string, []>("op_56_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_3_cast = add(x = var_142_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_3_cast")]; | |
| tensor<int32, [3]> var_147 = const()[name = tensor<string, []>("op_147"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_5_cast = reshape(shape = var_147, x = attn_weights_3_cast)[name = tensor<string, []>("input_5_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_7_cast = softmax(axis = var_5, x = input_5_cast)[name = tensor<string, []>("input_7_cast")]; | |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_1_cast = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast, y = value_states_3_cast)[name = tensor<string, []>("attn_output_1_cast")]; | |
| tensor<int32, [4]> var_152 = const()[name = tensor<string, []>("op_152"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_3_cast = reshape(shape = var_152, x = attn_output_1_cast)[name = tensor<string, []>("attn_output_3_cast")]; | |
| tensor<int32, [4]> attn_output_5_perm_0 = const()[name = tensor<string, []>("attn_output_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_155 = const()[name = tensor<string, []>("op_155"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_55 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast)[name = tensor<string, []>("transpose_55")]; | |
| tensor<fp16, [1, 77, 768]> input_9_cast = reshape(shape = var_155, x = transpose_55)[name = tensor<string, []>("input_9_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77276032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([768, 768])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77718656)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_3_cast = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_9_cast)[name = tensor<string, []>("hidden_states_3_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_11_cast = add(x = input_3_cast, y = hidden_states_3_cast)[name = tensor<string, []>("input_11_cast")]; | |
| tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77720256)))]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77721856)))]; | |
| tensor<fp16, [1, 77, 768]> input_13_cast = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast)[name = tensor<string, []>("input_13_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77723456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79492992))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([3072, 768])]; | |
| tensor<fp16, [3072]> text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79493184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79495552))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_15_cast = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_13_cast)[name = tensor<string, []>("input_15_cast")]; | |
| tensor<fp16, []> var_170_to_fp16 = const()[name = tensor<string, []>("op_170_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_171_cast = mul(x = input_15_cast, y = var_170_to_fp16)[name = tensor<string, []>("op_171_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_172_cast = sigmoid(x = var_171_cast)[name = tensor<string, []>("op_172_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_17_cast = mul(x = input_15_cast, y = var_172_cast)[name = tensor<string, []>("input_17_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79495744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81265280))), 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, []>(81265472)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_5_cast = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_17_cast)[name = tensor<string, []>("hidden_states_5_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_19_cast = add(x = input_11_cast, y = hidden_states_5_cast)[name = tensor<string, []>("input_19_cast")]; | |
| 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, []>(81267072)))]; | |
| 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, []>(81268672)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_7_cast = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast)[name = tensor<string, []>("hidden_states_7_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81270272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81712704))), 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, []>(81712896)))]; | |
| tensor<fp16, [1, 77, 768]> var_196_cast = 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)[name = tensor<string, []>("op_196_cast")]; | |
| tensor<fp16, []> var_197_to_fp16 = const()[name = tensor<string, []>("op_197_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_11_cast = mul(x = var_196_cast, y = var_197_to_fp16)[name = tensor<string, []>("tensor_11_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81714496))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82156928))), 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, []>(82157120)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_7_cast = 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)[name = tensor<string, []>("tensor_7_cast")]; | |
| tensor<int32, [4]> var_202 = const()[name = tensor<string, []>("op_202"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_203_cast = reshape(shape = var_202, x = tensor_7_cast)[name = tensor<string, []>("op_203_cast")]; | |
| tensor<int32, [4]> var_204_perm_0 = const()[name = tensor<string, []>("op_204_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82158720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82601152))), 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, []>(82601344)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_9_cast = 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)[name = tensor<string, []>("tensor_9_cast")]; | |
| tensor<int32, [4]> var_209 = const()[name = tensor<string, []>("op_209"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_210_cast = reshape(shape = var_209, x = tensor_9_cast)[name = tensor<string, []>("op_210_cast")]; | |
| tensor<int32, [4]> var_211_perm_0 = const()[name = tensor<string, []>("op_211_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_218 = const()[name = tensor<string, []>("op_218"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_219_cast = reshape(shape = var_218, x = tensor_11_cast)[name = tensor<string, []>("op_219_cast")]; | |
| tensor<int32, [4]> var_220_perm_0 = const()[name = tensor<string, []>("op_220_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_222 = const()[name = tensor<string, []>("op_222"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_54 = transpose(perm = var_220_perm_0, x = var_219_cast)[name = tensor<string, []>("transpose_54")]; | |
| tensor<fp16, [12, 77, 64]> query_states_3_cast = reshape(shape = var_222, x = transpose_54)[name = tensor<string, []>("query_states_3_cast")]; | |
| tensor<int32, [3]> var_224 = const()[name = tensor<string, []>("op_224"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_53 = transpose(perm = var_204_perm_0, x = var_203_cast)[name = tensor<string, []>("transpose_53")]; | |
| tensor<fp16, [12, 77, 64]> key_states_7_cast = reshape(shape = var_224, x = transpose_53)[name = tensor<string, []>("key_states_7_cast")]; | |
| tensor<int32, [3]> var_226 = const()[name = tensor<string, []>("op_226"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_52 = transpose(perm = var_211_perm_0, x = var_210_cast)[name = tensor<string, []>("transpose_52")]; | |
| tensor<fp16, [12, 77, 64]> value_states_7_cast = reshape(shape = var_226, x = transpose_52)[name = tensor<string, []>("value_states_7_cast")]; | |
| tensor<int32, [3]> var_229_perm_0 = const()[name = tensor<string, []>("op_229_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_7_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_7_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_51 = transpose(perm = var_229_perm_0, x = key_states_7_cast)[name = tensor<string, []>("transpose_51")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_7_cast = matmul(transpose_x = attn_weights_7_transpose_x_0, transpose_y = attn_weights_7_transpose_y_0, x = query_states_3_cast, y = transpose_51)[name = tensor<string, []>("attn_weights_7_cast")]; | |
| tensor<int32, [4]> var_231 = const()[name = tensor<string, []>("op_231"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_232_cast = reshape(shape = var_231, x = attn_weights_7_cast)[name = tensor<string, []>("op_232_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_9_cast = add(x = var_232_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_9_cast")]; | |
| tensor<int32, [3]> var_237 = const()[name = tensor<string, []>("op_237"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_21_cast = reshape(shape = var_237, x = attn_weights_9_cast)[name = tensor<string, []>("input_21_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_23_cast = softmax(axis = var_5, x = input_21_cast)[name = tensor<string, []>("input_23_cast")]; | |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_7_cast = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast, y = value_states_7_cast)[name = tensor<string, []>("attn_output_7_cast")]; | |
| tensor<int32, [4]> var_242 = const()[name = tensor<string, []>("op_242"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_9_cast = reshape(shape = var_242, x = attn_output_7_cast)[name = tensor<string, []>("attn_output_9_cast")]; | |
| tensor<int32, [4]> attn_output_11_perm_0 = const()[name = tensor<string, []>("attn_output_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_245 = const()[name = tensor<string, []>("op_245"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_50 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast)[name = tensor<string, []>("transpose_50")]; | |
| tensor<fp16, [1, 77, 768]> input_25_cast = reshape(shape = var_245, x = transpose_50)[name = tensor<string, []>("input_25_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82602944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83045376))), 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, []>(83045568)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_9_cast = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast)[name = tensor<string, []>("hidden_states_9_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_27_cast = add(x = input_19_cast, y = hidden_states_9_cast)[name = tensor<string, []>("input_27_cast")]; | |
| tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83047168)))]; | |
| 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, []>(83048768)))]; | |
| tensor<fp16, [1, 77, 768]> input_29_cast = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast)[name = tensor<string, []>("input_29_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83050368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84819904))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84820096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84822464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_31_cast = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_29_cast)[name = tensor<string, []>("input_31_cast")]; | |
| tensor<fp16, []> var_260_to_fp16 = const()[name = tensor<string, []>("op_260_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_261_cast = mul(x = input_31_cast, y = var_260_to_fp16)[name = tensor<string, []>("op_261_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_262_cast = sigmoid(x = var_261_cast)[name = tensor<string, []>("op_262_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_33_cast = mul(x = input_31_cast, y = var_262_cast)[name = tensor<string, []>("input_33_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84822656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86592192))), 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, []>(86592384)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_11_cast = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_33_cast)[name = tensor<string, []>("hidden_states_11_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_35_cast = add(x = input_27_cast, y = hidden_states_11_cast)[name = tensor<string, []>("input_35_cast")]; | |
| 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, []>(86593984)))]; | |
| 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, []>(86595584)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_13_cast = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast)[name = tensor<string, []>("hidden_states_13_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86597184))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87039616))), 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, []>(87039808)))]; | |
| tensor<fp16, [1, 77, 768]> var_286_cast = 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)[name = tensor<string, []>("op_286_cast")]; | |
| tensor<fp16, []> var_287_to_fp16 = const()[name = tensor<string, []>("op_287_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_17_cast = mul(x = var_286_cast, y = var_287_to_fp16)[name = tensor<string, []>("tensor_17_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87041408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87483840))), 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, []>(87484032)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_13_cast = 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)[name = tensor<string, []>("tensor_13_cast")]; | |
| tensor<int32, [4]> var_292 = const()[name = tensor<string, []>("op_292"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_293_cast = reshape(shape = var_292, x = tensor_13_cast)[name = tensor<string, []>("op_293_cast")]; | |
| tensor<int32, [4]> var_294_perm_0 = const()[name = tensor<string, []>("op_294_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87485632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87928064))), 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, []>(87928256)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_15_cast = 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)[name = tensor<string, []>("tensor_15_cast")]; | |
| tensor<int32, [4]> var_299 = const()[name = tensor<string, []>("op_299"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_300_cast = reshape(shape = var_299, x = tensor_15_cast)[name = tensor<string, []>("op_300_cast")]; | |
| tensor<int32, [4]> var_301_perm_0 = const()[name = tensor<string, []>("op_301_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_308 = const()[name = tensor<string, []>("op_308"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_309_cast = reshape(shape = var_308, x = tensor_17_cast)[name = tensor<string, []>("op_309_cast")]; | |
| tensor<int32, [4]> var_310_perm_0 = const()[name = tensor<string, []>("op_310_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_312 = const()[name = tensor<string, []>("op_312"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_49 = transpose(perm = var_310_perm_0, x = var_309_cast)[name = tensor<string, []>("transpose_49")]; | |
| tensor<fp16, [12, 77, 64]> query_states_5_cast = reshape(shape = var_312, x = transpose_49)[name = tensor<string, []>("query_states_5_cast")]; | |
| tensor<int32, [3]> var_314 = const()[name = tensor<string, []>("op_314"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_48 = transpose(perm = var_294_perm_0, x = var_293_cast)[name = tensor<string, []>("transpose_48")]; | |
| tensor<fp16, [12, 77, 64]> key_states_11_cast = reshape(shape = var_314, x = transpose_48)[name = tensor<string, []>("key_states_11_cast")]; | |
| tensor<int32, [3]> var_316 = const()[name = tensor<string, []>("op_316"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_47 = transpose(perm = var_301_perm_0, x = var_300_cast)[name = tensor<string, []>("transpose_47")]; | |
| tensor<fp16, [12, 77, 64]> value_states_11_cast = reshape(shape = var_316, x = transpose_47)[name = tensor<string, []>("value_states_11_cast")]; | |
| tensor<int32, [3]> var_319_perm_0 = const()[name = tensor<string, []>("op_319_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_13_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_13_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_46 = transpose(perm = var_319_perm_0, x = key_states_11_cast)[name = tensor<string, []>("transpose_46")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_13_cast = matmul(transpose_x = attn_weights_13_transpose_x_0, transpose_y = attn_weights_13_transpose_y_0, x = query_states_5_cast, y = transpose_46)[name = tensor<string, []>("attn_weights_13_cast")]; | |
| tensor<int32, [4]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_322_cast = reshape(shape = var_321, x = attn_weights_13_cast)[name = tensor<string, []>("op_322_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_15_cast = add(x = var_322_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_15_cast")]; | |
| tensor<int32, [3]> var_327 = const()[name = tensor<string, []>("op_327"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_37_cast = reshape(shape = var_327, x = attn_weights_15_cast)[name = tensor<string, []>("input_37_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_39_cast = softmax(axis = var_5, x = input_37_cast)[name = tensor<string, []>("input_39_cast")]; | |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_13_cast = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast, y = value_states_11_cast)[name = tensor<string, []>("attn_output_13_cast")]; | |
| tensor<int32, [4]> var_332 = const()[name = tensor<string, []>("op_332"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_15_cast = reshape(shape = var_332, x = attn_output_13_cast)[name = tensor<string, []>("attn_output_15_cast")]; | |
| tensor<int32, [4]> attn_output_17_perm_0 = const()[name = tensor<string, []>("attn_output_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_335 = const()[name = tensor<string, []>("op_335"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_45 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast)[name = tensor<string, []>("transpose_45")]; | |
| tensor<fp16, [1, 77, 768]> input_41_cast = reshape(shape = var_335, x = transpose_45)[name = tensor<string, []>("input_41_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87929856))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88372288))), 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, []>(88372480)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_15_cast = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_41_cast)[name = tensor<string, []>("hidden_states_15_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_43_cast = add(x = input_35_cast, y = hidden_states_15_cast)[name = tensor<string, []>("input_43_cast")]; | |
| tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88374080)))]; | |
| 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, []>(88375680)))]; | |
| tensor<fp16, [1, 77, 768]> input_45_cast = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast)[name = tensor<string, []>("input_45_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88377280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90146816))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90147008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90149376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_47_cast = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast)[name = tensor<string, []>("input_47_cast")]; | |
| tensor<fp16, []> var_350_to_fp16 = const()[name = tensor<string, []>("op_350_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_351_cast = mul(x = input_47_cast, y = var_350_to_fp16)[name = tensor<string, []>("op_351_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_352_cast = sigmoid(x = var_351_cast)[name = tensor<string, []>("op_352_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_49_cast = mul(x = input_47_cast, y = var_352_cast)[name = tensor<string, []>("input_49_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90149568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91919104))), 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, []>(91919296)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_17_cast = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast)[name = tensor<string, []>("hidden_states_17_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_51_cast = add(x = input_43_cast, y = hidden_states_17_cast)[name = tensor<string, []>("input_51_cast")]; | |
| 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, []>(91920896)))]; | |
| 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, []>(91922496)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_19_cast = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast)[name = tensor<string, []>("hidden_states_19_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91924096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92366528))), 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, []>(92366720)))]; | |
| tensor<fp16, [1, 77, 768]> var_376_cast = 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)[name = tensor<string, []>("op_376_cast")]; | |
| tensor<fp16, []> var_377_to_fp16 = const()[name = tensor<string, []>("op_377_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_23_cast = mul(x = var_376_cast, y = var_377_to_fp16)[name = tensor<string, []>("tensor_23_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92368320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92810752))), 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, []>(92810944)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_19_cast = 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)[name = tensor<string, []>("tensor_19_cast")]; | |
| tensor<int32, [4]> var_382 = const()[name = tensor<string, []>("op_382"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_383_cast = reshape(shape = var_382, x = tensor_19_cast)[name = tensor<string, []>("op_383_cast")]; | |
| tensor<int32, [4]> var_384_perm_0 = const()[name = tensor<string, []>("op_384_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92812544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93254976))), 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, []>(93255168)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_21_cast = 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)[name = tensor<string, []>("tensor_21_cast")]; | |
| tensor<int32, [4]> var_389 = const()[name = tensor<string, []>("op_389"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_390_cast = reshape(shape = var_389, x = tensor_21_cast)[name = tensor<string, []>("op_390_cast")]; | |
| tensor<int32, [4]> var_391_perm_0 = const()[name = tensor<string, []>("op_391_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_398 = const()[name = tensor<string, []>("op_398"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_399_cast = reshape(shape = var_398, x = tensor_23_cast)[name = tensor<string, []>("op_399_cast")]; | |
| tensor<int32, [4]> var_400_perm_0 = const()[name = tensor<string, []>("op_400_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_402 = const()[name = tensor<string, []>("op_402"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_44 = transpose(perm = var_400_perm_0, x = var_399_cast)[name = tensor<string, []>("transpose_44")]; | |
| tensor<fp16, [12, 77, 64]> query_states_7_cast = reshape(shape = var_402, x = transpose_44)[name = tensor<string, []>("query_states_7_cast")]; | |
| tensor<int32, [3]> var_404 = const()[name = tensor<string, []>("op_404"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_43 = transpose(perm = var_384_perm_0, x = var_383_cast)[name = tensor<string, []>("transpose_43")]; | |
| tensor<fp16, [12, 77, 64]> key_states_15_cast = reshape(shape = var_404, x = transpose_43)[name = tensor<string, []>("key_states_15_cast")]; | |
| tensor<int32, [3]> var_406 = const()[name = tensor<string, []>("op_406"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_42 = transpose(perm = var_391_perm_0, x = var_390_cast)[name = tensor<string, []>("transpose_42")]; | |
| tensor<fp16, [12, 77, 64]> value_states_15_cast = reshape(shape = var_406, x = transpose_42)[name = tensor<string, []>("value_states_15_cast")]; | |
| tensor<int32, [3]> var_409_perm_0 = const()[name = tensor<string, []>("op_409_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_19_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_19_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_41 = transpose(perm = var_409_perm_0, x = key_states_15_cast)[name = tensor<string, []>("transpose_41")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_19_cast = matmul(transpose_x = attn_weights_19_transpose_x_0, transpose_y = attn_weights_19_transpose_y_0, x = query_states_7_cast, y = transpose_41)[name = tensor<string, []>("attn_weights_19_cast")]; | |
| tensor<int32, [4]> var_411 = const()[name = tensor<string, []>("op_411"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_412_cast = reshape(shape = var_411, x = attn_weights_19_cast)[name = tensor<string, []>("op_412_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_21_cast = add(x = var_412_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_21_cast")]; | |
| tensor<int32, [3]> var_417 = const()[name = tensor<string, []>("op_417"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_53_cast = reshape(shape = var_417, x = attn_weights_21_cast)[name = tensor<string, []>("input_53_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_55_cast = softmax(axis = var_5, x = input_53_cast)[name = tensor<string, []>("input_55_cast")]; | |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_19_cast = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast, y = value_states_15_cast)[name = tensor<string, []>("attn_output_19_cast")]; | |
| tensor<int32, [4]> var_422 = const()[name = tensor<string, []>("op_422"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_21_cast = reshape(shape = var_422, x = attn_output_19_cast)[name = tensor<string, []>("attn_output_21_cast")]; | |
| tensor<int32, [4]> attn_output_23_perm_0 = const()[name = tensor<string, []>("attn_output_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_425 = const()[name = tensor<string, []>("op_425"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_40 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast)[name = tensor<string, []>("transpose_40")]; | |
| tensor<fp16, [1, 77, 768]> input_57_cast = reshape(shape = var_425, x = transpose_40)[name = tensor<string, []>("input_57_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93256768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93699200))), 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, []>(93699392)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_21_cast = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_57_cast)[name = tensor<string, []>("hidden_states_21_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_59_cast = add(x = input_51_cast, y = hidden_states_21_cast)[name = tensor<string, []>("input_59_cast")]; | |
| tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93700992)))]; | |
| 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, []>(93702592)))]; | |
| tensor<fp16, [1, 77, 768]> input_61_cast = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast)[name = tensor<string, []>("input_61_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93704192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95473728))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95473920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95476288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_63_cast = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_61_cast)[name = tensor<string, []>("input_63_cast")]; | |
| tensor<fp16, []> var_440_to_fp16 = const()[name = tensor<string, []>("op_440_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_441_cast = mul(x = input_63_cast, y = var_440_to_fp16)[name = tensor<string, []>("op_441_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_442_cast = sigmoid(x = var_441_cast)[name = tensor<string, []>("op_442_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_65_cast = mul(x = input_63_cast, y = var_442_cast)[name = tensor<string, []>("input_65_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95476480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97246016))), 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, []>(97246208)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_23_cast = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_65_cast)[name = tensor<string, []>("hidden_states_23_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_67_cast = add(x = input_59_cast, y = hidden_states_23_cast)[name = tensor<string, []>("input_67_cast")]; | |
| 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, []>(97247808)))]; | |
| 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, []>(97249408)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_25_cast = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast)[name = tensor<string, []>("hidden_states_25_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97251008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97693440))), 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, []>(97693632)))]; | |
| tensor<fp16, [1, 77, 768]> var_466_cast = 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)[name = tensor<string, []>("op_466_cast")]; | |
| tensor<fp16, []> var_467_to_fp16 = const()[name = tensor<string, []>("op_467_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_29_cast = mul(x = var_466_cast, y = var_467_to_fp16)[name = tensor<string, []>("tensor_29_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97695232))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98137664))), 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, []>(98137856)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_25_cast = 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)[name = tensor<string, []>("tensor_25_cast")]; | |
| tensor<int32, [4]> var_472 = const()[name = tensor<string, []>("op_472"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_473_cast = reshape(shape = var_472, x = tensor_25_cast)[name = tensor<string, []>("op_473_cast")]; | |
| tensor<int32, [4]> var_474_perm_0 = const()[name = tensor<string, []>("op_474_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98139456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98581888))), 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, []>(98582080)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_27_cast = 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)[name = tensor<string, []>("tensor_27_cast")]; | |
| tensor<int32, [4]> var_479 = const()[name = tensor<string, []>("op_479"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_480_cast = reshape(shape = var_479, x = tensor_27_cast)[name = tensor<string, []>("op_480_cast")]; | |
| tensor<int32, [4]> var_481_perm_0 = const()[name = tensor<string, []>("op_481_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_488 = const()[name = tensor<string, []>("op_488"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_489_cast = reshape(shape = var_488, x = tensor_29_cast)[name = tensor<string, []>("op_489_cast")]; | |
| tensor<int32, [4]> var_490_perm_0 = const()[name = tensor<string, []>("op_490_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_492 = const()[name = tensor<string, []>("op_492"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_39 = transpose(perm = var_490_perm_0, x = var_489_cast)[name = tensor<string, []>("transpose_39")]; | |
| tensor<fp16, [12, 77, 64]> query_states_9_cast = reshape(shape = var_492, x = transpose_39)[name = tensor<string, []>("query_states_9_cast")]; | |
| tensor<int32, [3]> var_494 = const()[name = tensor<string, []>("op_494"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_38 = transpose(perm = var_474_perm_0, x = var_473_cast)[name = tensor<string, []>("transpose_38")]; | |
| tensor<fp16, [12, 77, 64]> key_states_19_cast = reshape(shape = var_494, x = transpose_38)[name = tensor<string, []>("key_states_19_cast")]; | |
| tensor<int32, [3]> var_496 = const()[name = tensor<string, []>("op_496"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_37 = transpose(perm = var_481_perm_0, x = var_480_cast)[name = tensor<string, []>("transpose_37")]; | |
| tensor<fp16, [12, 77, 64]> value_states_19_cast = reshape(shape = var_496, x = transpose_37)[name = tensor<string, []>("value_states_19_cast")]; | |
| tensor<int32, [3]> var_499_perm_0 = const()[name = tensor<string, []>("op_499_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_25_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_25_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_36 = transpose(perm = var_499_perm_0, x = key_states_19_cast)[name = tensor<string, []>("transpose_36")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_25_cast = matmul(transpose_x = attn_weights_25_transpose_x_0, transpose_y = attn_weights_25_transpose_y_0, x = query_states_9_cast, y = transpose_36)[name = tensor<string, []>("attn_weights_25_cast")]; | |
| tensor<int32, [4]> var_501 = const()[name = tensor<string, []>("op_501"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_502_cast = reshape(shape = var_501, x = attn_weights_25_cast)[name = tensor<string, []>("op_502_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_27_cast = add(x = var_502_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_27_cast")]; | |
| tensor<int32, [3]> var_507 = const()[name = tensor<string, []>("op_507"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_69_cast = reshape(shape = var_507, x = attn_weights_27_cast)[name = tensor<string, []>("input_69_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_71_cast = softmax(axis = var_5, x = input_69_cast)[name = tensor<string, []>("input_71_cast")]; | |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_25_cast = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast, y = value_states_19_cast)[name = tensor<string, []>("attn_output_25_cast")]; | |
| tensor<int32, [4]> var_512 = const()[name = tensor<string, []>("op_512"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_27_cast = reshape(shape = var_512, x = attn_output_25_cast)[name = tensor<string, []>("attn_output_27_cast")]; | |
| tensor<int32, [4]> attn_output_29_perm_0 = const()[name = tensor<string, []>("attn_output_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_515 = const()[name = tensor<string, []>("op_515"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_35 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast)[name = tensor<string, []>("transpose_35")]; | |
| tensor<fp16, [1, 77, 768]> input_73_cast = reshape(shape = var_515, x = transpose_35)[name = tensor<string, []>("input_73_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98583680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99026112))), 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, []>(99026304)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_27_cast = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast)[name = tensor<string, []>("hidden_states_27_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_75_cast = add(x = input_67_cast, y = hidden_states_27_cast)[name = tensor<string, []>("input_75_cast")]; | |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99027904)))]; | |
| 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, []>(99029504)))]; | |
| tensor<fp16, [1, 77, 768]> input_77_cast = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast)[name = tensor<string, []>("input_77_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99031104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100800640))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100800832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100803200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_79_cast = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_77_cast)[name = tensor<string, []>("input_79_cast")]; | |
| tensor<fp16, []> var_530_to_fp16 = const()[name = tensor<string, []>("op_530_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_531_cast = mul(x = input_79_cast, y = var_530_to_fp16)[name = tensor<string, []>("op_531_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_532_cast = sigmoid(x = var_531_cast)[name = tensor<string, []>("op_532_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_81_cast = mul(x = input_79_cast, y = var_532_cast)[name = tensor<string, []>("input_81_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100803392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102572928))), 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, []>(102573120)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_29_cast = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_81_cast)[name = tensor<string, []>("hidden_states_29_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_83_cast = add(x = input_75_cast, y = hidden_states_29_cast)[name = tensor<string, []>("input_83_cast")]; | |
| 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, []>(102574720)))]; | |
| 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, []>(102576320)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_31_cast = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast)[name = tensor<string, []>("hidden_states_31_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102577920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103020352))), 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, []>(103020544)))]; | |
| tensor<fp16, [1, 77, 768]> var_556_cast = 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)[name = tensor<string, []>("op_556_cast")]; | |
| tensor<fp16, []> var_557_to_fp16 = const()[name = tensor<string, []>("op_557_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_35_cast = mul(x = var_556_cast, y = var_557_to_fp16)[name = tensor<string, []>("tensor_35_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103022144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103464576))), 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, []>(103464768)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_31_cast = 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)[name = tensor<string, []>("tensor_31_cast")]; | |
| tensor<int32, [4]> var_562 = const()[name = tensor<string, []>("op_562"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_563_cast = reshape(shape = var_562, x = tensor_31_cast)[name = tensor<string, []>("op_563_cast")]; | |
| tensor<int32, [4]> var_564_perm_0 = const()[name = tensor<string, []>("op_564_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103466368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103908800))), 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, []>(103908992)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_33_cast = 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)[name = tensor<string, []>("tensor_33_cast")]; | |
| tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_570_cast = reshape(shape = var_569, x = tensor_33_cast)[name = tensor<string, []>("op_570_cast")]; | |
| tensor<int32, [4]> var_571_perm_0 = const()[name = tensor<string, []>("op_571_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_578 = const()[name = tensor<string, []>("op_578"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_579_cast = reshape(shape = var_578, x = tensor_35_cast)[name = tensor<string, []>("op_579_cast")]; | |
| tensor<int32, [4]> var_580_perm_0 = const()[name = tensor<string, []>("op_580_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_582 = const()[name = tensor<string, []>("op_582"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_34 = transpose(perm = var_580_perm_0, x = var_579_cast)[name = tensor<string, []>("transpose_34")]; | |
| tensor<fp16, [12, 77, 64]> query_states_11_cast = reshape(shape = var_582, x = transpose_34)[name = tensor<string, []>("query_states_11_cast")]; | |
| tensor<int32, [3]> var_584 = const()[name = tensor<string, []>("op_584"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_33 = transpose(perm = var_564_perm_0, x = var_563_cast)[name = tensor<string, []>("transpose_33")]; | |
| tensor<fp16, [12, 77, 64]> key_states_23_cast = reshape(shape = var_584, x = transpose_33)[name = tensor<string, []>("key_states_23_cast")]; | |
| tensor<int32, [3]> var_586 = const()[name = tensor<string, []>("op_586"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_32 = transpose(perm = var_571_perm_0, x = var_570_cast)[name = tensor<string, []>("transpose_32")]; | |
| tensor<fp16, [12, 77, 64]> value_states_23_cast = reshape(shape = var_586, x = transpose_32)[name = tensor<string, []>("value_states_23_cast")]; | |
| tensor<int32, [3]> var_589_perm_0 = const()[name = tensor<string, []>("op_589_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_31_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_31_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_31 = transpose(perm = var_589_perm_0, x = key_states_23_cast)[name = tensor<string, []>("transpose_31")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_31_cast = matmul(transpose_x = attn_weights_31_transpose_x_0, transpose_y = attn_weights_31_transpose_y_0, x = query_states_11_cast, y = transpose_31)[name = tensor<string, []>("attn_weights_31_cast")]; | |
| tensor<int32, [4]> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_592_cast = reshape(shape = var_591, x = attn_weights_31_cast)[name = tensor<string, []>("op_592_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_33_cast = add(x = var_592_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_33_cast")]; | |
| tensor<int32, [3]> var_597 = const()[name = tensor<string, []>("op_597"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_85_cast = reshape(shape = var_597, x = attn_weights_33_cast)[name = tensor<string, []>("input_85_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_87_cast = softmax(axis = var_5, x = input_85_cast)[name = tensor<string, []>("input_87_cast")]; | |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_31_cast = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast, y = value_states_23_cast)[name = tensor<string, []>("attn_output_31_cast")]; | |
| tensor<int32, [4]> var_602 = const()[name = tensor<string, []>("op_602"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_33_cast = reshape(shape = var_602, x = attn_output_31_cast)[name = tensor<string, []>("attn_output_33_cast")]; | |
| tensor<int32, [4]> attn_output_35_perm_0 = const()[name = tensor<string, []>("attn_output_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_605 = const()[name = tensor<string, []>("op_605"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_30 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast)[name = tensor<string, []>("transpose_30")]; | |
| tensor<fp16, [1, 77, 768]> input_89_cast = reshape(shape = var_605, x = transpose_30)[name = tensor<string, []>("input_89_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103910592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104353024))), 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, []>(104353216)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_33_cast = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_89_cast)[name = tensor<string, []>("hidden_states_33_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_91_cast = add(x = input_83_cast, y = hidden_states_33_cast)[name = tensor<string, []>("input_91_cast")]; | |
| tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104354816)))]; | |
| 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, []>(104356416)))]; | |
| tensor<fp16, [1, 77, 768]> input_93_cast = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast)[name = tensor<string, []>("input_93_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104358016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106127552))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106127744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106130112))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_95_cast = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast)[name = tensor<string, []>("input_95_cast")]; | |
| tensor<fp16, []> var_620_to_fp16 = const()[name = tensor<string, []>("op_620_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_621_cast = mul(x = input_95_cast, y = var_620_to_fp16)[name = tensor<string, []>("op_621_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_622_cast = sigmoid(x = var_621_cast)[name = tensor<string, []>("op_622_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_97_cast = mul(x = input_95_cast, y = var_622_cast)[name = tensor<string, []>("input_97_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106130304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107899840))), 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, []>(107900032)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_35_cast = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast)[name = tensor<string, []>("hidden_states_35_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_99_cast = add(x = input_91_cast, y = hidden_states_35_cast)[name = tensor<string, []>("input_99_cast")]; | |
| 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, []>(107901632)))]; | |
| 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, []>(107903232)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_37_cast = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast)[name = tensor<string, []>("hidden_states_37_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107904832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108347264))), 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, []>(108347456)))]; | |
| tensor<fp16, [1, 77, 768]> var_646_cast = 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)[name = tensor<string, []>("op_646_cast")]; | |
| tensor<fp16, []> var_647_to_fp16 = const()[name = tensor<string, []>("op_647_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_41_cast = mul(x = var_646_cast, y = var_647_to_fp16)[name = tensor<string, []>("tensor_41_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108349056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108791488))), 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, []>(108791680)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_37_cast = 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)[name = tensor<string, []>("tensor_37_cast")]; | |
| tensor<int32, [4]> var_652 = const()[name = tensor<string, []>("op_652"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_653_cast = reshape(shape = var_652, x = tensor_37_cast)[name = tensor<string, []>("op_653_cast")]; | |
| tensor<int32, [4]> var_654_perm_0 = const()[name = tensor<string, []>("op_654_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108793280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109235712))), 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, []>(109235904)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_39_cast = 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)[name = tensor<string, []>("tensor_39_cast")]; | |
| tensor<int32, [4]> var_659 = const()[name = tensor<string, []>("op_659"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_660_cast = reshape(shape = var_659, x = tensor_39_cast)[name = tensor<string, []>("op_660_cast")]; | |
| tensor<int32, [4]> var_661_perm_0 = const()[name = tensor<string, []>("op_661_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_668 = const()[name = tensor<string, []>("op_668"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_669_cast = reshape(shape = var_668, x = tensor_41_cast)[name = tensor<string, []>("op_669_cast")]; | |
| tensor<int32, [4]> var_670_perm_0 = const()[name = tensor<string, []>("op_670_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_672 = const()[name = tensor<string, []>("op_672"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_29 = transpose(perm = var_670_perm_0, x = var_669_cast)[name = tensor<string, []>("transpose_29")]; | |
| tensor<fp16, [12, 77, 64]> query_states_13_cast = reshape(shape = var_672, x = transpose_29)[name = tensor<string, []>("query_states_13_cast")]; | |
| tensor<int32, [3]> var_674 = const()[name = tensor<string, []>("op_674"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_28 = transpose(perm = var_654_perm_0, x = var_653_cast)[name = tensor<string, []>("transpose_28")]; | |
| tensor<fp16, [12, 77, 64]> key_states_27_cast = reshape(shape = var_674, x = transpose_28)[name = tensor<string, []>("key_states_27_cast")]; | |
| tensor<int32, [3]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_27 = transpose(perm = var_661_perm_0, x = var_660_cast)[name = tensor<string, []>("transpose_27")]; | |
| tensor<fp16, [12, 77, 64]> value_states_27_cast = reshape(shape = var_676, x = transpose_27)[name = tensor<string, []>("value_states_27_cast")]; | |
| tensor<int32, [3]> var_679_perm_0 = const()[name = tensor<string, []>("op_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_37_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_37_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_26 = transpose(perm = var_679_perm_0, x = key_states_27_cast)[name = tensor<string, []>("transpose_26")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_37_cast = matmul(transpose_x = attn_weights_37_transpose_x_0, transpose_y = attn_weights_37_transpose_y_0, x = query_states_13_cast, y = transpose_26)[name = tensor<string, []>("attn_weights_37_cast")]; | |
| tensor<int32, [4]> var_681 = const()[name = tensor<string, []>("op_681"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_682_cast = reshape(shape = var_681, x = attn_weights_37_cast)[name = tensor<string, []>("op_682_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_39_cast = add(x = var_682_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_39_cast")]; | |
| tensor<int32, [3]> var_687 = const()[name = tensor<string, []>("op_687"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_101_cast = reshape(shape = var_687, x = attn_weights_39_cast)[name = tensor<string, []>("input_101_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_103_cast = softmax(axis = var_5, x = input_101_cast)[name = tensor<string, []>("input_103_cast")]; | |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_37_cast = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast, y = value_states_27_cast)[name = tensor<string, []>("attn_output_37_cast")]; | |
| tensor<int32, [4]> var_692 = const()[name = tensor<string, []>("op_692"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_39_cast = reshape(shape = var_692, x = attn_output_37_cast)[name = tensor<string, []>("attn_output_39_cast")]; | |
| tensor<int32, [4]> attn_output_41_perm_0 = const()[name = tensor<string, []>("attn_output_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_695 = const()[name = tensor<string, []>("op_695"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_25 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast)[name = tensor<string, []>("transpose_25")]; | |
| tensor<fp16, [1, 77, 768]> input_105_cast = reshape(shape = var_695, x = transpose_25)[name = tensor<string, []>("input_105_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109237504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109679936))), 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, []>(109680128)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_39_cast = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_105_cast)[name = tensor<string, []>("hidden_states_39_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_107_cast = add(x = input_99_cast, y = hidden_states_39_cast)[name = tensor<string, []>("input_107_cast")]; | |
| tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109681728)))]; | |
| 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, []>(109683328)))]; | |
| tensor<fp16, [1, 77, 768]> input_109_cast = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast)[name = tensor<string, []>("input_109_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109684928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111454464))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111454656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111457024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_111_cast = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_109_cast)[name = tensor<string, []>("input_111_cast")]; | |
| tensor<fp16, []> var_710_to_fp16 = const()[name = tensor<string, []>("op_710_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_711_cast = mul(x = input_111_cast, y = var_710_to_fp16)[name = tensor<string, []>("op_711_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_712_cast = sigmoid(x = var_711_cast)[name = tensor<string, []>("op_712_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_113_cast = mul(x = input_111_cast, y = var_712_cast)[name = tensor<string, []>("input_113_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111457216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113226752))), 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, []>(113226944)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_41_cast = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_113_cast)[name = tensor<string, []>("hidden_states_41_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_115_cast = add(x = input_107_cast, y = hidden_states_41_cast)[name = tensor<string, []>("input_115_cast")]; | |
| 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, []>(113228544)))]; | |
| 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, []>(113230144)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_43_cast = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast)[name = tensor<string, []>("hidden_states_43_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113231744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113674176))), 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, []>(113674368)))]; | |
| tensor<fp16, [1, 77, 768]> var_736_cast = 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)[name = tensor<string, []>("op_736_cast")]; | |
| tensor<fp16, []> var_737_to_fp16 = const()[name = tensor<string, []>("op_737_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_47_cast = mul(x = var_736_cast, y = var_737_to_fp16)[name = tensor<string, []>("tensor_47_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113675968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114118400))), 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, []>(114118592)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_43_cast = 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)[name = tensor<string, []>("tensor_43_cast")]; | |
| tensor<int32, [4]> var_742 = const()[name = tensor<string, []>("op_742"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_743_cast = reshape(shape = var_742, x = tensor_43_cast)[name = tensor<string, []>("op_743_cast")]; | |
| tensor<int32, [4]> var_744_perm_0 = const()[name = tensor<string, []>("op_744_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114120192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114562624))), 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, []>(114562816)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_45_cast = 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)[name = tensor<string, []>("tensor_45_cast")]; | |
| tensor<int32, [4]> var_749 = const()[name = tensor<string, []>("op_749"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_750_cast = reshape(shape = var_749, x = tensor_45_cast)[name = tensor<string, []>("op_750_cast")]; | |
| tensor<int32, [4]> var_751_perm_0 = const()[name = tensor<string, []>("op_751_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_758 = const()[name = tensor<string, []>("op_758"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_759_cast = reshape(shape = var_758, x = tensor_47_cast)[name = tensor<string, []>("op_759_cast")]; | |
| tensor<int32, [4]> var_760_perm_0 = const()[name = tensor<string, []>("op_760_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_762 = const()[name = tensor<string, []>("op_762"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_24 = transpose(perm = var_760_perm_0, x = var_759_cast)[name = tensor<string, []>("transpose_24")]; | |
| tensor<fp16, [12, 77, 64]> query_states_15_cast = reshape(shape = var_762, x = transpose_24)[name = tensor<string, []>("query_states_15_cast")]; | |
| tensor<int32, [3]> var_764 = const()[name = tensor<string, []>("op_764"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_23 = transpose(perm = var_744_perm_0, x = var_743_cast)[name = tensor<string, []>("transpose_23")]; | |
| tensor<fp16, [12, 77, 64]> key_states_31_cast = reshape(shape = var_764, x = transpose_23)[name = tensor<string, []>("key_states_31_cast")]; | |
| tensor<int32, [3]> var_766 = const()[name = tensor<string, []>("op_766"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_22 = transpose(perm = var_751_perm_0, x = var_750_cast)[name = tensor<string, []>("transpose_22")]; | |
| tensor<fp16, [12, 77, 64]> value_states_31_cast = reshape(shape = var_766, x = transpose_22)[name = tensor<string, []>("value_states_31_cast")]; | |
| tensor<int32, [3]> var_769_perm_0 = const()[name = tensor<string, []>("op_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_43_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_43_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_21 = transpose(perm = var_769_perm_0, x = key_states_31_cast)[name = tensor<string, []>("transpose_21")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_43_cast = matmul(transpose_x = attn_weights_43_transpose_x_0, transpose_y = attn_weights_43_transpose_y_0, x = query_states_15_cast, y = transpose_21)[name = tensor<string, []>("attn_weights_43_cast")]; | |
| tensor<int32, [4]> var_771 = const()[name = tensor<string, []>("op_771"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_772_cast = reshape(shape = var_771, x = attn_weights_43_cast)[name = tensor<string, []>("op_772_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_45_cast = add(x = var_772_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_45_cast")]; | |
| tensor<int32, [3]> var_777 = const()[name = tensor<string, []>("op_777"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_117_cast = reshape(shape = var_777, x = attn_weights_45_cast)[name = tensor<string, []>("input_117_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_119_cast = softmax(axis = var_5, x = input_117_cast)[name = tensor<string, []>("input_119_cast")]; | |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_43_cast = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast, y = value_states_31_cast)[name = tensor<string, []>("attn_output_43_cast")]; | |
| tensor<int32, [4]> var_782 = const()[name = tensor<string, []>("op_782"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_45_cast = reshape(shape = var_782, x = attn_output_43_cast)[name = tensor<string, []>("attn_output_45_cast")]; | |
| tensor<int32, [4]> attn_output_47_perm_0 = const()[name = tensor<string, []>("attn_output_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_785 = const()[name = tensor<string, []>("op_785"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_20 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast)[name = tensor<string, []>("transpose_20")]; | |
| tensor<fp16, [1, 77, 768]> input_121_cast = reshape(shape = var_785, x = transpose_20)[name = tensor<string, []>("input_121_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114564416))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115006848))), 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, []>(115007040)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_45_cast = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast)[name = tensor<string, []>("hidden_states_45_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_123_cast = add(x = input_115_cast, y = hidden_states_45_cast)[name = tensor<string, []>("input_123_cast")]; | |
| tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115008640)))]; | |
| 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, []>(115010240)))]; | |
| tensor<fp16, [1, 77, 768]> input_125_cast = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast)[name = tensor<string, []>("input_125_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115011840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116781376))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116781568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116783936))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_127_cast = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_125_cast)[name = tensor<string, []>("input_127_cast")]; | |
| tensor<fp16, []> var_800_to_fp16 = const()[name = tensor<string, []>("op_800_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_801_cast = mul(x = input_127_cast, y = var_800_to_fp16)[name = tensor<string, []>("op_801_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_802_cast = sigmoid(x = var_801_cast)[name = tensor<string, []>("op_802_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_129_cast = mul(x = input_127_cast, y = var_802_cast)[name = tensor<string, []>("input_129_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(116784128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118553664))), 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, []>(118553856)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_47_cast = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_129_cast)[name = tensor<string, []>("hidden_states_47_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_131_cast = add(x = input_123_cast, y = hidden_states_47_cast)[name = tensor<string, []>("input_131_cast")]; | |
| 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, []>(118555456)))]; | |
| 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, []>(118557056)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_49_cast = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast)[name = tensor<string, []>("hidden_states_49_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118558656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119001088))), 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, []>(119001280)))]; | |
| tensor<fp16, [1, 77, 768]> var_826_cast = 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)[name = tensor<string, []>("op_826_cast")]; | |
| tensor<fp16, []> var_827_to_fp16 = const()[name = tensor<string, []>("op_827_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_53_cast = mul(x = var_826_cast, y = var_827_to_fp16)[name = tensor<string, []>("tensor_53_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119002880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119445312))), 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, []>(119445504)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_49_cast = 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)[name = tensor<string, []>("tensor_49_cast")]; | |
| tensor<int32, [4]> var_832 = const()[name = tensor<string, []>("op_832"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_833_cast = reshape(shape = var_832, x = tensor_49_cast)[name = tensor<string, []>("op_833_cast")]; | |
| tensor<int32, [4]> var_834_perm_0 = const()[name = tensor<string, []>("op_834_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119447104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119889536))), 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, []>(119889728)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_51_cast = 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)[name = tensor<string, []>("tensor_51_cast")]; | |
| tensor<int32, [4]> var_839 = const()[name = tensor<string, []>("op_839"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_840_cast = reshape(shape = var_839, x = tensor_51_cast)[name = tensor<string, []>("op_840_cast")]; | |
| tensor<int32, [4]> var_841_perm_0 = const()[name = tensor<string, []>("op_841_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_848 = const()[name = tensor<string, []>("op_848"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_849_cast = reshape(shape = var_848, x = tensor_53_cast)[name = tensor<string, []>("op_849_cast")]; | |
| tensor<int32, [4]> var_850_perm_0 = const()[name = tensor<string, []>("op_850_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_852 = const()[name = tensor<string, []>("op_852"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_19 = transpose(perm = var_850_perm_0, x = var_849_cast)[name = tensor<string, []>("transpose_19")]; | |
| tensor<fp16, [12, 77, 64]> query_states_17_cast = reshape(shape = var_852, x = transpose_19)[name = tensor<string, []>("query_states_17_cast")]; | |
| tensor<int32, [3]> var_854 = const()[name = tensor<string, []>("op_854"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_18 = transpose(perm = var_834_perm_0, x = var_833_cast)[name = tensor<string, []>("transpose_18")]; | |
| tensor<fp16, [12, 77, 64]> key_states_35_cast = reshape(shape = var_854, x = transpose_18)[name = tensor<string, []>("key_states_35_cast")]; | |
| tensor<int32, [3]> var_856 = const()[name = tensor<string, []>("op_856"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_17 = transpose(perm = var_841_perm_0, x = var_840_cast)[name = tensor<string, []>("transpose_17")]; | |
| tensor<fp16, [12, 77, 64]> value_states_35_cast = reshape(shape = var_856, x = transpose_17)[name = tensor<string, []>("value_states_35_cast")]; | |
| tensor<int32, [3]> var_859_perm_0 = const()[name = tensor<string, []>("op_859_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_49_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_49_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_16 = transpose(perm = var_859_perm_0, x = key_states_35_cast)[name = tensor<string, []>("transpose_16")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_49_cast = matmul(transpose_x = attn_weights_49_transpose_x_0, transpose_y = attn_weights_49_transpose_y_0, x = query_states_17_cast, y = transpose_16)[name = tensor<string, []>("attn_weights_49_cast")]; | |
| tensor<int32, [4]> var_861 = const()[name = tensor<string, []>("op_861"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_862_cast = reshape(shape = var_861, x = attn_weights_49_cast)[name = tensor<string, []>("op_862_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_51_cast = add(x = var_862_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_51_cast")]; | |
| tensor<int32, [3]> var_867 = const()[name = tensor<string, []>("op_867"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_133_cast = reshape(shape = var_867, x = attn_weights_51_cast)[name = tensor<string, []>("input_133_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_135_cast = softmax(axis = var_5, x = input_133_cast)[name = tensor<string, []>("input_135_cast")]; | |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_49_cast = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast, y = value_states_35_cast)[name = tensor<string, []>("attn_output_49_cast")]; | |
| tensor<int32, [4]> var_872 = const()[name = tensor<string, []>("op_872"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_51_cast = reshape(shape = var_872, x = attn_output_49_cast)[name = tensor<string, []>("attn_output_51_cast")]; | |
| tensor<int32, [4]> attn_output_53_perm_0 = const()[name = tensor<string, []>("attn_output_53_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_875 = const()[name = tensor<string, []>("op_875"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_15 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast)[name = tensor<string, []>("transpose_15")]; | |
| tensor<fp16, [1, 77, 768]> input_137_cast = reshape(shape = var_875, x = transpose_15)[name = tensor<string, []>("input_137_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119891328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120333760))), 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, []>(120333952)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_51_cast = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_137_cast)[name = tensor<string, []>("hidden_states_51_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_139_cast = add(x = input_131_cast, y = hidden_states_51_cast)[name = tensor<string, []>("input_139_cast")]; | |
| tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120335552)))]; | |
| 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, []>(120337152)))]; | |
| tensor<fp16, [1, 77, 768]> input_141_cast = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast)[name = tensor<string, []>("input_141_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120338752))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122108288))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122108480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122110848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_143_cast = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast)[name = tensor<string, []>("input_143_cast")]; | |
| tensor<fp16, []> var_890_to_fp16 = const()[name = tensor<string, []>("op_890_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_891_cast = mul(x = input_143_cast, y = var_890_to_fp16)[name = tensor<string, []>("op_891_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_892_cast = sigmoid(x = var_891_cast)[name = tensor<string, []>("op_892_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_145_cast = mul(x = input_143_cast, y = var_892_cast)[name = tensor<string, []>("input_145_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122111040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123880576))), 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, []>(123880768)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_53_cast = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast)[name = tensor<string, []>("hidden_states_53_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_147_cast = add(x = input_139_cast, y = hidden_states_53_cast)[name = tensor<string, []>("input_147_cast")]; | |
| 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, []>(123882368)))]; | |
| 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, []>(123883968)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_55_cast = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast)[name = tensor<string, []>("hidden_states_55_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123885568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124328000))), 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, []>(124328192)))]; | |
| tensor<fp16, [1, 77, 768]> var_916_cast = 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)[name = tensor<string, []>("op_916_cast")]; | |
| tensor<fp16, []> var_917_to_fp16 = const()[name = tensor<string, []>("op_917_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_59_cast = mul(x = var_916_cast, y = var_917_to_fp16)[name = tensor<string, []>("tensor_59_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124329792))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124772224))), 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, []>(124772416)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_55_cast = 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)[name = tensor<string, []>("tensor_55_cast")]; | |
| tensor<int32, [4]> var_922 = const()[name = tensor<string, []>("op_922"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_923_cast = reshape(shape = var_922, x = tensor_55_cast)[name = tensor<string, []>("op_923_cast")]; | |
| tensor<int32, [4]> var_924_perm_0 = const()[name = tensor<string, []>("op_924_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124774016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125216448))), 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, []>(125216640)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_57_cast = 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)[name = tensor<string, []>("tensor_57_cast")]; | |
| tensor<int32, [4]> var_929 = const()[name = tensor<string, []>("op_929"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_930_cast = reshape(shape = var_929, x = tensor_57_cast)[name = tensor<string, []>("op_930_cast")]; | |
| tensor<int32, [4]> var_931_perm_0 = const()[name = tensor<string, []>("op_931_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_938 = const()[name = tensor<string, []>("op_938"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_939_cast = reshape(shape = var_938, x = tensor_59_cast)[name = tensor<string, []>("op_939_cast")]; | |
| tensor<int32, [4]> var_940_perm_0 = const()[name = tensor<string, []>("op_940_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_942 = const()[name = tensor<string, []>("op_942"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_14 = transpose(perm = var_940_perm_0, x = var_939_cast)[name = tensor<string, []>("transpose_14")]; | |
| tensor<fp16, [12, 77, 64]> query_states_19_cast = reshape(shape = var_942, x = transpose_14)[name = tensor<string, []>("query_states_19_cast")]; | |
| tensor<int32, [3]> var_944 = const()[name = tensor<string, []>("op_944"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_13 = transpose(perm = var_924_perm_0, x = var_923_cast)[name = tensor<string, []>("transpose_13")]; | |
| tensor<fp16, [12, 77, 64]> key_states_39_cast = reshape(shape = var_944, x = transpose_13)[name = tensor<string, []>("key_states_39_cast")]; | |
| tensor<int32, [3]> var_946 = const()[name = tensor<string, []>("op_946"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_12 = transpose(perm = var_931_perm_0, x = var_930_cast)[name = tensor<string, []>("transpose_12")]; | |
| tensor<fp16, [12, 77, 64]> value_states_39_cast = reshape(shape = var_946, x = transpose_12)[name = tensor<string, []>("value_states_39_cast")]; | |
| tensor<int32, [3]> var_949_perm_0 = const()[name = tensor<string, []>("op_949_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_55_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_55_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_11 = transpose(perm = var_949_perm_0, x = key_states_39_cast)[name = tensor<string, []>("transpose_11")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_55_cast = matmul(transpose_x = attn_weights_55_transpose_x_0, transpose_y = attn_weights_55_transpose_y_0, x = query_states_19_cast, y = transpose_11)[name = tensor<string, []>("attn_weights_55_cast")]; | |
| tensor<int32, [4]> var_951 = const()[name = tensor<string, []>("op_951"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_952_cast = reshape(shape = var_951, x = attn_weights_55_cast)[name = tensor<string, []>("op_952_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_57_cast = add(x = var_952_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_57_cast")]; | |
| tensor<int32, [3]> var_957 = const()[name = tensor<string, []>("op_957"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_149_cast = reshape(shape = var_957, x = attn_weights_57_cast)[name = tensor<string, []>("input_149_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_151_cast = softmax(axis = var_5, x = input_149_cast)[name = tensor<string, []>("input_151_cast")]; | |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_55_cast = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast, y = value_states_39_cast)[name = tensor<string, []>("attn_output_55_cast")]; | |
| tensor<int32, [4]> var_962 = const()[name = tensor<string, []>("op_962"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_57_cast = reshape(shape = var_962, x = attn_output_55_cast)[name = tensor<string, []>("attn_output_57_cast")]; | |
| tensor<int32, [4]> attn_output_59_perm_0 = const()[name = tensor<string, []>("attn_output_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_965 = const()[name = tensor<string, []>("op_965"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_10 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast)[name = tensor<string, []>("transpose_10")]; | |
| tensor<fp16, [1, 77, 768]> input_153_cast = reshape(shape = var_965, x = transpose_10)[name = tensor<string, []>("input_153_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125218240))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125660672))), 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, []>(125660864)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_57_cast = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_153_cast)[name = tensor<string, []>("hidden_states_57_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_155_cast = add(x = input_147_cast, y = hidden_states_57_cast)[name = tensor<string, []>("input_155_cast")]; | |
| tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125662464)))]; | |
| 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, []>(125664064)))]; | |
| tensor<fp16, [1, 77, 768]> input_157_cast = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast)[name = tensor<string, []>("input_157_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125665664))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127435200))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127435392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127437760))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_159_cast = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_157_cast)[name = tensor<string, []>("input_159_cast")]; | |
| tensor<fp16, []> var_980_to_fp16 = const()[name = tensor<string, []>("op_980_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_981_cast = mul(x = input_159_cast, y = var_980_to_fp16)[name = tensor<string, []>("op_981_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_982_cast = sigmoid(x = var_981_cast)[name = tensor<string, []>("op_982_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_161_cast = mul(x = input_159_cast, y = var_982_cast)[name = tensor<string, []>("input_161_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127437952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129207488))), 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, []>(129207680)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_59_cast = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_161_cast)[name = tensor<string, []>("hidden_states_59_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_163_cast = add(x = input_155_cast, y = hidden_states_59_cast)[name = tensor<string, []>("input_163_cast")]; | |
| 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, []>(129209280)))]; | |
| 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, []>(129210880)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_61_cast = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast)[name = tensor<string, []>("hidden_states_61_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129212480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129654912))), 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, []>(129655104)))]; | |
| tensor<fp16, [1, 77, 768]> var_1006_cast = 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)[name = tensor<string, []>("op_1006_cast")]; | |
| tensor<fp16, []> var_1007_to_fp16 = const()[name = tensor<string, []>("op_1007_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_65_cast = mul(x = var_1006_cast, y = var_1007_to_fp16)[name = tensor<string, []>("tensor_65_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129656704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130099136))), 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, []>(130099328)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_61_cast = 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)[name = tensor<string, []>("tensor_61_cast")]; | |
| tensor<int32, [4]> var_1012 = const()[name = tensor<string, []>("op_1012"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1013_cast = reshape(shape = var_1012, x = tensor_61_cast)[name = tensor<string, []>("op_1013_cast")]; | |
| tensor<int32, [4]> var_1014_perm_0 = const()[name = tensor<string, []>("op_1014_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130100928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130543360))), 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, []>(130543552)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_63_cast = 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)[name = tensor<string, []>("tensor_63_cast")]; | |
| tensor<int32, [4]> var_1019 = const()[name = tensor<string, []>("op_1019"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1020_cast = reshape(shape = var_1019, x = tensor_63_cast)[name = tensor<string, []>("op_1020_cast")]; | |
| tensor<int32, [4]> var_1021_perm_0 = const()[name = tensor<string, []>("op_1021_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1028 = const()[name = tensor<string, []>("op_1028"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1029_cast = reshape(shape = var_1028, x = tensor_65_cast)[name = tensor<string, []>("op_1029_cast")]; | |
| tensor<int32, [4]> var_1030_perm_0 = const()[name = tensor<string, []>("op_1030_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1032 = const()[name = tensor<string, []>("op_1032"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_9 = transpose(perm = var_1030_perm_0, x = var_1029_cast)[name = tensor<string, []>("transpose_9")]; | |
| tensor<fp16, [12, 77, 64]> query_states_21_cast = reshape(shape = var_1032, x = transpose_9)[name = tensor<string, []>("query_states_21_cast")]; | |
| tensor<int32, [3]> var_1034 = const()[name = tensor<string, []>("op_1034"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_8 = transpose(perm = var_1014_perm_0, x = var_1013_cast)[name = tensor<string, []>("transpose_8")]; | |
| tensor<fp16, [12, 77, 64]> key_states_43_cast = reshape(shape = var_1034, x = transpose_8)[name = tensor<string, []>("key_states_43_cast")]; | |
| tensor<int32, [3]> var_1036 = const()[name = tensor<string, []>("op_1036"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_7 = transpose(perm = var_1021_perm_0, x = var_1020_cast)[name = tensor<string, []>("transpose_7")]; | |
| tensor<fp16, [12, 77, 64]> value_states_43_cast = reshape(shape = var_1036, x = transpose_7)[name = tensor<string, []>("value_states_43_cast")]; | |
| tensor<int32, [3]> var_1039_perm_0 = const()[name = tensor<string, []>("op_1039_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_61_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_61_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_6 = transpose(perm = var_1039_perm_0, x = key_states_43_cast)[name = tensor<string, []>("transpose_6")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_61_cast = matmul(transpose_x = attn_weights_61_transpose_x_0, transpose_y = attn_weights_61_transpose_y_0, x = query_states_21_cast, y = transpose_6)[name = tensor<string, []>("attn_weights_61_cast")]; | |
| tensor<int32, [4]> var_1041 = const()[name = tensor<string, []>("op_1041"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1042_cast = reshape(shape = var_1041, x = attn_weights_61_cast)[name = tensor<string, []>("op_1042_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_63_cast = add(x = var_1042_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_63_cast")]; | |
| tensor<int32, [3]> var_1047 = const()[name = tensor<string, []>("op_1047"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_165_cast = reshape(shape = var_1047, x = attn_weights_63_cast)[name = tensor<string, []>("input_165_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_167_cast = softmax(axis = var_5, x = input_165_cast)[name = tensor<string, []>("input_167_cast")]; | |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_61_cast = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast, y = value_states_43_cast)[name = tensor<string, []>("attn_output_61_cast")]; | |
| tensor<int32, [4]> var_1052 = const()[name = tensor<string, []>("op_1052"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_63_cast = reshape(shape = var_1052, x = attn_output_61_cast)[name = tensor<string, []>("attn_output_63_cast")]; | |
| tensor<int32, [4]> attn_output_65_perm_0 = const()[name = tensor<string, []>("attn_output_65_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1055 = const()[name = tensor<string, []>("op_1055"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_5 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast)[name = tensor<string, []>("transpose_5")]; | |
| tensor<fp16, [1, 77, 768]> input_169_cast = reshape(shape = var_1055, x = transpose_5)[name = tensor<string, []>("input_169_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130545152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130987584))), 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, []>(130987776)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_63_cast = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast)[name = tensor<string, []>("hidden_states_63_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_171_cast = add(x = input_163_cast, y = hidden_states_63_cast)[name = tensor<string, []>("input_171_cast")]; | |
| tensor<int32, [1]> input_173_axes_0 = const()[name = tensor<string, []>("input_173_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130989376)))]; | |
| 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, []>(130990976)))]; | |
| tensor<fp16, [1, 77, 768]> input_173_cast = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast)[name = tensor<string, []>("input_173_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130992576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132762112))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132762304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132764672))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_175_cast = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_173_cast)[name = tensor<string, []>("input_175_cast")]; | |
| tensor<fp16, []> var_1070_to_fp16 = const()[name = tensor<string, []>("op_1070_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1071_cast = mul(x = input_175_cast, y = var_1070_to_fp16)[name = tensor<string, []>("op_1071_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_1072_cast = sigmoid(x = var_1071_cast)[name = tensor<string, []>("op_1072_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_177_cast = mul(x = input_175_cast, y = var_1072_cast)[name = tensor<string, []>("input_177_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132764864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134534400))), 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, []>(134534592)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_65_cast = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_177_cast)[name = tensor<string, []>("hidden_states_65_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_179_cast = add(x = input_171_cast, y = hidden_states_65_cast)[name = tensor<string, []>("input_179_cast")]; | |
| 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, []>(134536192)))]; | |
| 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, []>(134537792)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_67_cast = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast)[name = tensor<string, []>("hidden_states_67_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134539392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134981824))), 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, []>(134982016)))]; | |
| tensor<fp16, [1, 77, 768]> var_1096_cast = 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)[name = tensor<string, []>("op_1096_cast")]; | |
| tensor<fp16, []> var_1097_to_fp16 = const()[name = tensor<string, []>("op_1097_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; | |
| tensor<fp16, [1, 77, 768]> tensor_cast = mul(x = var_1096_cast, y = var_1097_to_fp16)[name = tensor<string, []>("tensor_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134983616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135426048))), 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, []>(135426240)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_67_cast = 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)[name = tensor<string, []>("tensor_67_cast")]; | |
| tensor<int32, [4]> var_1102 = const()[name = tensor<string, []>("op_1102"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1103_cast = reshape(shape = var_1102, x = tensor_67_cast)[name = tensor<string, []>("op_1103_cast")]; | |
| tensor<int32, [4]> var_1104_perm_0 = const()[name = tensor<string, []>("op_1104_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135427840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135870272))), 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, []>(135870464)))]; | |
| tensor<fp16, [1, 77, 768]> tensor_69_cast = 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)[name = tensor<string, []>("tensor_69_cast")]; | |
| tensor<int32, [4]> var_1109 = const()[name = tensor<string, []>("op_1109"), val = tensor<int32, [4]>([1, -1, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1110_cast = reshape(shape = var_1109, x = tensor_69_cast)[name = tensor<string, []>("op_1110_cast")]; | |
| tensor<int32, [4]> var_1111_perm_0 = const()[name = tensor<string, []>("op_1111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> var_1118 = const()[name = tensor<string, []>("op_1118"), val = tensor<int32, [4]>([1, 77, 12, 64])]; | |
| tensor<fp16, [1, 77, 12, 64]> var_1119_cast = reshape(shape = var_1118, x = tensor_cast)[name = tensor<string, []>("op_1119_cast")]; | |
| tensor<int32, [4]> var_1120_perm_0 = const()[name = tensor<string, []>("op_1120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1122 = const()[name = tensor<string, []>("op_1122"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_4 = transpose(perm = var_1120_perm_0, x = var_1119_cast)[name = tensor<string, []>("transpose_4")]; | |
| tensor<fp16, [12, 77, 64]> query_states_cast = reshape(shape = var_1122, x = transpose_4)[name = tensor<string, []>("query_states_cast")]; | |
| tensor<int32, [3]> var_1124 = const()[name = tensor<string, []>("op_1124"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_3 = transpose(perm = var_1104_perm_0, x = var_1103_cast)[name = tensor<string, []>("transpose_3")]; | |
| tensor<fp16, [12, 77, 64]> key_states_cast = reshape(shape = var_1124, x = transpose_3)[name = tensor<string, []>("key_states_cast")]; | |
| tensor<int32, [3]> var_1126 = const()[name = tensor<string, []>("op_1126"), val = tensor<int32, [3]>([12, -1, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> transpose_2 = transpose(perm = var_1111_perm_0, x = var_1110_cast)[name = tensor<string, []>("transpose_2")]; | |
| tensor<fp16, [12, 77, 64]> value_states_cast = reshape(shape = var_1126, x = transpose_2)[name = tensor<string, []>("value_states_cast")]; | |
| tensor<int32, [3]> var_1129_perm_0 = const()[name = tensor<string, []>("op_1129_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<bool, []> attn_weights_67_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_weights_67_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 64, 77]> transpose_1 = transpose(perm = var_1129_perm_0, x = key_states_cast)[name = tensor<string, []>("transpose_1")]; | |
| tensor<fp16, [12, 77, 77]> attn_weights_67_cast = matmul(transpose_x = attn_weights_67_transpose_x_0, transpose_y = attn_weights_67_transpose_y_0, x = query_states_cast, y = transpose_1)[name = tensor<string, []>("attn_weights_67_cast")]; | |
| tensor<int32, [4]> var_1131 = const()[name = tensor<string, []>("op_1131"), val = tensor<int32, [4]>([1, 12, 77, 77])]; | |
| tensor<fp16, [1, 12, 77, 77]> var_1132_cast = reshape(shape = var_1131, x = attn_weights_67_cast)[name = tensor<string, []>("op_1132_cast")]; | |
| tensor<fp16, [1, 12, 77, 77]> attn_weights_69_cast = add(x = var_1132_cast, y = op_56_to_fp16_palettized)[name = tensor<string, []>("attn_weights_69_cast")]; | |
| tensor<int32, [3]> var_1137 = const()[name = tensor<string, []>("op_1137"), val = tensor<int32, [3]>([12, 77, 77])]; | |
| tensor<fp16, [12, 77, 77]> input_181_cast = reshape(shape = var_1137, x = attn_weights_69_cast)[name = tensor<string, []>("input_181_cast")]; | |
| tensor<fp16, [12, 77, 77]> input_183_cast = softmax(axis = var_5, x = input_181_cast)[name = tensor<string, []>("input_183_cast")]; | |
| tensor<bool, []> attn_output_67_transpose_x_0 = const()[name = tensor<string, []>("attn_output_67_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> attn_output_67_transpose_y_0 = const()[name = tensor<string, []>("attn_output_67_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [12, 77, 64]> attn_output_67_cast = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast, y = value_states_cast)[name = tensor<string, []>("attn_output_67_cast")]; | |
| tensor<int32, [4]> var_1142 = const()[name = tensor<string, []>("op_1142"), val = tensor<int32, [4]>([1, 12, 77, 64])]; | |
| tensor<fp16, [1, 12, 77, 64]> attn_output_69_cast = reshape(shape = var_1142, x = attn_output_67_cast)[name = tensor<string, []>("attn_output_69_cast")]; | |
| tensor<int32, [4]> attn_output_perm_0 = const()[name = tensor<string, []>("attn_output_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> var_1145 = const()[name = tensor<string, []>("op_1145"), val = tensor<int32, [3]>([1, 77, 768])]; | |
| tensor<fp16, [1, 77, 12, 64]> transpose_0 = transpose(perm = attn_output_perm_0, x = attn_output_69_cast)[name = tensor<string, []>("transpose_0")]; | |
| tensor<fp16, [1, 77, 768]> input_185_cast = reshape(shape = var_1145, x = transpose_0)[name = tensor<string, []>("input_185_cast")]; | |
| tensor<fp16, [768, 768]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [442368]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135872064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136314496))), 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, []>(136314688)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_69_cast = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_185_cast)[name = tensor<string, []>("hidden_states_69_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_187_cast = add(x = input_179_cast, y = hidden_states_69_cast)[name = tensor<string, []>("input_187_cast")]; | |
| tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136316288)))]; | |
| 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, []>(136317888)))]; | |
| tensor<fp16, [1, 77, 768]> input_189_cast = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast)[name = tensor<string, []>("input_189_cast")]; | |
| tensor<fp16, [3072, 768]> text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136319488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138089024))), 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_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138089216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138091584))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([3072])]; | |
| tensor<fp16, [1, 77, 3072]> input_191_cast = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast)[name = tensor<string, []>("input_191_cast")]; | |
| tensor<fp16, []> var_1160_to_fp16 = const()[name = tensor<string, []>("op_1160_to_fp16"), val = tensor<fp16, []>(0x1.b3cp+0)]; | |
| tensor<fp16, [1, 77, 3072]> var_1161_cast = mul(x = input_191_cast, y = var_1160_to_fp16)[name = tensor<string, []>("op_1161_cast")]; | |
| tensor<fp16, [1, 77, 3072]> var_1162_cast = sigmoid(x = var_1161_cast)[name = tensor<string, []>("op_1162_cast")]; | |
| tensor<fp16, [1, 77, 3072]> input_193_cast = mul(x = input_191_cast, y = var_1162_cast)[name = tensor<string, []>("input_193_cast")]; | |
| tensor<fp16, [768, 3072]> text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1769472]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138091776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139861312))), 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, []>(139861504)))]; | |
| tensor<fp16, [1, 77, 768]> hidden_states_cast = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast)[name = tensor<string, []>("hidden_states_cast")]; | |
| tensor<fp16, [1, 77, 768]> input_cast = add(x = input_187_cast, y = hidden_states_cast)[name = tensor<string, []>("input_cast")]; | |
| 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, []>(139863104)))]; | |
| 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, []>(139864704)))]; | |
| tensor<fp16, [1, 77, 768]> last_hidden_state_cast = 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)[name = tensor<string, []>("last_hidden_state_cast")]; | |
| tensor<string, []> last_hidden_state_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("last_hidden_state_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<int32, [1]> var_1173 = const()[name = tensor<string, []>("op_1173"), val = tensor<int32, [1]>([0])]; | |
| tensor<int32, [1]> var_1175 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_2)[name = tensor<string, []>("op_1175")]; | |
| 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_1173, var_1175))[name = tensor<string, []>("stack_0")]; | |
| tensor<int32, []> var_1177_transpose_batch_dims_0 = const()[name = tensor<string, []>("op_1177_transpose_batch_dims_0"), val = tensor<int32, []>(0)]; | |
| tensor<fp16, [1, 768]> var_1177_transpose_cast = gather_nd(batch_dims = var_1177_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast)[name = tensor<string, []>("op_1177_transpose_cast")]; | |
| tensor<string, []> var_1177_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_1177_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<fp32, [1, 77, 768]> last_hidden_state = cast(dtype = last_hidden_state_cast_to_fp32_dtype_0, x = last_hidden_state_cast)[name = tensor<string, []>("cast_0")]; | |
| tensor<fp32, [1, 768]> pooled_outputs = cast(dtype = var_1177_cast_to_fp32_dtype_0, x = var_1177_transpose_cast)[name = tensor<string, []>("cast_1")]; | |
| } -> (last_hidden_state, pooled_outputs); | |
| } |