| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.4"}, {"coremlc-version", "1839.0.0"}, {"coremltools-component-torch", "2.0.1+cu117"}, {"coremltools-version", "7.0b1"}})] |
| { |
| 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, 1280]> 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, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| tensor<int32, [1, 77]> cast_1322 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor<string, []>("cast_1322")]; |
| tensor<fp16, [1, 77, 1280]> inputs_embeds_cast = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_1322, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_cast")]; |
| tensor<fp16, [1, 77, 1280]> position_embeddings_to_fp16 = const()[name = tensor<string, []>("position_embeddings_to_fp16"), val = tensor<fp16, [1, 77, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126484608)))]; |
| tensor<fp16, [1, 77, 1280]> input_3_cast = add(x = inputs_embeds_cast, y = position_embeddings_to_fp16)[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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126681792)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126684416)))]; |
| tensor<fp16, []> var_12_to_fp16 = const()[name = tensor<string, []>("op_12_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126687040)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129963904)))]; |
| tensor<fp16, [1, 77, 1280]> var_128_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, x = hidden_states_1_cast)[name = tensor<string, []>("op_128_cast")]; |
| tensor<fp16, []> var_129_to_fp16 = const()[name = tensor<string, []>("op_129_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_5_cast = mul(x = var_128_cast, y = var_129_to_fp16)[name = tensor<string, []>("tensor_5_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129966528)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133243392)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_1_cast)[name = tensor<string, []>("tensor_1_cast")]; |
| tensor<int32, [4]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_135_cast = reshape(shape = var_134, x = tensor_1_cast)[name = tensor<string, []>("op_135_cast")]; |
| tensor<int32, [4]> var_136_perm_0 = const()[name = tensor<string, []>("op_136_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133246016)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136522880)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_1_cast)[name = tensor<string, []>("tensor_3_cast")]; |
| tensor<int32, [4]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_142_cast = reshape(shape = var_141, x = tensor_3_cast)[name = tensor<string, []>("op_142_cast")]; |
| tensor<int32, [4]> var_143_perm_0 = const()[name = tensor<string, []>("op_143_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_150 = const()[name = tensor<string, []>("op_150"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_151_cast = reshape(shape = var_150, x = tensor_5_cast)[name = tensor<string, []>("op_151_cast")]; |
| tensor<int32, [4]> var_152_perm_0 = const()[name = tensor<string, []>("op_152_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_154 = const()[name = tensor<string, []>("op_154"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_158 = transpose(perm = var_152_perm_0, x = var_151_cast)[name = tensor<string, []>("transpose_158")]; |
| tensor<fp16, [20, 77, 64]> query_states_1_cast = reshape(shape = var_154, x = transpose_158)[name = tensor<string, []>("query_states_1_cast")]; |
| tensor<int32, [3]> var_156 = const()[name = tensor<string, []>("op_156"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_160 = transpose(perm = var_136_perm_0, x = var_135_cast)[name = tensor<string, []>("transpose_160")]; |
| tensor<fp16, [20, 77, 64]> key_states_3_cast = reshape(shape = var_156, x = transpose_160)[name = tensor<string, []>("key_states_3_cast")]; |
| tensor<int32, [3]> var_158 = const()[name = tensor<string, []>("op_158"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_159 = transpose(perm = var_143_perm_0, x = var_142_cast)[name = tensor<string, []>("transpose_159")]; |
| tensor<fp16, [20, 77, 64]> value_states_3_cast = reshape(shape = var_158, x = transpose_159)[name = tensor<string, []>("value_states_3_cast")]; |
| tensor<int32, [3]> var_161_perm_0 = const()[name = tensor<string, []>("op_161_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, [20, 64, 77]> transpose_157 = transpose(perm = var_161_perm_0, x = key_states_3_cast)[name = tensor<string, []>("transpose_157")]; |
| tensor<fp16, [20, 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_157)[name = tensor<string, []>("attn_weights_1_cast")]; |
| tensor<int32, [4]> var_163 = const()[name = tensor<string, []>("op_163"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_164_cast = reshape(shape = var_163, x = attn_weights_1_cast)[name = tensor<string, []>("op_164_cast")]; |
| tensor<fp16, [1, 1, 77, 77]> causal_attention_mask_to_fp16 = const()[name = tensor<string, []>("causal_attention_mask_to_fp16"), val = tensor<fp16, [1, 1, 77, 77]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136525504)))]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_3_cast = add(x = var_164_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_3_cast")]; |
| tensor<int32, [3]> var_169 = const()[name = tensor<string, []>("op_169"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_5_cast = reshape(shape = var_169, x = attn_weights_3_cast)[name = tensor<string, []>("input_5_cast")]; |
| tensor<fp16, [20, 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, [20, 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_174 = const()[name = tensor<string, []>("op_174"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_3_cast = reshape(shape = var_174, 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_177 = const()[name = tensor<string, []>("op_177"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_156 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast)[name = tensor<string, []>("transpose_156")]; |
| tensor<fp16, [1, 77, 1280]> input_9_cast = reshape(shape = var_177, x = transpose_156)[name = tensor<string, []>("input_9_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136537472)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139814336)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_9_cast)[name = tensor<string, []>("hidden_states_3_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139816960)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139819584)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139822208)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152929472)))]; |
| tensor<fp16, [1, 77, 5120]> input_15_cast = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16, x = input_13_cast)[name = tensor<string, []>("input_15_cast")]; |
| tensor<string, []> input_17_mode_0 = const()[name = tensor<string, []>("input_17_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_17_cast = gelu(mode = input_17_mode_0, x = input_15_cast)[name = tensor<string, []>("input_17_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152939776)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166047040)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_17_cast)[name = tensor<string, []>("hidden_states_5_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166049664)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166052288)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166054912)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169331776)))]; |
| tensor<fp16, [1, 77, 1280]> var_215_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, x = hidden_states_7_cast)[name = tensor<string, []>("op_215_cast")]; |
| tensor<fp16, []> var_216_to_fp16 = const()[name = tensor<string, []>("op_216_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_11_cast = mul(x = var_215_cast, y = var_216_to_fp16)[name = tensor<string, []>("tensor_11_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169334400)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172611264)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_7_cast)[name = tensor<string, []>("tensor_7_cast")]; |
| tensor<int32, [4]> var_221 = const()[name = tensor<string, []>("op_221"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_222_cast = reshape(shape = var_221, x = tensor_7_cast)[name = tensor<string, []>("op_222_cast")]; |
| tensor<int32, [4]> var_223_perm_0 = const()[name = tensor<string, []>("op_223_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172613888)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175890752)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_7_cast)[name = tensor<string, []>("tensor_9_cast")]; |
| tensor<int32, [4]> var_228 = const()[name = tensor<string, []>("op_228"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_229_cast = reshape(shape = var_228, x = tensor_9_cast)[name = tensor<string, []>("op_229_cast")]; |
| tensor<int32, [4]> var_230_perm_0 = const()[name = tensor<string, []>("op_230_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_237 = const()[name = tensor<string, []>("op_237"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_238_cast = reshape(shape = var_237, x = tensor_11_cast)[name = tensor<string, []>("op_238_cast")]; |
| tensor<int32, [4]> var_239_perm_0 = const()[name = tensor<string, []>("op_239_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_241 = const()[name = tensor<string, []>("op_241"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_153 = transpose(perm = var_239_perm_0, x = var_238_cast)[name = tensor<string, []>("transpose_153")]; |
| tensor<fp16, [20, 77, 64]> query_states_3_cast = reshape(shape = var_241, x = transpose_153)[name = tensor<string, []>("query_states_3_cast")]; |
| tensor<int32, [3]> var_243 = const()[name = tensor<string, []>("op_243"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_155 = transpose(perm = var_223_perm_0, x = var_222_cast)[name = tensor<string, []>("transpose_155")]; |
| tensor<fp16, [20, 77, 64]> key_states_7_cast = reshape(shape = var_243, x = transpose_155)[name = tensor<string, []>("key_states_7_cast")]; |
| tensor<int32, [3]> var_245 = const()[name = tensor<string, []>("op_245"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_154 = transpose(perm = var_230_perm_0, x = var_229_cast)[name = tensor<string, []>("transpose_154")]; |
| tensor<fp16, [20, 77, 64]> value_states_7_cast = reshape(shape = var_245, x = transpose_154)[name = tensor<string, []>("value_states_7_cast")]; |
| tensor<int32, [3]> var_248_perm_0 = const()[name = tensor<string, []>("op_248_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, [20, 64, 77]> transpose_152 = transpose(perm = var_248_perm_0, x = key_states_7_cast)[name = tensor<string, []>("transpose_152")]; |
| tensor<fp16, [20, 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_152)[name = tensor<string, []>("attn_weights_7_cast")]; |
| tensor<int32, [4]> var_250 = const()[name = tensor<string, []>("op_250"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_251_cast = reshape(shape = var_250, x = attn_weights_7_cast)[name = tensor<string, []>("op_251_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_9_cast = add(x = var_251_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_9_cast")]; |
| tensor<int32, [3]> var_256 = const()[name = tensor<string, []>("op_256"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_21_cast = reshape(shape = var_256, x = attn_weights_9_cast)[name = tensor<string, []>("input_21_cast")]; |
| tensor<fp16, [20, 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, [20, 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_261 = const()[name = tensor<string, []>("op_261"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_9_cast = reshape(shape = var_261, 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_264 = const()[name = tensor<string, []>("op_264"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_151 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast)[name = tensor<string, []>("transpose_151")]; |
| tensor<fp16, [1, 77, 1280]> input_25_cast = reshape(shape = var_264, x = transpose_151)[name = tensor<string, []>("input_25_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175893376)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179170240)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_25_cast)[name = tensor<string, []>("hidden_states_9_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179172864)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179175488)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179178112)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192285376)))]; |
| tensor<fp16, [1, 77, 5120]> input_31_cast = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16, x = input_29_cast)[name = tensor<string, []>("input_31_cast")]; |
| tensor<string, []> input_33_mode_0 = const()[name = tensor<string, []>("input_33_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_33_cast = gelu(mode = input_33_mode_0, x = input_31_cast)[name = tensor<string, []>("input_33_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192295680)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205402944)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_33_cast)[name = tensor<string, []>("hidden_states_11_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205405568)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205408192)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205410816)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208687680)))]; |
| tensor<fp16, [1, 77, 1280]> var_302_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, x = hidden_states_13_cast)[name = tensor<string, []>("op_302_cast")]; |
| tensor<fp16, []> var_303_to_fp16 = const()[name = tensor<string, []>("op_303_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_17_cast = mul(x = var_302_cast, y = var_303_to_fp16)[name = tensor<string, []>("tensor_17_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208690304)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211967168)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_13_cast)[name = tensor<string, []>("tensor_13_cast")]; |
| tensor<int32, [4]> var_308 = const()[name = tensor<string, []>("op_308"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_309_cast = reshape(shape = var_308, x = tensor_13_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<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211969792)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215246656)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_13_cast)[name = tensor<string, []>("tensor_15_cast")]; |
| tensor<int32, [4]> var_315 = const()[name = tensor<string, []>("op_315"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_316_cast = reshape(shape = var_315, x = tensor_15_cast)[name = tensor<string, []>("op_316_cast")]; |
| tensor<int32, [4]> var_317_perm_0 = const()[name = tensor<string, []>("op_317_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_324 = const()[name = tensor<string, []>("op_324"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_325_cast = reshape(shape = var_324, x = tensor_17_cast)[name = tensor<string, []>("op_325_cast")]; |
| tensor<int32, [4]> var_326_perm_0 = const()[name = tensor<string, []>("op_326_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_328 = const()[name = tensor<string, []>("op_328"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_148 = transpose(perm = var_326_perm_0, x = var_325_cast)[name = tensor<string, []>("transpose_148")]; |
| tensor<fp16, [20, 77, 64]> query_states_5_cast = reshape(shape = var_328, x = transpose_148)[name = tensor<string, []>("query_states_5_cast")]; |
| tensor<int32, [3]> var_330 = const()[name = tensor<string, []>("op_330"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_150 = transpose(perm = var_310_perm_0, x = var_309_cast)[name = tensor<string, []>("transpose_150")]; |
| tensor<fp16, [20, 77, 64]> key_states_11_cast = reshape(shape = var_330, x = transpose_150)[name = tensor<string, []>("key_states_11_cast")]; |
| tensor<int32, [3]> var_332 = const()[name = tensor<string, []>("op_332"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_149 = transpose(perm = var_317_perm_0, x = var_316_cast)[name = tensor<string, []>("transpose_149")]; |
| tensor<fp16, [20, 77, 64]> value_states_11_cast = reshape(shape = var_332, x = transpose_149)[name = tensor<string, []>("value_states_11_cast")]; |
| tensor<int32, [3]> var_335_perm_0 = const()[name = tensor<string, []>("op_335_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, [20, 64, 77]> transpose_147 = transpose(perm = var_335_perm_0, x = key_states_11_cast)[name = tensor<string, []>("transpose_147")]; |
| tensor<fp16, [20, 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_147)[name = tensor<string, []>("attn_weights_13_cast")]; |
| tensor<int32, [4]> var_337 = const()[name = tensor<string, []>("op_337"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_338_cast = reshape(shape = var_337, x = attn_weights_13_cast)[name = tensor<string, []>("op_338_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_15_cast = add(x = var_338_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_15_cast")]; |
| tensor<int32, [3]> var_343 = const()[name = tensor<string, []>("op_343"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_37_cast = reshape(shape = var_343, x = attn_weights_15_cast)[name = tensor<string, []>("input_37_cast")]; |
| tensor<fp16, [20, 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, [20, 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_348 = const()[name = tensor<string, []>("op_348"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_15_cast = reshape(shape = var_348, 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_351 = const()[name = tensor<string, []>("op_351"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_146 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast)[name = tensor<string, []>("transpose_146")]; |
| tensor<fp16, [1, 77, 1280]> input_41_cast = reshape(shape = var_351, x = transpose_146)[name = tensor<string, []>("input_41_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215249280)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218526144)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_41_cast)[name = tensor<string, []>("hidden_states_15_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218528768)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218531392)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218534016)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231641280)))]; |
| tensor<fp16, [1, 77, 5120]> input_47_cast = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16, x = input_45_cast)[name = tensor<string, []>("input_47_cast")]; |
| tensor<string, []> input_49_mode_0 = const()[name = tensor<string, []>("input_49_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_49_cast = gelu(mode = input_49_mode_0, x = input_47_cast)[name = tensor<string, []>("input_49_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231651584)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244758848)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_49_cast)[name = tensor<string, []>("hidden_states_17_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244761472)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244764096)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244766720)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(248043584)))]; |
| tensor<fp16, [1, 77, 1280]> var_389_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, x = hidden_states_19_cast)[name = tensor<string, []>("op_389_cast")]; |
| tensor<fp16, []> var_390_to_fp16 = const()[name = tensor<string, []>("op_390_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_23_cast = mul(x = var_389_cast, y = var_390_to_fp16)[name = tensor<string, []>("tensor_23_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(248046208)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251323072)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_19_cast)[name = tensor<string, []>("tensor_19_cast")]; |
| tensor<int32, [4]> var_395 = const()[name = tensor<string, []>("op_395"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_396_cast = reshape(shape = var_395, x = tensor_19_cast)[name = tensor<string, []>("op_396_cast")]; |
| tensor<int32, [4]> var_397_perm_0 = const()[name = tensor<string, []>("op_397_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251325696)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254602560)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_19_cast)[name = tensor<string, []>("tensor_21_cast")]; |
| tensor<int32, [4]> var_402 = const()[name = tensor<string, []>("op_402"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_403_cast = reshape(shape = var_402, x = tensor_21_cast)[name = tensor<string, []>("op_403_cast")]; |
| tensor<int32, [4]> var_404_perm_0 = const()[name = tensor<string, []>("op_404_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_411 = const()[name = tensor<string, []>("op_411"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_412_cast = reshape(shape = var_411, x = tensor_23_cast)[name = tensor<string, []>("op_412_cast")]; |
| tensor<int32, [4]> var_413_perm_0 = const()[name = tensor<string, []>("op_413_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_415 = const()[name = tensor<string, []>("op_415"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_143 = transpose(perm = var_413_perm_0, x = var_412_cast)[name = tensor<string, []>("transpose_143")]; |
| tensor<fp16, [20, 77, 64]> query_states_7_cast = reshape(shape = var_415, x = transpose_143)[name = tensor<string, []>("query_states_7_cast")]; |
| tensor<int32, [3]> var_417 = const()[name = tensor<string, []>("op_417"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_145 = transpose(perm = var_397_perm_0, x = var_396_cast)[name = tensor<string, []>("transpose_145")]; |
| tensor<fp16, [20, 77, 64]> key_states_15_cast = reshape(shape = var_417, x = transpose_145)[name = tensor<string, []>("key_states_15_cast")]; |
| tensor<int32, [3]> var_419 = const()[name = tensor<string, []>("op_419"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_144 = transpose(perm = var_404_perm_0, x = var_403_cast)[name = tensor<string, []>("transpose_144")]; |
| tensor<fp16, [20, 77, 64]> value_states_15_cast = reshape(shape = var_419, x = transpose_144)[name = tensor<string, []>("value_states_15_cast")]; |
| tensor<int32, [3]> var_422_perm_0 = const()[name = tensor<string, []>("op_422_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, [20, 64, 77]> transpose_142 = transpose(perm = var_422_perm_0, x = key_states_15_cast)[name = tensor<string, []>("transpose_142")]; |
| tensor<fp16, [20, 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_142)[name = tensor<string, []>("attn_weights_19_cast")]; |
| tensor<int32, [4]> var_424 = const()[name = tensor<string, []>("op_424"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_425_cast = reshape(shape = var_424, x = attn_weights_19_cast)[name = tensor<string, []>("op_425_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_21_cast = add(x = var_425_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_21_cast")]; |
| tensor<int32, [3]> var_430 = const()[name = tensor<string, []>("op_430"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_53_cast = reshape(shape = var_430, x = attn_weights_21_cast)[name = tensor<string, []>("input_53_cast")]; |
| tensor<fp16, [20, 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, [20, 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_435 = const()[name = tensor<string, []>("op_435"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_21_cast = reshape(shape = var_435, 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_438 = const()[name = tensor<string, []>("op_438"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_141 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast)[name = tensor<string, []>("transpose_141")]; |
| tensor<fp16, [1, 77, 1280]> input_57_cast = reshape(shape = var_438, x = transpose_141)[name = tensor<string, []>("input_57_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254605184)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257882048)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_57_cast)[name = tensor<string, []>("hidden_states_21_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257884672)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257887296)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257889920)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(270997184)))]; |
| tensor<fp16, [1, 77, 5120]> input_63_cast = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16, x = input_61_cast)[name = tensor<string, []>("input_63_cast")]; |
| tensor<string, []> input_65_mode_0 = const()[name = tensor<string, []>("input_65_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_65_cast = gelu(mode = input_65_mode_0, x = input_63_cast)[name = tensor<string, []>("input_65_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271007488)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284114752)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_65_cast)[name = tensor<string, []>("hidden_states_23_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284117376)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284120000)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284122624)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287399488)))]; |
| tensor<fp16, [1, 77, 1280]> var_476_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, x = hidden_states_25_cast)[name = tensor<string, []>("op_476_cast")]; |
| tensor<fp16, []> var_477_to_fp16 = const()[name = tensor<string, []>("op_477_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_29_cast = mul(x = var_476_cast, y = var_477_to_fp16)[name = tensor<string, []>("tensor_29_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287402112)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290678976)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_25_cast)[name = tensor<string, []>("tensor_25_cast")]; |
| tensor<int32, [4]> var_482 = const()[name = tensor<string, []>("op_482"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_483_cast = reshape(shape = var_482, x = tensor_25_cast)[name = tensor<string, []>("op_483_cast")]; |
| tensor<int32, [4]> var_484_perm_0 = const()[name = tensor<string, []>("op_484_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290681600)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293958464)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_25_cast)[name = tensor<string, []>("tensor_27_cast")]; |
| tensor<int32, [4]> var_489 = const()[name = tensor<string, []>("op_489"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_490_cast = reshape(shape = var_489, x = tensor_27_cast)[name = tensor<string, []>("op_490_cast")]; |
| tensor<int32, [4]> var_491_perm_0 = const()[name = tensor<string, []>("op_491_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_498 = const()[name = tensor<string, []>("op_498"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_499_cast = reshape(shape = var_498, x = tensor_29_cast)[name = tensor<string, []>("op_499_cast")]; |
| tensor<int32, [4]> var_500_perm_0 = const()[name = tensor<string, []>("op_500_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_502 = const()[name = tensor<string, []>("op_502"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_138 = transpose(perm = var_500_perm_0, x = var_499_cast)[name = tensor<string, []>("transpose_138")]; |
| tensor<fp16, [20, 77, 64]> query_states_9_cast = reshape(shape = var_502, x = transpose_138)[name = tensor<string, []>("query_states_9_cast")]; |
| tensor<int32, [3]> var_504 = const()[name = tensor<string, []>("op_504"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_140 = transpose(perm = var_484_perm_0, x = var_483_cast)[name = tensor<string, []>("transpose_140")]; |
| tensor<fp16, [20, 77, 64]> key_states_19_cast = reshape(shape = var_504, x = transpose_140)[name = tensor<string, []>("key_states_19_cast")]; |
| tensor<int32, [3]> var_506 = const()[name = tensor<string, []>("op_506"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_139 = transpose(perm = var_491_perm_0, x = var_490_cast)[name = tensor<string, []>("transpose_139")]; |
| tensor<fp16, [20, 77, 64]> value_states_19_cast = reshape(shape = var_506, x = transpose_139)[name = tensor<string, []>("value_states_19_cast")]; |
| tensor<int32, [3]> var_509_perm_0 = const()[name = tensor<string, []>("op_509_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, [20, 64, 77]> transpose_137 = transpose(perm = var_509_perm_0, x = key_states_19_cast)[name = tensor<string, []>("transpose_137")]; |
| tensor<fp16, [20, 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_137)[name = tensor<string, []>("attn_weights_25_cast")]; |
| tensor<int32, [4]> var_511 = const()[name = tensor<string, []>("op_511"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_512_cast = reshape(shape = var_511, x = attn_weights_25_cast)[name = tensor<string, []>("op_512_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_27_cast = add(x = var_512_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_27_cast")]; |
| tensor<int32, [3]> var_517 = const()[name = tensor<string, []>("op_517"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_69_cast = reshape(shape = var_517, x = attn_weights_27_cast)[name = tensor<string, []>("input_69_cast")]; |
| tensor<fp16, [20, 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, [20, 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_522 = const()[name = tensor<string, []>("op_522"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_27_cast = reshape(shape = var_522, 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_525 = const()[name = tensor<string, []>("op_525"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_136 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast)[name = tensor<string, []>("transpose_136")]; |
| tensor<fp16, [1, 77, 1280]> input_73_cast = reshape(shape = var_525, x = transpose_136)[name = tensor<string, []>("input_73_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293961088)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297237952)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_73_cast)[name = tensor<string, []>("hidden_states_27_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297240576)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297243200)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297245824)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310353088)))]; |
| tensor<fp16, [1, 77, 5120]> input_79_cast = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16, x = input_77_cast)[name = tensor<string, []>("input_79_cast")]; |
| tensor<string, []> input_81_mode_0 = const()[name = tensor<string, []>("input_81_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_81_cast = gelu(mode = input_81_mode_0, x = input_79_cast)[name = tensor<string, []>("input_81_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310363392)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323470656)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_81_cast)[name = tensor<string, []>("hidden_states_29_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323473280)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323475904)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323478528)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326755392)))]; |
| tensor<fp16, [1, 77, 1280]> var_563_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, x = hidden_states_31_cast)[name = tensor<string, []>("op_563_cast")]; |
| tensor<fp16, []> var_564_to_fp16 = const()[name = tensor<string, []>("op_564_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_35_cast = mul(x = var_563_cast, y = var_564_to_fp16)[name = tensor<string, []>("tensor_35_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(326758016)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330034880)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_31_cast)[name = tensor<string, []>("tensor_31_cast")]; |
| tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_570_cast = reshape(shape = var_569, x = tensor_31_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<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(330037504)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333314368)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_31_cast)[name = tensor<string, []>("tensor_33_cast")]; |
| tensor<int32, [4]> var_576 = const()[name = tensor<string, []>("op_576"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_577_cast = reshape(shape = var_576, x = tensor_33_cast)[name = tensor<string, []>("op_577_cast")]; |
| tensor<int32, [4]> var_578_perm_0 = const()[name = tensor<string, []>("op_578_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_585 = const()[name = tensor<string, []>("op_585"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_586_cast = reshape(shape = var_585, x = tensor_35_cast)[name = tensor<string, []>("op_586_cast")]; |
| tensor<int32, [4]> var_587_perm_0 = const()[name = tensor<string, []>("op_587_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_589 = const()[name = tensor<string, []>("op_589"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_133 = transpose(perm = var_587_perm_0, x = var_586_cast)[name = tensor<string, []>("transpose_133")]; |
| tensor<fp16, [20, 77, 64]> query_states_11_cast = reshape(shape = var_589, x = transpose_133)[name = tensor<string, []>("query_states_11_cast")]; |
| tensor<int32, [3]> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_135 = transpose(perm = var_571_perm_0, x = var_570_cast)[name = tensor<string, []>("transpose_135")]; |
| tensor<fp16, [20, 77, 64]> key_states_23_cast = reshape(shape = var_591, x = transpose_135)[name = tensor<string, []>("key_states_23_cast")]; |
| tensor<int32, [3]> var_593 = const()[name = tensor<string, []>("op_593"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_134 = transpose(perm = var_578_perm_0, x = var_577_cast)[name = tensor<string, []>("transpose_134")]; |
| tensor<fp16, [20, 77, 64]> value_states_23_cast = reshape(shape = var_593, x = transpose_134)[name = tensor<string, []>("value_states_23_cast")]; |
| tensor<int32, [3]> var_596_perm_0 = const()[name = tensor<string, []>("op_596_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, [20, 64, 77]> transpose_132 = transpose(perm = var_596_perm_0, x = key_states_23_cast)[name = tensor<string, []>("transpose_132")]; |
| tensor<fp16, [20, 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_132)[name = tensor<string, []>("attn_weights_31_cast")]; |
| tensor<int32, [4]> var_598 = const()[name = tensor<string, []>("op_598"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_599_cast = reshape(shape = var_598, x = attn_weights_31_cast)[name = tensor<string, []>("op_599_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_33_cast = add(x = var_599_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_33_cast")]; |
| tensor<int32, [3]> var_604 = const()[name = tensor<string, []>("op_604"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_85_cast = reshape(shape = var_604, x = attn_weights_33_cast)[name = tensor<string, []>("input_85_cast")]; |
| tensor<fp16, [20, 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, [20, 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_609 = const()[name = tensor<string, []>("op_609"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_33_cast = reshape(shape = var_609, 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_612 = const()[name = tensor<string, []>("op_612"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_131 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast)[name = tensor<string, []>("transpose_131")]; |
| tensor<fp16, [1, 77, 1280]> input_89_cast = reshape(shape = var_612, x = transpose_131)[name = tensor<string, []>("input_89_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333316992)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336593856)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_89_cast)[name = tensor<string, []>("hidden_states_33_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336596480)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336599104)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(336601728)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(349708992)))]; |
| tensor<fp16, [1, 77, 5120]> input_95_cast = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16, x = input_93_cast)[name = tensor<string, []>("input_95_cast")]; |
| tensor<string, []> input_97_mode_0 = const()[name = tensor<string, []>("input_97_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_97_cast = gelu(mode = input_97_mode_0, x = input_95_cast)[name = tensor<string, []>("input_97_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(349719296)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362826560)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_97_cast)[name = tensor<string, []>("hidden_states_35_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362829184)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362831808)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(362834432)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(366111296)))]; |
| tensor<fp16, [1, 77, 1280]> var_650_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, x = hidden_states_37_cast)[name = tensor<string, []>("op_650_cast")]; |
| tensor<fp16, []> var_651_to_fp16 = const()[name = tensor<string, []>("op_651_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_41_cast = mul(x = var_650_cast, y = var_651_to_fp16)[name = tensor<string, []>("tensor_41_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(366113920)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(369390784)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_37_cast)[name = tensor<string, []>("tensor_37_cast")]; |
| tensor<int32, [4]> var_656 = const()[name = tensor<string, []>("op_656"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_657_cast = reshape(shape = var_656, x = tensor_37_cast)[name = tensor<string, []>("op_657_cast")]; |
| tensor<int32, [4]> var_658_perm_0 = const()[name = tensor<string, []>("op_658_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(369393408)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372670272)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_37_cast)[name = tensor<string, []>("tensor_39_cast")]; |
| tensor<int32, [4]> var_663 = const()[name = tensor<string, []>("op_663"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_664_cast = reshape(shape = var_663, x = tensor_39_cast)[name = tensor<string, []>("op_664_cast")]; |
| tensor<int32, [4]> var_665_perm_0 = const()[name = tensor<string, []>("op_665_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_672 = const()[name = tensor<string, []>("op_672"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_673_cast = reshape(shape = var_672, x = tensor_41_cast)[name = tensor<string, []>("op_673_cast")]; |
| tensor<int32, [4]> var_674_perm_0 = const()[name = tensor<string, []>("op_674_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_128 = transpose(perm = var_674_perm_0, x = var_673_cast)[name = tensor<string, []>("transpose_128")]; |
| tensor<fp16, [20, 77, 64]> query_states_13_cast = reshape(shape = var_676, x = transpose_128)[name = tensor<string, []>("query_states_13_cast")]; |
| tensor<int32, [3]> var_678 = const()[name = tensor<string, []>("op_678"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_130 = transpose(perm = var_658_perm_0, x = var_657_cast)[name = tensor<string, []>("transpose_130")]; |
| tensor<fp16, [20, 77, 64]> key_states_27_cast = reshape(shape = var_678, x = transpose_130)[name = tensor<string, []>("key_states_27_cast")]; |
| tensor<int32, [3]> var_680 = const()[name = tensor<string, []>("op_680"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_129 = transpose(perm = var_665_perm_0, x = var_664_cast)[name = tensor<string, []>("transpose_129")]; |
| tensor<fp16, [20, 77, 64]> value_states_27_cast = reshape(shape = var_680, x = transpose_129)[name = tensor<string, []>("value_states_27_cast")]; |
| tensor<int32, [3]> var_683_perm_0 = const()[name = tensor<string, []>("op_683_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, [20, 64, 77]> transpose_127 = transpose(perm = var_683_perm_0, x = key_states_27_cast)[name = tensor<string, []>("transpose_127")]; |
| tensor<fp16, [20, 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_127)[name = tensor<string, []>("attn_weights_37_cast")]; |
| tensor<int32, [4]> var_685 = const()[name = tensor<string, []>("op_685"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_686_cast = reshape(shape = var_685, x = attn_weights_37_cast)[name = tensor<string, []>("op_686_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_39_cast = add(x = var_686_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_39_cast")]; |
| tensor<int32, [3]> var_691 = const()[name = tensor<string, []>("op_691"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_101_cast = reshape(shape = var_691, x = attn_weights_39_cast)[name = tensor<string, []>("input_101_cast")]; |
| tensor<fp16, [20, 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, [20, 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_696 = const()[name = tensor<string, []>("op_696"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_39_cast = reshape(shape = var_696, 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_699 = const()[name = tensor<string, []>("op_699"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_126 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast)[name = tensor<string, []>("transpose_126")]; |
| tensor<fp16, [1, 77, 1280]> input_105_cast = reshape(shape = var_699, x = transpose_126)[name = tensor<string, []>("input_105_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(372672896)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375949760)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_105_cast)[name = tensor<string, []>("hidden_states_39_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375952384)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375955008)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375957632)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389064896)))]; |
| tensor<fp16, [1, 77, 5120]> input_111_cast = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16, x = input_109_cast)[name = tensor<string, []>("input_111_cast")]; |
| tensor<string, []> input_113_mode_0 = const()[name = tensor<string, []>("input_113_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_113_cast = gelu(mode = input_113_mode_0, x = input_111_cast)[name = tensor<string, []>("input_113_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389075200)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(402182464)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_113_cast)[name = tensor<string, []>("hidden_states_41_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(402185088)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(402187712)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(402190336)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(405467200)))]; |
| tensor<fp16, [1, 77, 1280]> var_737_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, x = hidden_states_43_cast)[name = tensor<string, []>("op_737_cast")]; |
| tensor<fp16, []> var_738_to_fp16 = const()[name = tensor<string, []>("op_738_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_47_cast = mul(x = var_737_cast, y = var_738_to_fp16)[name = tensor<string, []>("tensor_47_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(405469824)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408746688)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_43_cast)[name = tensor<string, []>("tensor_43_cast")]; |
| tensor<int32, [4]> var_743 = const()[name = tensor<string, []>("op_743"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_744_cast = reshape(shape = var_743, x = tensor_43_cast)[name = tensor<string, []>("op_744_cast")]; |
| tensor<int32, [4]> var_745_perm_0 = const()[name = tensor<string, []>("op_745_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(408749312)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412026176)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_43_cast)[name = tensor<string, []>("tensor_45_cast")]; |
| tensor<int32, [4]> var_750 = const()[name = tensor<string, []>("op_750"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_751_cast = reshape(shape = var_750, x = tensor_45_cast)[name = tensor<string, []>("op_751_cast")]; |
| tensor<int32, [4]> var_752_perm_0 = const()[name = tensor<string, []>("op_752_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_759 = const()[name = tensor<string, []>("op_759"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_760_cast = reshape(shape = var_759, x = tensor_47_cast)[name = tensor<string, []>("op_760_cast")]; |
| tensor<int32, [4]> var_761_perm_0 = const()[name = tensor<string, []>("op_761_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_763 = const()[name = tensor<string, []>("op_763"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_123 = transpose(perm = var_761_perm_0, x = var_760_cast)[name = tensor<string, []>("transpose_123")]; |
| tensor<fp16, [20, 77, 64]> query_states_15_cast = reshape(shape = var_763, x = transpose_123)[name = tensor<string, []>("query_states_15_cast")]; |
| tensor<int32, [3]> var_765 = const()[name = tensor<string, []>("op_765"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_125 = transpose(perm = var_745_perm_0, x = var_744_cast)[name = tensor<string, []>("transpose_125")]; |
| tensor<fp16, [20, 77, 64]> key_states_31_cast = reshape(shape = var_765, x = transpose_125)[name = tensor<string, []>("key_states_31_cast")]; |
| tensor<int32, [3]> var_767 = const()[name = tensor<string, []>("op_767"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_124 = transpose(perm = var_752_perm_0, x = var_751_cast)[name = tensor<string, []>("transpose_124")]; |
| tensor<fp16, [20, 77, 64]> value_states_31_cast = reshape(shape = var_767, x = transpose_124)[name = tensor<string, []>("value_states_31_cast")]; |
| tensor<int32, [3]> var_770_perm_0 = const()[name = tensor<string, []>("op_770_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, [20, 64, 77]> transpose_122 = transpose(perm = var_770_perm_0, x = key_states_31_cast)[name = tensor<string, []>("transpose_122")]; |
| tensor<fp16, [20, 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_122)[name = tensor<string, []>("attn_weights_43_cast")]; |
| tensor<int32, [4]> var_772 = const()[name = tensor<string, []>("op_772"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_773_cast = reshape(shape = var_772, x = attn_weights_43_cast)[name = tensor<string, []>("op_773_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_45_cast = add(x = var_773_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_45_cast")]; |
| tensor<int32, [3]> var_778 = const()[name = tensor<string, []>("op_778"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_117_cast = reshape(shape = var_778, x = attn_weights_45_cast)[name = tensor<string, []>("input_117_cast")]; |
| tensor<fp16, [20, 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, [20, 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_783 = const()[name = tensor<string, []>("op_783"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_45_cast = reshape(shape = var_783, 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_786 = const()[name = tensor<string, []>("op_786"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_121 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast)[name = tensor<string, []>("transpose_121")]; |
| tensor<fp16, [1, 77, 1280]> input_121_cast = reshape(shape = var_786, x = transpose_121)[name = tensor<string, []>("input_121_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412028800)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415305664)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_121_cast)[name = tensor<string, []>("hidden_states_45_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415308288)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415310912)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415313536)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428420800)))]; |
| tensor<fp16, [1, 77, 5120]> input_127_cast = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16, x = input_125_cast)[name = tensor<string, []>("input_127_cast")]; |
| tensor<string, []> input_129_mode_0 = const()[name = tensor<string, []>("input_129_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_129_cast = gelu(mode = input_129_mode_0, x = input_127_cast)[name = tensor<string, []>("input_129_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428431104)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(441538368)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_129_cast)[name = tensor<string, []>("hidden_states_47_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(441540992)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(441543616)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(441546240)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444823104)))]; |
| tensor<fp16, [1, 77, 1280]> var_824_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, x = hidden_states_49_cast)[name = tensor<string, []>("op_824_cast")]; |
| tensor<fp16, []> var_825_to_fp16 = const()[name = tensor<string, []>("op_825_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_53_cast = mul(x = var_824_cast, y = var_825_to_fp16)[name = tensor<string, []>("tensor_53_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(444825728)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448102592)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_49_cast)[name = tensor<string, []>("tensor_49_cast")]; |
| tensor<int32, [4]> var_830 = const()[name = tensor<string, []>("op_830"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_831_cast = reshape(shape = var_830, x = tensor_49_cast)[name = tensor<string, []>("op_831_cast")]; |
| tensor<int32, [4]> var_832_perm_0 = const()[name = tensor<string, []>("op_832_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448105216)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(451382080)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_49_cast)[name = tensor<string, []>("tensor_51_cast")]; |
| tensor<int32, [4]> var_837 = const()[name = tensor<string, []>("op_837"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_838_cast = reshape(shape = var_837, x = tensor_51_cast)[name = tensor<string, []>("op_838_cast")]; |
| tensor<int32, [4]> var_839_perm_0 = const()[name = tensor<string, []>("op_839_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_846 = const()[name = tensor<string, []>("op_846"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_847_cast = reshape(shape = var_846, x = tensor_53_cast)[name = tensor<string, []>("op_847_cast")]; |
| tensor<int32, [4]> var_848_perm_0 = const()[name = tensor<string, []>("op_848_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_850 = const()[name = tensor<string, []>("op_850"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_118 = transpose(perm = var_848_perm_0, x = var_847_cast)[name = tensor<string, []>("transpose_118")]; |
| tensor<fp16, [20, 77, 64]> query_states_17_cast = reshape(shape = var_850, x = transpose_118)[name = tensor<string, []>("query_states_17_cast")]; |
| tensor<int32, [3]> var_852 = const()[name = tensor<string, []>("op_852"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_120 = transpose(perm = var_832_perm_0, x = var_831_cast)[name = tensor<string, []>("transpose_120")]; |
| tensor<fp16, [20, 77, 64]> key_states_35_cast = reshape(shape = var_852, x = transpose_120)[name = tensor<string, []>("key_states_35_cast")]; |
| tensor<int32, [3]> var_854 = const()[name = tensor<string, []>("op_854"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_119 = transpose(perm = var_839_perm_0, x = var_838_cast)[name = tensor<string, []>("transpose_119")]; |
| tensor<fp16, [20, 77, 64]> value_states_35_cast = reshape(shape = var_854, x = transpose_119)[name = tensor<string, []>("value_states_35_cast")]; |
| tensor<int32, [3]> var_857_perm_0 = const()[name = tensor<string, []>("op_857_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, [20, 64, 77]> transpose_117 = transpose(perm = var_857_perm_0, x = key_states_35_cast)[name = tensor<string, []>("transpose_117")]; |
| tensor<fp16, [20, 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_117)[name = tensor<string, []>("attn_weights_49_cast")]; |
| tensor<int32, [4]> var_859 = const()[name = tensor<string, []>("op_859"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_860_cast = reshape(shape = var_859, x = attn_weights_49_cast)[name = tensor<string, []>("op_860_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_51_cast = add(x = var_860_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_51_cast")]; |
| tensor<int32, [3]> var_865 = const()[name = tensor<string, []>("op_865"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_133_cast = reshape(shape = var_865, x = attn_weights_51_cast)[name = tensor<string, []>("input_133_cast")]; |
| tensor<fp16, [20, 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, [20, 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_870 = const()[name = tensor<string, []>("op_870"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_51_cast = reshape(shape = var_870, 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_873 = const()[name = tensor<string, []>("op_873"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_116 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast)[name = tensor<string, []>("transpose_116")]; |
| tensor<fp16, [1, 77, 1280]> input_137_cast = reshape(shape = var_873, x = transpose_116)[name = tensor<string, []>("input_137_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(451384704)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454661568)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_137_cast)[name = tensor<string, []>("hidden_states_51_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454664192)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454666816)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(454669440)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467776704)))]; |
| tensor<fp16, [1, 77, 5120]> input_143_cast = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16, x = input_141_cast)[name = tensor<string, []>("input_143_cast")]; |
| tensor<string, []> input_145_mode_0 = const()[name = tensor<string, []>("input_145_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_145_cast = gelu(mode = input_145_mode_0, x = input_143_cast)[name = tensor<string, []>("input_145_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467787008)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480894272)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_145_cast)[name = tensor<string, []>("hidden_states_53_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480896896)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480899520)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(480902144)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(484179008)))]; |
| tensor<fp16, [1, 77, 1280]> var_911_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, x = hidden_states_55_cast)[name = tensor<string, []>("op_911_cast")]; |
| tensor<fp16, []> var_912_to_fp16 = const()[name = tensor<string, []>("op_912_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_59_cast = mul(x = var_911_cast, y = var_912_to_fp16)[name = tensor<string, []>("tensor_59_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(484181632)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487458496)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_55_cast)[name = tensor<string, []>("tensor_55_cast")]; |
| tensor<int32, [4]> var_917 = const()[name = tensor<string, []>("op_917"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_918_cast = reshape(shape = var_917, x = tensor_55_cast)[name = tensor<string, []>("op_918_cast")]; |
| tensor<int32, [4]> var_919_perm_0 = const()[name = tensor<string, []>("op_919_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487461120)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490737984)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_55_cast)[name = tensor<string, []>("tensor_57_cast")]; |
| tensor<int32, [4]> var_924 = const()[name = tensor<string, []>("op_924"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_925_cast = reshape(shape = var_924, x = tensor_57_cast)[name = tensor<string, []>("op_925_cast")]; |
| tensor<int32, [4]> var_926_perm_0 = const()[name = tensor<string, []>("op_926_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_933 = const()[name = tensor<string, []>("op_933"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_934_cast = reshape(shape = var_933, x = tensor_59_cast)[name = tensor<string, []>("op_934_cast")]; |
| tensor<int32, [4]> var_935_perm_0 = const()[name = tensor<string, []>("op_935_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_937 = const()[name = tensor<string, []>("op_937"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_113 = transpose(perm = var_935_perm_0, x = var_934_cast)[name = tensor<string, []>("transpose_113")]; |
| tensor<fp16, [20, 77, 64]> query_states_19_cast = reshape(shape = var_937, x = transpose_113)[name = tensor<string, []>("query_states_19_cast")]; |
| tensor<int32, [3]> var_939 = const()[name = tensor<string, []>("op_939"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_115 = transpose(perm = var_919_perm_0, x = var_918_cast)[name = tensor<string, []>("transpose_115")]; |
| tensor<fp16, [20, 77, 64]> key_states_39_cast = reshape(shape = var_939, x = transpose_115)[name = tensor<string, []>("key_states_39_cast")]; |
| tensor<int32, [3]> var_941 = const()[name = tensor<string, []>("op_941"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_114 = transpose(perm = var_926_perm_0, x = var_925_cast)[name = tensor<string, []>("transpose_114")]; |
| tensor<fp16, [20, 77, 64]> value_states_39_cast = reshape(shape = var_941, x = transpose_114)[name = tensor<string, []>("value_states_39_cast")]; |
| tensor<int32, [3]> var_944_perm_0 = const()[name = tensor<string, []>("op_944_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, [20, 64, 77]> transpose_112 = transpose(perm = var_944_perm_0, x = key_states_39_cast)[name = tensor<string, []>("transpose_112")]; |
| tensor<fp16, [20, 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_112)[name = tensor<string, []>("attn_weights_55_cast")]; |
| tensor<int32, [4]> var_946 = const()[name = tensor<string, []>("op_946"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_947_cast = reshape(shape = var_946, x = attn_weights_55_cast)[name = tensor<string, []>("op_947_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_57_cast = add(x = var_947_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_57_cast")]; |
| tensor<int32, [3]> var_952 = const()[name = tensor<string, []>("op_952"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_149_cast = reshape(shape = var_952, x = attn_weights_57_cast)[name = tensor<string, []>("input_149_cast")]; |
| tensor<fp16, [20, 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, [20, 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_957 = const()[name = tensor<string, []>("op_957"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_57_cast = reshape(shape = var_957, 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_960 = const()[name = tensor<string, []>("op_960"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_111 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast)[name = tensor<string, []>("transpose_111")]; |
| tensor<fp16, [1, 77, 1280]> input_153_cast = reshape(shape = var_960, x = transpose_111)[name = tensor<string, []>("input_153_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490740608)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494017472)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_153_cast)[name = tensor<string, []>("hidden_states_57_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494020096)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494022720)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(494025344)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(507132608)))]; |
| tensor<fp16, [1, 77, 5120]> input_159_cast = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16, x = input_157_cast)[name = tensor<string, []>("input_159_cast")]; |
| tensor<string, []> input_161_mode_0 = const()[name = tensor<string, []>("input_161_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_161_cast = gelu(mode = input_161_mode_0, x = input_159_cast)[name = tensor<string, []>("input_161_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(507142912)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520250176)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_161_cast)[name = tensor<string, []>("hidden_states_59_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520252800)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520255424)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520258048)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(523534912)))]; |
| tensor<fp16, [1, 77, 1280]> var_998_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, x = hidden_states_61_cast)[name = tensor<string, []>("op_998_cast")]; |
| tensor<fp16, []> var_999_to_fp16 = const()[name = tensor<string, []>("op_999_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_65_cast = mul(x = var_998_cast, y = var_999_to_fp16)[name = tensor<string, []>("tensor_65_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(523537536)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526814400)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_61_cast)[name = tensor<string, []>("tensor_61_cast")]; |
| tensor<int32, [4]> var_1004 = const()[name = tensor<string, []>("op_1004"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1005_cast = reshape(shape = var_1004, x = tensor_61_cast)[name = tensor<string, []>("op_1005_cast")]; |
| tensor<int32, [4]> var_1006_perm_0 = const()[name = tensor<string, []>("op_1006_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526817024)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(530093888)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_61_cast)[name = tensor<string, []>("tensor_63_cast")]; |
| tensor<int32, [4]> var_1011 = const()[name = tensor<string, []>("op_1011"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1012_cast = reshape(shape = var_1011, x = tensor_63_cast)[name = tensor<string, []>("op_1012_cast")]; |
| tensor<int32, [4]> var_1013_perm_0 = const()[name = tensor<string, []>("op_1013_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1020 = const()[name = tensor<string, []>("op_1020"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1021_cast = reshape(shape = var_1020, x = tensor_65_cast)[name = tensor<string, []>("op_1021_cast")]; |
| tensor<int32, [4]> var_1022_perm_0 = const()[name = tensor<string, []>("op_1022_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1024 = const()[name = tensor<string, []>("op_1024"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_108 = transpose(perm = var_1022_perm_0, x = var_1021_cast)[name = tensor<string, []>("transpose_108")]; |
| tensor<fp16, [20, 77, 64]> query_states_21_cast = reshape(shape = var_1024, x = transpose_108)[name = tensor<string, []>("query_states_21_cast")]; |
| tensor<int32, [3]> var_1026 = const()[name = tensor<string, []>("op_1026"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_110 = transpose(perm = var_1006_perm_0, x = var_1005_cast)[name = tensor<string, []>("transpose_110")]; |
| tensor<fp16, [20, 77, 64]> key_states_43_cast = reshape(shape = var_1026, x = transpose_110)[name = tensor<string, []>("key_states_43_cast")]; |
| tensor<int32, [3]> var_1028 = const()[name = tensor<string, []>("op_1028"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_109 = transpose(perm = var_1013_perm_0, x = var_1012_cast)[name = tensor<string, []>("transpose_109")]; |
| tensor<fp16, [20, 77, 64]> value_states_43_cast = reshape(shape = var_1028, x = transpose_109)[name = tensor<string, []>("value_states_43_cast")]; |
| tensor<int32, [3]> var_1031_perm_0 = const()[name = tensor<string, []>("op_1031_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, [20, 64, 77]> transpose_107 = transpose(perm = var_1031_perm_0, x = key_states_43_cast)[name = tensor<string, []>("transpose_107")]; |
| tensor<fp16, [20, 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_107)[name = tensor<string, []>("attn_weights_61_cast")]; |
| tensor<int32, [4]> var_1033 = const()[name = tensor<string, []>("op_1033"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1034_cast = reshape(shape = var_1033, x = attn_weights_61_cast)[name = tensor<string, []>("op_1034_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_63_cast = add(x = var_1034_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_63_cast")]; |
| tensor<int32, [3]> var_1039 = const()[name = tensor<string, []>("op_1039"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_165_cast = reshape(shape = var_1039, x = attn_weights_63_cast)[name = tensor<string, []>("input_165_cast")]; |
| tensor<fp16, [20, 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, [20, 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_1044 = const()[name = tensor<string, []>("op_1044"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_63_cast = reshape(shape = var_1044, 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_1047 = const()[name = tensor<string, []>("op_1047"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_106 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast)[name = tensor<string, []>("transpose_106")]; |
| tensor<fp16, [1, 77, 1280]> input_169_cast = reshape(shape = var_1047, x = transpose_106)[name = tensor<string, []>("input_169_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(530096512)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533373376)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_169_cast)[name = tensor<string, []>("hidden_states_63_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533376000)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533378624)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(533381248)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(546488512)))]; |
| tensor<fp16, [1, 77, 5120]> input_175_cast = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16, x = input_173_cast)[name = tensor<string, []>("input_175_cast")]; |
| tensor<string, []> input_177_mode_0 = const()[name = tensor<string, []>("input_177_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_177_cast = gelu(mode = input_177_mode_0, x = input_175_cast)[name = tensor<string, []>("input_177_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(546498816)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559606080)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_177_cast)[name = tensor<string, []>("hidden_states_65_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559608704)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559611328)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [1280, 1280]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559613952)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562890816)))]; |
| tensor<fp16, [1, 77, 1280]> var_1085_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, x = hidden_states_67_cast)[name = tensor<string, []>("op_1085_cast")]; |
| tensor<fp16, []> var_1086_to_fp16 = const()[name = tensor<string, []>("op_1086_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_71_cast = mul(x = var_1085_cast, y = var_1086_to_fp16)[name = tensor<string, []>("tensor_71_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562893440)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566170304)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_67_cast)[name = tensor<string, []>("tensor_67_cast")]; |
| tensor<int32, [4]> var_1091 = const()[name = tensor<string, []>("op_1091"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1092_cast = reshape(shape = var_1091, x = tensor_67_cast)[name = tensor<string, []>("op_1092_cast")]; |
| tensor<int32, [4]> var_1093_perm_0 = const()[name = tensor<string, []>("op_1093_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566172928)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(569449792)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = hidden_states_67_cast)[name = tensor<string, []>("tensor_69_cast")]; |
| tensor<int32, [4]> var_1098 = const()[name = tensor<string, []>("op_1098"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1099_cast = reshape(shape = var_1098, x = tensor_69_cast)[name = tensor<string, []>("op_1099_cast")]; |
| tensor<int32, [4]> var_1100_perm_0 = const()[name = tensor<string, []>("op_1100_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1107 = const()[name = tensor<string, []>("op_1107"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1108_cast = reshape(shape = var_1107, x = tensor_71_cast)[name = tensor<string, []>("op_1108_cast")]; |
| tensor<int32, [4]> var_1109_perm_0 = const()[name = tensor<string, []>("op_1109_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1111 = const()[name = tensor<string, []>("op_1111"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_103 = transpose(perm = var_1109_perm_0, x = var_1108_cast)[name = tensor<string, []>("transpose_103")]; |
| tensor<fp16, [20, 77, 64]> query_states_23_cast = reshape(shape = var_1111, x = transpose_103)[name = tensor<string, []>("query_states_23_cast")]; |
| tensor<int32, [3]> var_1113 = const()[name = tensor<string, []>("op_1113"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_105 = transpose(perm = var_1093_perm_0, x = var_1092_cast)[name = tensor<string, []>("transpose_105")]; |
| tensor<fp16, [20, 77, 64]> key_states_47_cast = reshape(shape = var_1113, x = transpose_105)[name = tensor<string, []>("key_states_47_cast")]; |
| tensor<int32, [3]> var_1115 = const()[name = tensor<string, []>("op_1115"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_104 = transpose(perm = var_1100_perm_0, x = var_1099_cast)[name = tensor<string, []>("transpose_104")]; |
| tensor<fp16, [20, 77, 64]> value_states_47_cast = reshape(shape = var_1115, x = transpose_104)[name = tensor<string, []>("value_states_47_cast")]; |
| tensor<int32, [3]> var_1118_perm_0 = const()[name = tensor<string, []>("op_1118_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, [20, 64, 77]> transpose_102 = transpose(perm = var_1118_perm_0, x = key_states_47_cast)[name = tensor<string, []>("transpose_102")]; |
| tensor<fp16, [20, 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_23_cast, y = transpose_102)[name = tensor<string, []>("attn_weights_67_cast")]; |
| tensor<int32, [4]> var_1120 = const()[name = tensor<string, []>("op_1120"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1121_cast = reshape(shape = var_1120, x = attn_weights_67_cast)[name = tensor<string, []>("op_1121_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_69_cast = add(x = var_1121_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_69_cast")]; |
| tensor<int32, [3]> var_1126 = const()[name = tensor<string, []>("op_1126"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_181_cast = reshape(shape = var_1126, x = attn_weights_69_cast)[name = tensor<string, []>("input_181_cast")]; |
| tensor<fp16, [20, 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, [20, 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_47_cast)[name = tensor<string, []>("attn_output_67_cast")]; |
| tensor<int32, [4]> var_1131 = const()[name = tensor<string, []>("op_1131"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_69_cast = reshape(shape = var_1131, x = attn_output_67_cast)[name = tensor<string, []>("attn_output_69_cast")]; |
| tensor<int32, [4]> attn_output_71_perm_0 = const()[name = tensor<string, []>("attn_output_71_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1134 = const()[name = tensor<string, []>("op_1134"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_101 = transpose(perm = attn_output_71_perm_0, x = attn_output_69_cast)[name = tensor<string, []>("transpose_101")]; |
| tensor<fp16, [1, 77, 1280]> input_185_cast = reshape(shape = var_1134, x = transpose_101)[name = tensor<string, []>("input_185_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(569452416)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572729280)))]; |
| tensor<fp16, [1, 77, 1280]> 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, x = input_185_cast)[name = tensor<string, []>("hidden_states_69_cast")]; |
| tensor<fp16, [1, 77, 1280]> 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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572731904)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572734528)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_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, [5120, 1280]> text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(572737152)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585844416)))]; |
| tensor<fp16, [1, 77, 5120]> input_191_cast = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16, x = input_189_cast)[name = tensor<string, []>("input_191_cast")]; |
| tensor<string, []> input_193_mode_0 = const()[name = tensor<string, []>("input_193_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_193_cast = gelu(mode = input_193_mode_0, x = input_191_cast)[name = tensor<string, []>("input_193_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585854720)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598961984)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_71_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, x = input_193_cast)[name = tensor<string, []>("hidden_states_71_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_195_cast = add(x = input_187_cast, y = hidden_states_71_cast)[name = tensor<string, []>("input_195_cast")]; |
| tensor<int32, [1]> hidden_states_73_axes_0 = const()[name = tensor<string, []>("hidden_states_73_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598964608)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598967232)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_73_cast = layer_norm(axes = hidden_states_73_axes_0, beta = text_encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16, x = input_195_cast)[name = tensor<string, []>("hidden_states_73_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(598969856)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602246720)))]; |
| tensor<fp16, [1, 77, 1280]> var_1172_cast = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16, x = hidden_states_73_cast)[name = tensor<string, []>("op_1172_cast")]; |
| tensor<fp16, []> var_1173_to_fp16 = const()[name = tensor<string, []>("op_1173_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_77_cast = mul(x = var_1172_cast, y = var_1173_to_fp16)[name = tensor<string, []>("tensor_77_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602249344)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(605526208)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_73_cast = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16, x = hidden_states_73_cast)[name = tensor<string, []>("tensor_73_cast")]; |
| tensor<int32, [4]> var_1178 = const()[name = tensor<string, []>("op_1178"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1179_cast = reshape(shape = var_1178, x = tensor_73_cast)[name = tensor<string, []>("op_1179_cast")]; |
| tensor<int32, [4]> var_1180_perm_0 = const()[name = tensor<string, []>("op_1180_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(605528832)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608805696)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_75_cast = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16, x = hidden_states_73_cast)[name = tensor<string, []>("tensor_75_cast")]; |
| tensor<int32, [4]> var_1185 = const()[name = tensor<string, []>("op_1185"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1186_cast = reshape(shape = var_1185, x = tensor_75_cast)[name = tensor<string, []>("op_1186_cast")]; |
| tensor<int32, [4]> var_1187_perm_0 = const()[name = tensor<string, []>("op_1187_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1194 = const()[name = tensor<string, []>("op_1194"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1195_cast = reshape(shape = var_1194, x = tensor_77_cast)[name = tensor<string, []>("op_1195_cast")]; |
| tensor<int32, [4]> var_1196_perm_0 = const()[name = tensor<string, []>("op_1196_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1198 = const()[name = tensor<string, []>("op_1198"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_98 = transpose(perm = var_1196_perm_0, x = var_1195_cast)[name = tensor<string, []>("transpose_98")]; |
| tensor<fp16, [20, 77, 64]> query_states_25_cast = reshape(shape = var_1198, x = transpose_98)[name = tensor<string, []>("query_states_25_cast")]; |
| tensor<int32, [3]> var_1200 = const()[name = tensor<string, []>("op_1200"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_100 = transpose(perm = var_1180_perm_0, x = var_1179_cast)[name = tensor<string, []>("transpose_100")]; |
| tensor<fp16, [20, 77, 64]> key_states_51_cast = reshape(shape = var_1200, x = transpose_100)[name = tensor<string, []>("key_states_51_cast")]; |
| tensor<int32, [3]> var_1202 = const()[name = tensor<string, []>("op_1202"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_99 = transpose(perm = var_1187_perm_0, x = var_1186_cast)[name = tensor<string, []>("transpose_99")]; |
| tensor<fp16, [20, 77, 64]> value_states_51_cast = reshape(shape = var_1202, x = transpose_99)[name = tensor<string, []>("value_states_51_cast")]; |
| tensor<int32, [3]> var_1205_perm_0 = const()[name = tensor<string, []>("op_1205_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_73_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_73_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_73_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_73_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_97 = transpose(perm = var_1205_perm_0, x = key_states_51_cast)[name = tensor<string, []>("transpose_97")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_73_cast = matmul(transpose_x = attn_weights_73_transpose_x_0, transpose_y = attn_weights_73_transpose_y_0, x = query_states_25_cast, y = transpose_97)[name = tensor<string, []>("attn_weights_73_cast")]; |
| tensor<int32, [4]> var_1207 = const()[name = tensor<string, []>("op_1207"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1208_cast = reshape(shape = var_1207, x = attn_weights_73_cast)[name = tensor<string, []>("op_1208_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_75_cast = add(x = var_1208_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_75_cast")]; |
| tensor<int32, [3]> var_1213 = const()[name = tensor<string, []>("op_1213"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_197_cast = reshape(shape = var_1213, x = attn_weights_75_cast)[name = tensor<string, []>("input_197_cast")]; |
| tensor<fp16, [20, 77, 77]> input_199_cast = softmax(axis = var_5, x = input_197_cast)[name = tensor<string, []>("input_199_cast")]; |
| tensor<bool, []> attn_output_73_transpose_x_0 = const()[name = tensor<string, []>("attn_output_73_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_73_transpose_y_0 = const()[name = tensor<string, []>("attn_output_73_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_73_cast = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = input_199_cast, y = value_states_51_cast)[name = tensor<string, []>("attn_output_73_cast")]; |
| tensor<int32, [4]> var_1218 = const()[name = tensor<string, []>("op_1218"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_75_cast = reshape(shape = var_1218, x = attn_output_73_cast)[name = tensor<string, []>("attn_output_75_cast")]; |
| tensor<int32, [4]> attn_output_77_perm_0 = const()[name = tensor<string, []>("attn_output_77_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1221 = const()[name = tensor<string, []>("op_1221"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_96 = transpose(perm = attn_output_77_perm_0, x = attn_output_75_cast)[name = tensor<string, []>("transpose_96")]; |
| tensor<fp16, [1, 77, 1280]> input_201_cast = reshape(shape = var_1221, x = transpose_96)[name = tensor<string, []>("input_201_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(608808320)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612085184)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_75_cast = linear(bias = text_encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16, x = input_201_cast)[name = tensor<string, []>("hidden_states_75_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_203_cast = add(x = input_195_cast, y = hidden_states_75_cast)[name = tensor<string, []>("input_203_cast")]; |
| tensor<int32, [1]> input_205_axes_0 = const()[name = tensor<string, []>("input_205_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612087808)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612090432)))]; |
| tensor<fp16, [1, 77, 1280]> input_205_cast = layer_norm(axes = input_205_axes_0, beta = text_encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16, x = input_203_cast)[name = tensor<string, []>("input_205_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612093056)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625200320)))]; |
| tensor<fp16, [1, 77, 5120]> input_207_cast = linear(bias = text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16, x = input_205_cast)[name = tensor<string, []>("input_207_cast")]; |
| tensor<string, []> input_209_mode_0 = const()[name = tensor<string, []>("input_209_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_209_cast = gelu(mode = input_209_mode_0, x = input_207_cast)[name = tensor<string, []>("input_209_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(625210624)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638317888)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_77_cast = linear(bias = text_encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16, x = input_209_cast)[name = tensor<string, []>("hidden_states_77_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_211_cast = add(x = input_203_cast, y = hidden_states_77_cast)[name = tensor<string, []>("input_211_cast")]; |
| tensor<int32, [1]> hidden_states_79_axes_0 = const()[name = tensor<string, []>("hidden_states_79_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638320512)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638323136)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_79_cast = layer_norm(axes = hidden_states_79_axes_0, beta = text_encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16, x = input_211_cast)[name = tensor<string, []>("hidden_states_79_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(638325760)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641602624)))]; |
| tensor<fp16, [1, 77, 1280]> var_1259_cast = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16, x = hidden_states_79_cast)[name = tensor<string, []>("op_1259_cast")]; |
| tensor<fp16, []> var_1260_to_fp16 = const()[name = tensor<string, []>("op_1260_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_83_cast = mul(x = var_1259_cast, y = var_1260_to_fp16)[name = tensor<string, []>("tensor_83_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(641605248)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644882112)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_79_cast = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16, x = hidden_states_79_cast)[name = tensor<string, []>("tensor_79_cast")]; |
| tensor<int32, [4]> var_1265 = const()[name = tensor<string, []>("op_1265"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1266_cast = reshape(shape = var_1265, x = tensor_79_cast)[name = tensor<string, []>("op_1266_cast")]; |
| tensor<int32, [4]> var_1267_perm_0 = const()[name = tensor<string, []>("op_1267_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(644884736)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(648161600)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_81_cast = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16, x = hidden_states_79_cast)[name = tensor<string, []>("tensor_81_cast")]; |
| tensor<int32, [4]> var_1272 = const()[name = tensor<string, []>("op_1272"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1273_cast = reshape(shape = var_1272, x = tensor_81_cast)[name = tensor<string, []>("op_1273_cast")]; |
| tensor<int32, [4]> var_1274_perm_0 = const()[name = tensor<string, []>("op_1274_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1281 = const()[name = tensor<string, []>("op_1281"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1282_cast = reshape(shape = var_1281, x = tensor_83_cast)[name = tensor<string, []>("op_1282_cast")]; |
| tensor<int32, [4]> var_1283_perm_0 = const()[name = tensor<string, []>("op_1283_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1285 = const()[name = tensor<string, []>("op_1285"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_93 = transpose(perm = var_1283_perm_0, x = var_1282_cast)[name = tensor<string, []>("transpose_93")]; |
| tensor<fp16, [20, 77, 64]> query_states_27_cast = reshape(shape = var_1285, x = transpose_93)[name = tensor<string, []>("query_states_27_cast")]; |
| tensor<int32, [3]> var_1287 = const()[name = tensor<string, []>("op_1287"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_95 = transpose(perm = var_1267_perm_0, x = var_1266_cast)[name = tensor<string, []>("transpose_95")]; |
| tensor<fp16, [20, 77, 64]> key_states_55_cast = reshape(shape = var_1287, x = transpose_95)[name = tensor<string, []>("key_states_55_cast")]; |
| tensor<int32, [3]> var_1289 = const()[name = tensor<string, []>("op_1289"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_94 = transpose(perm = var_1274_perm_0, x = var_1273_cast)[name = tensor<string, []>("transpose_94")]; |
| tensor<fp16, [20, 77, 64]> value_states_55_cast = reshape(shape = var_1289, x = transpose_94)[name = tensor<string, []>("value_states_55_cast")]; |
| tensor<int32, [3]> var_1292_perm_0 = const()[name = tensor<string, []>("op_1292_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_79_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_79_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_79_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_79_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_92 = transpose(perm = var_1292_perm_0, x = key_states_55_cast)[name = tensor<string, []>("transpose_92")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_79_cast = matmul(transpose_x = attn_weights_79_transpose_x_0, transpose_y = attn_weights_79_transpose_y_0, x = query_states_27_cast, y = transpose_92)[name = tensor<string, []>("attn_weights_79_cast")]; |
| tensor<int32, [4]> var_1294 = const()[name = tensor<string, []>("op_1294"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1295_cast = reshape(shape = var_1294, x = attn_weights_79_cast)[name = tensor<string, []>("op_1295_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_81_cast = add(x = var_1295_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_81_cast")]; |
| tensor<int32, [3]> var_1300 = const()[name = tensor<string, []>("op_1300"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_213_cast = reshape(shape = var_1300, x = attn_weights_81_cast)[name = tensor<string, []>("input_213_cast")]; |
| tensor<fp16, [20, 77, 77]> input_215_cast = softmax(axis = var_5, x = input_213_cast)[name = tensor<string, []>("input_215_cast")]; |
| tensor<bool, []> attn_output_79_transpose_x_0 = const()[name = tensor<string, []>("attn_output_79_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_79_transpose_y_0 = const()[name = tensor<string, []>("attn_output_79_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_79_cast = matmul(transpose_x = attn_output_79_transpose_x_0, transpose_y = attn_output_79_transpose_y_0, x = input_215_cast, y = value_states_55_cast)[name = tensor<string, []>("attn_output_79_cast")]; |
| tensor<int32, [4]> var_1305 = const()[name = tensor<string, []>("op_1305"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_81_cast = reshape(shape = var_1305, x = attn_output_79_cast)[name = tensor<string, []>("attn_output_81_cast")]; |
| tensor<int32, [4]> attn_output_83_perm_0 = const()[name = tensor<string, []>("attn_output_83_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1308 = const()[name = tensor<string, []>("op_1308"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_91 = transpose(perm = attn_output_83_perm_0, x = attn_output_81_cast)[name = tensor<string, []>("transpose_91")]; |
| tensor<fp16, [1, 77, 1280]> input_217_cast = reshape(shape = var_1308, x = transpose_91)[name = tensor<string, []>("input_217_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(648164224)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651441088)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_81_cast = linear(bias = text_encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16, x = input_217_cast)[name = tensor<string, []>("hidden_states_81_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_219_cast = add(x = input_211_cast, y = hidden_states_81_cast)[name = tensor<string, []>("input_219_cast")]; |
| tensor<int32, [1]> input_221_axes_0 = const()[name = tensor<string, []>("input_221_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651443712)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651446336)))]; |
| tensor<fp16, [1, 77, 1280]> input_221_cast = layer_norm(axes = input_221_axes_0, beta = text_encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16, x = input_219_cast)[name = tensor<string, []>("input_221_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651448960)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(664556224)))]; |
| tensor<fp16, [1, 77, 5120]> input_223_cast = linear(bias = text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16, x = input_221_cast)[name = tensor<string, []>("input_223_cast")]; |
| tensor<string, []> input_225_mode_0 = const()[name = tensor<string, []>("input_225_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_225_cast = gelu(mode = input_225_mode_0, x = input_223_cast)[name = tensor<string, []>("input_225_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(664566528)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677673792)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_83_cast = linear(bias = text_encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16, x = input_225_cast)[name = tensor<string, []>("hidden_states_83_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_227_cast = add(x = input_219_cast, y = hidden_states_83_cast)[name = tensor<string, []>("input_227_cast")]; |
| tensor<int32, [1]> hidden_states_85_axes_0 = const()[name = tensor<string, []>("hidden_states_85_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677676416)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677679040)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_85_cast = layer_norm(axes = hidden_states_85_axes_0, beta = text_encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16, x = input_227_cast)[name = tensor<string, []>("hidden_states_85_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(677681664)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(680958528)))]; |
| tensor<fp16, [1, 77, 1280]> var_1346_cast = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16, x = hidden_states_85_cast)[name = tensor<string, []>("op_1346_cast")]; |
| tensor<fp16, []> var_1347_to_fp16 = const()[name = tensor<string, []>("op_1347_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_89_cast = mul(x = var_1346_cast, y = var_1347_to_fp16)[name = tensor<string, []>("tensor_89_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(680961152)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684238016)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_85_cast = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16, x = hidden_states_85_cast)[name = tensor<string, []>("tensor_85_cast")]; |
| tensor<int32, [4]> var_1352 = const()[name = tensor<string, []>("op_1352"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1353_cast = reshape(shape = var_1352, x = tensor_85_cast)[name = tensor<string, []>("op_1353_cast")]; |
| tensor<int32, [4]> var_1354_perm_0 = const()[name = tensor<string, []>("op_1354_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(684240640)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(687517504)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_87_cast = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16, x = hidden_states_85_cast)[name = tensor<string, []>("tensor_87_cast")]; |
| tensor<int32, [4]> var_1359 = const()[name = tensor<string, []>("op_1359"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1360_cast = reshape(shape = var_1359, x = tensor_87_cast)[name = tensor<string, []>("op_1360_cast")]; |
| tensor<int32, [4]> var_1361_perm_0 = const()[name = tensor<string, []>("op_1361_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1368 = const()[name = tensor<string, []>("op_1368"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1369_cast = reshape(shape = var_1368, x = tensor_89_cast)[name = tensor<string, []>("op_1369_cast")]; |
| tensor<int32, [4]> var_1370_perm_0 = const()[name = tensor<string, []>("op_1370_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1372 = const()[name = tensor<string, []>("op_1372"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_88 = transpose(perm = var_1370_perm_0, x = var_1369_cast)[name = tensor<string, []>("transpose_88")]; |
| tensor<fp16, [20, 77, 64]> query_states_29_cast = reshape(shape = var_1372, x = transpose_88)[name = tensor<string, []>("query_states_29_cast")]; |
| tensor<int32, [3]> var_1374 = const()[name = tensor<string, []>("op_1374"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_90 = transpose(perm = var_1354_perm_0, x = var_1353_cast)[name = tensor<string, []>("transpose_90")]; |
| tensor<fp16, [20, 77, 64]> key_states_59_cast = reshape(shape = var_1374, x = transpose_90)[name = tensor<string, []>("key_states_59_cast")]; |
| tensor<int32, [3]> var_1376 = const()[name = tensor<string, []>("op_1376"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_89 = transpose(perm = var_1361_perm_0, x = var_1360_cast)[name = tensor<string, []>("transpose_89")]; |
| tensor<fp16, [20, 77, 64]> value_states_59_cast = reshape(shape = var_1376, x = transpose_89)[name = tensor<string, []>("value_states_59_cast")]; |
| tensor<int32, [3]> var_1379_perm_0 = const()[name = tensor<string, []>("op_1379_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_85_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_85_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_85_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_85_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_87 = transpose(perm = var_1379_perm_0, x = key_states_59_cast)[name = tensor<string, []>("transpose_87")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_85_cast = matmul(transpose_x = attn_weights_85_transpose_x_0, transpose_y = attn_weights_85_transpose_y_0, x = query_states_29_cast, y = transpose_87)[name = tensor<string, []>("attn_weights_85_cast")]; |
| tensor<int32, [4]> var_1381 = const()[name = tensor<string, []>("op_1381"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1382_cast = reshape(shape = var_1381, x = attn_weights_85_cast)[name = tensor<string, []>("op_1382_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_87_cast = add(x = var_1382_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_87_cast")]; |
| tensor<int32, [3]> var_1387 = const()[name = tensor<string, []>("op_1387"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_229_cast = reshape(shape = var_1387, x = attn_weights_87_cast)[name = tensor<string, []>("input_229_cast")]; |
| tensor<fp16, [20, 77, 77]> input_231_cast = softmax(axis = var_5, x = input_229_cast)[name = tensor<string, []>("input_231_cast")]; |
| tensor<bool, []> attn_output_85_transpose_x_0 = const()[name = tensor<string, []>("attn_output_85_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_85_transpose_y_0 = const()[name = tensor<string, []>("attn_output_85_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_85_cast = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = input_231_cast, y = value_states_59_cast)[name = tensor<string, []>("attn_output_85_cast")]; |
| tensor<int32, [4]> var_1392 = const()[name = tensor<string, []>("op_1392"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_87_cast = reshape(shape = var_1392, x = attn_output_85_cast)[name = tensor<string, []>("attn_output_87_cast")]; |
| tensor<int32, [4]> attn_output_89_perm_0 = const()[name = tensor<string, []>("attn_output_89_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1395 = const()[name = tensor<string, []>("op_1395"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_86 = transpose(perm = attn_output_89_perm_0, x = attn_output_87_cast)[name = tensor<string, []>("transpose_86")]; |
| tensor<fp16, [1, 77, 1280]> input_233_cast = reshape(shape = var_1395, x = transpose_86)[name = tensor<string, []>("input_233_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(687520128)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(690796992)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_87_cast = linear(bias = text_encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16, x = input_233_cast)[name = tensor<string, []>("hidden_states_87_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_235_cast = add(x = input_227_cast, y = hidden_states_87_cast)[name = tensor<string, []>("input_235_cast")]; |
| tensor<int32, [1]> input_237_axes_0 = const()[name = tensor<string, []>("input_237_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(690799616)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(690802240)))]; |
| tensor<fp16, [1, 77, 1280]> input_237_cast = layer_norm(axes = input_237_axes_0, beta = text_encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16, x = input_235_cast)[name = tensor<string, []>("input_237_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(690804864)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703912128)))]; |
| tensor<fp16, [1, 77, 5120]> input_239_cast = linear(bias = text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16, x = input_237_cast)[name = tensor<string, []>("input_239_cast")]; |
| tensor<string, []> input_241_mode_0 = const()[name = tensor<string, []>("input_241_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_241_cast = gelu(mode = input_241_mode_0, x = input_239_cast)[name = tensor<string, []>("input_241_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703922432)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717029696)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_89_cast = linear(bias = text_encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16, x = input_241_cast)[name = tensor<string, []>("hidden_states_89_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_243_cast = add(x = input_235_cast, y = hidden_states_89_cast)[name = tensor<string, []>("input_243_cast")]; |
| tensor<int32, [1]> hidden_states_91_axes_0 = const()[name = tensor<string, []>("hidden_states_91_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717032320)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717034944)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_91_cast = layer_norm(axes = hidden_states_91_axes_0, beta = text_encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16, x = input_243_cast)[name = tensor<string, []>("hidden_states_91_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717037568)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(720314432)))]; |
| tensor<fp16, [1, 77, 1280]> var_1433_cast = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16, x = hidden_states_91_cast)[name = tensor<string, []>("op_1433_cast")]; |
| tensor<fp16, []> var_1434_to_fp16 = const()[name = tensor<string, []>("op_1434_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_95_cast = mul(x = var_1433_cast, y = var_1434_to_fp16)[name = tensor<string, []>("tensor_95_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(720317056)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723593920)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_91_cast = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16, x = hidden_states_91_cast)[name = tensor<string, []>("tensor_91_cast")]; |
| tensor<int32, [4]> var_1439 = const()[name = tensor<string, []>("op_1439"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1440_cast = reshape(shape = var_1439, x = tensor_91_cast)[name = tensor<string, []>("op_1440_cast")]; |
| tensor<int32, [4]> var_1441_perm_0 = const()[name = tensor<string, []>("op_1441_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723596544)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(726873408)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_93_cast = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16, x = hidden_states_91_cast)[name = tensor<string, []>("tensor_93_cast")]; |
| tensor<int32, [4]> var_1446 = const()[name = tensor<string, []>("op_1446"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1447_cast = reshape(shape = var_1446, x = tensor_93_cast)[name = tensor<string, []>("op_1447_cast")]; |
| tensor<int32, [4]> var_1448_perm_0 = const()[name = tensor<string, []>("op_1448_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1455 = const()[name = tensor<string, []>("op_1455"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1456_cast = reshape(shape = var_1455, x = tensor_95_cast)[name = tensor<string, []>("op_1456_cast")]; |
| tensor<int32, [4]> var_1457_perm_0 = const()[name = tensor<string, []>("op_1457_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1459 = const()[name = tensor<string, []>("op_1459"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_83 = transpose(perm = var_1457_perm_0, x = var_1456_cast)[name = tensor<string, []>("transpose_83")]; |
| tensor<fp16, [20, 77, 64]> query_states_31_cast = reshape(shape = var_1459, x = transpose_83)[name = tensor<string, []>("query_states_31_cast")]; |
| tensor<int32, [3]> var_1461 = const()[name = tensor<string, []>("op_1461"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_85 = transpose(perm = var_1441_perm_0, x = var_1440_cast)[name = tensor<string, []>("transpose_85")]; |
| tensor<fp16, [20, 77, 64]> key_states_63_cast = reshape(shape = var_1461, x = transpose_85)[name = tensor<string, []>("key_states_63_cast")]; |
| tensor<int32, [3]> var_1463 = const()[name = tensor<string, []>("op_1463"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_84 = transpose(perm = var_1448_perm_0, x = var_1447_cast)[name = tensor<string, []>("transpose_84")]; |
| tensor<fp16, [20, 77, 64]> value_states_63_cast = reshape(shape = var_1463, x = transpose_84)[name = tensor<string, []>("value_states_63_cast")]; |
| tensor<int32, [3]> var_1466_perm_0 = const()[name = tensor<string, []>("op_1466_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_91_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_91_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_91_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_91_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_82 = transpose(perm = var_1466_perm_0, x = key_states_63_cast)[name = tensor<string, []>("transpose_82")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_91_cast = matmul(transpose_x = attn_weights_91_transpose_x_0, transpose_y = attn_weights_91_transpose_y_0, x = query_states_31_cast, y = transpose_82)[name = tensor<string, []>("attn_weights_91_cast")]; |
| tensor<int32, [4]> var_1468 = const()[name = tensor<string, []>("op_1468"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1469_cast = reshape(shape = var_1468, x = attn_weights_91_cast)[name = tensor<string, []>("op_1469_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_93_cast = add(x = var_1469_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_93_cast")]; |
| tensor<int32, [3]> var_1474 = const()[name = tensor<string, []>("op_1474"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_245_cast = reshape(shape = var_1474, x = attn_weights_93_cast)[name = tensor<string, []>("input_245_cast")]; |
| tensor<fp16, [20, 77, 77]> input_247_cast = softmax(axis = var_5, x = input_245_cast)[name = tensor<string, []>("input_247_cast")]; |
| tensor<bool, []> attn_output_91_transpose_x_0 = const()[name = tensor<string, []>("attn_output_91_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_91_transpose_y_0 = const()[name = tensor<string, []>("attn_output_91_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_91_cast = matmul(transpose_x = attn_output_91_transpose_x_0, transpose_y = attn_output_91_transpose_y_0, x = input_247_cast, y = value_states_63_cast)[name = tensor<string, []>("attn_output_91_cast")]; |
| tensor<int32, [4]> var_1479 = const()[name = tensor<string, []>("op_1479"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_93_cast = reshape(shape = var_1479, x = attn_output_91_cast)[name = tensor<string, []>("attn_output_93_cast")]; |
| tensor<int32, [4]> attn_output_95_perm_0 = const()[name = tensor<string, []>("attn_output_95_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1482 = const()[name = tensor<string, []>("op_1482"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_81 = transpose(perm = attn_output_95_perm_0, x = attn_output_93_cast)[name = tensor<string, []>("transpose_81")]; |
| tensor<fp16, [1, 77, 1280]> input_249_cast = reshape(shape = var_1482, x = transpose_81)[name = tensor<string, []>("input_249_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(726876032)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(730152896)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_93_cast = linear(bias = text_encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16, x = input_249_cast)[name = tensor<string, []>("hidden_states_93_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_251_cast = add(x = input_243_cast, y = hidden_states_93_cast)[name = tensor<string, []>("input_251_cast")]; |
| tensor<int32, [1]> input_253_axes_0 = const()[name = tensor<string, []>("input_253_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(730155520)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(730158144)))]; |
| tensor<fp16, [1, 77, 1280]> input_253_cast = layer_norm(axes = input_253_axes_0, beta = text_encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16, x = input_251_cast)[name = tensor<string, []>("input_253_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(730160768)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(743268032)))]; |
| tensor<fp16, [1, 77, 5120]> input_255_cast = linear(bias = text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16, x = input_253_cast)[name = tensor<string, []>("input_255_cast")]; |
| tensor<string, []> input_257_mode_0 = const()[name = tensor<string, []>("input_257_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_257_cast = gelu(mode = input_257_mode_0, x = input_255_cast)[name = tensor<string, []>("input_257_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(743278336)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(756385600)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_95_cast = linear(bias = text_encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16, x = input_257_cast)[name = tensor<string, []>("hidden_states_95_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_259_cast = add(x = input_251_cast, y = hidden_states_95_cast)[name = tensor<string, []>("input_259_cast")]; |
| tensor<int32, [1]> hidden_states_97_axes_0 = const()[name = tensor<string, []>("hidden_states_97_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(756388224)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(756390848)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_97_cast = layer_norm(axes = hidden_states_97_axes_0, beta = text_encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16, x = input_259_cast)[name = tensor<string, []>("hidden_states_97_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(756393472)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759670336)))]; |
| tensor<fp16, [1, 77, 1280]> var_1520_cast = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16, x = hidden_states_97_cast)[name = tensor<string, []>("op_1520_cast")]; |
| tensor<fp16, []> var_1521_to_fp16 = const()[name = tensor<string, []>("op_1521_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_101_cast = mul(x = var_1520_cast, y = var_1521_to_fp16)[name = tensor<string, []>("tensor_101_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759672960)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762949824)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_97_cast = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16, x = hidden_states_97_cast)[name = tensor<string, []>("tensor_97_cast")]; |
| tensor<int32, [4]> var_1526 = const()[name = tensor<string, []>("op_1526"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1527_cast = reshape(shape = var_1526, x = tensor_97_cast)[name = tensor<string, []>("op_1527_cast")]; |
| tensor<int32, [4]> var_1528_perm_0 = const()[name = tensor<string, []>("op_1528_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762952448)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766229312)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_99_cast = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16, x = hidden_states_97_cast)[name = tensor<string, []>("tensor_99_cast")]; |
| tensor<int32, [4]> var_1533 = const()[name = tensor<string, []>("op_1533"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1534_cast = reshape(shape = var_1533, x = tensor_99_cast)[name = tensor<string, []>("op_1534_cast")]; |
| tensor<int32, [4]> var_1535_perm_0 = const()[name = tensor<string, []>("op_1535_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1542 = const()[name = tensor<string, []>("op_1542"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1543_cast = reshape(shape = var_1542, x = tensor_101_cast)[name = tensor<string, []>("op_1543_cast")]; |
| tensor<int32, [4]> var_1544_perm_0 = const()[name = tensor<string, []>("op_1544_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1546 = const()[name = tensor<string, []>("op_1546"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_78 = transpose(perm = var_1544_perm_0, x = var_1543_cast)[name = tensor<string, []>("transpose_78")]; |
| tensor<fp16, [20, 77, 64]> query_states_33_cast = reshape(shape = var_1546, x = transpose_78)[name = tensor<string, []>("query_states_33_cast")]; |
| tensor<int32, [3]> var_1548 = const()[name = tensor<string, []>("op_1548"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_80 = transpose(perm = var_1528_perm_0, x = var_1527_cast)[name = tensor<string, []>("transpose_80")]; |
| tensor<fp16, [20, 77, 64]> key_states_67_cast = reshape(shape = var_1548, x = transpose_80)[name = tensor<string, []>("key_states_67_cast")]; |
| tensor<int32, [3]> var_1550 = const()[name = tensor<string, []>("op_1550"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_79 = transpose(perm = var_1535_perm_0, x = var_1534_cast)[name = tensor<string, []>("transpose_79")]; |
| tensor<fp16, [20, 77, 64]> value_states_67_cast = reshape(shape = var_1550, x = transpose_79)[name = tensor<string, []>("value_states_67_cast")]; |
| tensor<int32, [3]> var_1553_perm_0 = const()[name = tensor<string, []>("op_1553_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_97_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_97_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_97_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_97_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_77 = transpose(perm = var_1553_perm_0, x = key_states_67_cast)[name = tensor<string, []>("transpose_77")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_97_cast = matmul(transpose_x = attn_weights_97_transpose_x_0, transpose_y = attn_weights_97_transpose_y_0, x = query_states_33_cast, y = transpose_77)[name = tensor<string, []>("attn_weights_97_cast")]; |
| tensor<int32, [4]> var_1555 = const()[name = tensor<string, []>("op_1555"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1556_cast = reshape(shape = var_1555, x = attn_weights_97_cast)[name = tensor<string, []>("op_1556_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_99_cast = add(x = var_1556_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_99_cast")]; |
| tensor<int32, [3]> var_1561 = const()[name = tensor<string, []>("op_1561"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_261_cast = reshape(shape = var_1561, x = attn_weights_99_cast)[name = tensor<string, []>("input_261_cast")]; |
| tensor<fp16, [20, 77, 77]> input_263_cast = softmax(axis = var_5, x = input_261_cast)[name = tensor<string, []>("input_263_cast")]; |
| tensor<bool, []> attn_output_97_transpose_x_0 = const()[name = tensor<string, []>("attn_output_97_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_97_transpose_y_0 = const()[name = tensor<string, []>("attn_output_97_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_97_cast = matmul(transpose_x = attn_output_97_transpose_x_0, transpose_y = attn_output_97_transpose_y_0, x = input_263_cast, y = value_states_67_cast)[name = tensor<string, []>("attn_output_97_cast")]; |
| tensor<int32, [4]> var_1566 = const()[name = tensor<string, []>("op_1566"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_99_cast = reshape(shape = var_1566, x = attn_output_97_cast)[name = tensor<string, []>("attn_output_99_cast")]; |
| tensor<int32, [4]> attn_output_101_perm_0 = const()[name = tensor<string, []>("attn_output_101_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1569 = const()[name = tensor<string, []>("op_1569"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_76 = transpose(perm = attn_output_101_perm_0, x = attn_output_99_cast)[name = tensor<string, []>("transpose_76")]; |
| tensor<fp16, [1, 77, 1280]> input_265_cast = reshape(shape = var_1569, x = transpose_76)[name = tensor<string, []>("input_265_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766231936)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(769508800)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_99_cast = linear(bias = text_encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16, x = input_265_cast)[name = tensor<string, []>("hidden_states_99_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_267_cast = add(x = input_259_cast, y = hidden_states_99_cast)[name = tensor<string, []>("input_267_cast")]; |
| tensor<int32, [1]> input_269_axes_0 = const()[name = tensor<string, []>("input_269_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(769511424)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(769514048)))]; |
| tensor<fp16, [1, 77, 1280]> input_269_cast = layer_norm(axes = input_269_axes_0, beta = text_encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16, x = input_267_cast)[name = tensor<string, []>("input_269_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(769516672)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782623936)))]; |
| tensor<fp16, [1, 77, 5120]> input_271_cast = linear(bias = text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16, x = input_269_cast)[name = tensor<string, []>("input_271_cast")]; |
| tensor<string, []> input_273_mode_0 = const()[name = tensor<string, []>("input_273_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_273_cast = gelu(mode = input_273_mode_0, x = input_271_cast)[name = tensor<string, []>("input_273_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782634240)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795741504)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_101_cast = linear(bias = text_encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16, x = input_273_cast)[name = tensor<string, []>("hidden_states_101_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_275_cast = add(x = input_267_cast, y = hidden_states_101_cast)[name = tensor<string, []>("input_275_cast")]; |
| tensor<int32, [1]> hidden_states_103_axes_0 = const()[name = tensor<string, []>("hidden_states_103_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795744128)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795746752)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_103_cast = layer_norm(axes = hidden_states_103_axes_0, beta = text_encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16, x = input_275_cast)[name = tensor<string, []>("hidden_states_103_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795749376)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(799026240)))]; |
| tensor<fp16, [1, 77, 1280]> var_1607_cast = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16, x = hidden_states_103_cast)[name = tensor<string, []>("op_1607_cast")]; |
| tensor<fp16, []> var_1608_to_fp16 = const()[name = tensor<string, []>("op_1608_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_107_cast = mul(x = var_1607_cast, y = var_1608_to_fp16)[name = tensor<string, []>("tensor_107_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(799028864)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(802305728)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_103_cast = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16, x = hidden_states_103_cast)[name = tensor<string, []>("tensor_103_cast")]; |
| tensor<int32, [4]> var_1613 = const()[name = tensor<string, []>("op_1613"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1614_cast = reshape(shape = var_1613, x = tensor_103_cast)[name = tensor<string, []>("op_1614_cast")]; |
| tensor<int32, [4]> var_1615_perm_0 = const()[name = tensor<string, []>("op_1615_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(802308352)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805585216)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_105_cast = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16, x = hidden_states_103_cast)[name = tensor<string, []>("tensor_105_cast")]; |
| tensor<int32, [4]> var_1620 = const()[name = tensor<string, []>("op_1620"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1621_cast = reshape(shape = var_1620, x = tensor_105_cast)[name = tensor<string, []>("op_1621_cast")]; |
| tensor<int32, [4]> var_1622_perm_0 = const()[name = tensor<string, []>("op_1622_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1629 = const()[name = tensor<string, []>("op_1629"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1630_cast = reshape(shape = var_1629, x = tensor_107_cast)[name = tensor<string, []>("op_1630_cast")]; |
| tensor<int32, [4]> var_1631_perm_0 = const()[name = tensor<string, []>("op_1631_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1633 = const()[name = tensor<string, []>("op_1633"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_73 = transpose(perm = var_1631_perm_0, x = var_1630_cast)[name = tensor<string, []>("transpose_73")]; |
| tensor<fp16, [20, 77, 64]> query_states_35_cast = reshape(shape = var_1633, x = transpose_73)[name = tensor<string, []>("query_states_35_cast")]; |
| tensor<int32, [3]> var_1635 = const()[name = tensor<string, []>("op_1635"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_75 = transpose(perm = var_1615_perm_0, x = var_1614_cast)[name = tensor<string, []>("transpose_75")]; |
| tensor<fp16, [20, 77, 64]> key_states_71_cast = reshape(shape = var_1635, x = transpose_75)[name = tensor<string, []>("key_states_71_cast")]; |
| tensor<int32, [3]> var_1637 = const()[name = tensor<string, []>("op_1637"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_74 = transpose(perm = var_1622_perm_0, x = var_1621_cast)[name = tensor<string, []>("transpose_74")]; |
| tensor<fp16, [20, 77, 64]> value_states_71_cast = reshape(shape = var_1637, x = transpose_74)[name = tensor<string, []>("value_states_71_cast")]; |
| tensor<int32, [3]> var_1640_perm_0 = const()[name = tensor<string, []>("op_1640_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_103_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_103_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_103_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_103_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_72 = transpose(perm = var_1640_perm_0, x = key_states_71_cast)[name = tensor<string, []>("transpose_72")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_103_cast = matmul(transpose_x = attn_weights_103_transpose_x_0, transpose_y = attn_weights_103_transpose_y_0, x = query_states_35_cast, y = transpose_72)[name = tensor<string, []>("attn_weights_103_cast")]; |
| tensor<int32, [4]> var_1642 = const()[name = tensor<string, []>("op_1642"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1643_cast = reshape(shape = var_1642, x = attn_weights_103_cast)[name = tensor<string, []>("op_1643_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_105_cast = add(x = var_1643_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_105_cast")]; |
| tensor<int32, [3]> var_1648 = const()[name = tensor<string, []>("op_1648"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_277_cast = reshape(shape = var_1648, x = attn_weights_105_cast)[name = tensor<string, []>("input_277_cast")]; |
| tensor<fp16, [20, 77, 77]> input_279_cast = softmax(axis = var_5, x = input_277_cast)[name = tensor<string, []>("input_279_cast")]; |
| tensor<bool, []> attn_output_103_transpose_x_0 = const()[name = tensor<string, []>("attn_output_103_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_103_transpose_y_0 = const()[name = tensor<string, []>("attn_output_103_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_103_cast = matmul(transpose_x = attn_output_103_transpose_x_0, transpose_y = attn_output_103_transpose_y_0, x = input_279_cast, y = value_states_71_cast)[name = tensor<string, []>("attn_output_103_cast")]; |
| tensor<int32, [4]> var_1653 = const()[name = tensor<string, []>("op_1653"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_105_cast = reshape(shape = var_1653, x = attn_output_103_cast)[name = tensor<string, []>("attn_output_105_cast")]; |
| tensor<int32, [4]> attn_output_107_perm_0 = const()[name = tensor<string, []>("attn_output_107_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1656 = const()[name = tensor<string, []>("op_1656"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_71 = transpose(perm = attn_output_107_perm_0, x = attn_output_105_cast)[name = tensor<string, []>("transpose_71")]; |
| tensor<fp16, [1, 77, 1280]> input_281_cast = reshape(shape = var_1656, x = transpose_71)[name = tensor<string, []>("input_281_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(805587840)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808864704)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_105_cast = linear(bias = text_encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16, x = input_281_cast)[name = tensor<string, []>("hidden_states_105_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_283_cast = add(x = input_275_cast, y = hidden_states_105_cast)[name = tensor<string, []>("input_283_cast")]; |
| tensor<int32, [1]> input_285_axes_0 = const()[name = tensor<string, []>("input_285_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808867328)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808869952)))]; |
| tensor<fp16, [1, 77, 1280]> input_285_cast = layer_norm(axes = input_285_axes_0, beta = text_encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16, x = input_283_cast)[name = tensor<string, []>("input_285_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808872576)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(821979840)))]; |
| tensor<fp16, [1, 77, 5120]> input_287_cast = linear(bias = text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16, x = input_285_cast)[name = tensor<string, []>("input_287_cast")]; |
| tensor<string, []> input_289_mode_0 = const()[name = tensor<string, []>("input_289_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_289_cast = gelu(mode = input_289_mode_0, x = input_287_cast)[name = tensor<string, []>("input_289_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(821990144)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835097408)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_107_cast = linear(bias = text_encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16, x = input_289_cast)[name = tensor<string, []>("hidden_states_107_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_291_cast = add(x = input_283_cast, y = hidden_states_107_cast)[name = tensor<string, []>("input_291_cast")]; |
| tensor<int32, [1]> hidden_states_109_axes_0 = const()[name = tensor<string, []>("hidden_states_109_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835100032)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835102656)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_109_cast = layer_norm(axes = hidden_states_109_axes_0, beta = text_encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16, x = input_291_cast)[name = tensor<string, []>("hidden_states_109_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(835105280)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838382144)))]; |
| tensor<fp16, [1, 77, 1280]> var_1694_cast = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16, x = hidden_states_109_cast)[name = tensor<string, []>("op_1694_cast")]; |
| tensor<fp16, []> var_1695_to_fp16 = const()[name = tensor<string, []>("op_1695_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_113_cast = mul(x = var_1694_cast, y = var_1695_to_fp16)[name = tensor<string, []>("tensor_113_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838384768)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841661632)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_109_cast = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16, x = hidden_states_109_cast)[name = tensor<string, []>("tensor_109_cast")]; |
| tensor<int32, [4]> var_1700 = const()[name = tensor<string, []>("op_1700"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1701_cast = reshape(shape = var_1700, x = tensor_109_cast)[name = tensor<string, []>("op_1701_cast")]; |
| tensor<int32, [4]> var_1702_perm_0 = const()[name = tensor<string, []>("op_1702_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(841664256)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844941120)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_111_cast = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16, x = hidden_states_109_cast)[name = tensor<string, []>("tensor_111_cast")]; |
| tensor<int32, [4]> var_1707 = const()[name = tensor<string, []>("op_1707"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1708_cast = reshape(shape = var_1707, x = tensor_111_cast)[name = tensor<string, []>("op_1708_cast")]; |
| tensor<int32, [4]> var_1709_perm_0 = const()[name = tensor<string, []>("op_1709_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1716 = const()[name = tensor<string, []>("op_1716"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1717_cast = reshape(shape = var_1716, x = tensor_113_cast)[name = tensor<string, []>("op_1717_cast")]; |
| tensor<int32, [4]> var_1718_perm_0 = const()[name = tensor<string, []>("op_1718_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1720 = const()[name = tensor<string, []>("op_1720"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_68 = transpose(perm = var_1718_perm_0, x = var_1717_cast)[name = tensor<string, []>("transpose_68")]; |
| tensor<fp16, [20, 77, 64]> query_states_37_cast = reshape(shape = var_1720, x = transpose_68)[name = tensor<string, []>("query_states_37_cast")]; |
| tensor<int32, [3]> var_1722 = const()[name = tensor<string, []>("op_1722"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_70 = transpose(perm = var_1702_perm_0, x = var_1701_cast)[name = tensor<string, []>("transpose_70")]; |
| tensor<fp16, [20, 77, 64]> key_states_75_cast = reshape(shape = var_1722, x = transpose_70)[name = tensor<string, []>("key_states_75_cast")]; |
| tensor<int32, [3]> var_1724 = const()[name = tensor<string, []>("op_1724"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_69 = transpose(perm = var_1709_perm_0, x = var_1708_cast)[name = tensor<string, []>("transpose_69")]; |
| tensor<fp16, [20, 77, 64]> value_states_75_cast = reshape(shape = var_1724, x = transpose_69)[name = tensor<string, []>("value_states_75_cast")]; |
| tensor<int32, [3]> var_1727_perm_0 = const()[name = tensor<string, []>("op_1727_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_109_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_109_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_109_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_109_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_67 = transpose(perm = var_1727_perm_0, x = key_states_75_cast)[name = tensor<string, []>("transpose_67")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_109_cast = matmul(transpose_x = attn_weights_109_transpose_x_0, transpose_y = attn_weights_109_transpose_y_0, x = query_states_37_cast, y = transpose_67)[name = tensor<string, []>("attn_weights_109_cast")]; |
| tensor<int32, [4]> var_1729 = const()[name = tensor<string, []>("op_1729"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1730_cast = reshape(shape = var_1729, x = attn_weights_109_cast)[name = tensor<string, []>("op_1730_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_111_cast = add(x = var_1730_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_111_cast")]; |
| tensor<int32, [3]> var_1735 = const()[name = tensor<string, []>("op_1735"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_293_cast = reshape(shape = var_1735, x = attn_weights_111_cast)[name = tensor<string, []>("input_293_cast")]; |
| tensor<fp16, [20, 77, 77]> input_295_cast = softmax(axis = var_5, x = input_293_cast)[name = tensor<string, []>("input_295_cast")]; |
| tensor<bool, []> attn_output_109_transpose_x_0 = const()[name = tensor<string, []>("attn_output_109_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_109_transpose_y_0 = const()[name = tensor<string, []>("attn_output_109_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_109_cast = matmul(transpose_x = attn_output_109_transpose_x_0, transpose_y = attn_output_109_transpose_y_0, x = input_295_cast, y = value_states_75_cast)[name = tensor<string, []>("attn_output_109_cast")]; |
| tensor<int32, [4]> var_1740 = const()[name = tensor<string, []>("op_1740"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_111_cast = reshape(shape = var_1740, x = attn_output_109_cast)[name = tensor<string, []>("attn_output_111_cast")]; |
| tensor<int32, [4]> attn_output_113_perm_0 = const()[name = tensor<string, []>("attn_output_113_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1743 = const()[name = tensor<string, []>("op_1743"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_66 = transpose(perm = attn_output_113_perm_0, x = attn_output_111_cast)[name = tensor<string, []>("transpose_66")]; |
| tensor<fp16, [1, 77, 1280]> input_297_cast = reshape(shape = var_1743, x = transpose_66)[name = tensor<string, []>("input_297_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844943744)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(848220608)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_111_cast = linear(bias = text_encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16, x = input_297_cast)[name = tensor<string, []>("hidden_states_111_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_299_cast = add(x = input_291_cast, y = hidden_states_111_cast)[name = tensor<string, []>("input_299_cast")]; |
| tensor<int32, [1]> input_301_axes_0 = const()[name = tensor<string, []>("input_301_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(848223232)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(848225856)))]; |
| tensor<fp16, [1, 77, 1280]> input_301_cast = layer_norm(axes = input_301_axes_0, beta = text_encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16, x = input_299_cast)[name = tensor<string, []>("input_301_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(848228480)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(861335744)))]; |
| tensor<fp16, [1, 77, 5120]> input_303_cast = linear(bias = text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16, x = input_301_cast)[name = tensor<string, []>("input_303_cast")]; |
| tensor<string, []> input_305_mode_0 = const()[name = tensor<string, []>("input_305_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_305_cast = gelu(mode = input_305_mode_0, x = input_303_cast)[name = tensor<string, []>("input_305_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(861346048)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(874453312)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_113_cast = linear(bias = text_encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16, x = input_305_cast)[name = tensor<string, []>("hidden_states_113_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_307_cast = add(x = input_299_cast, y = hidden_states_113_cast)[name = tensor<string, []>("input_307_cast")]; |
| tensor<int32, [1]> hidden_states_115_axes_0 = const()[name = tensor<string, []>("hidden_states_115_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(874455936)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(874458560)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_115_cast = layer_norm(axes = hidden_states_115_axes_0, beta = text_encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16, x = input_307_cast)[name = tensor<string, []>("hidden_states_115_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(874461184)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(877738048)))]; |
| tensor<fp16, [1, 77, 1280]> var_1781_cast = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16, x = hidden_states_115_cast)[name = tensor<string, []>("op_1781_cast")]; |
| tensor<fp16, []> var_1782_to_fp16 = const()[name = tensor<string, []>("op_1782_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_119_cast = mul(x = var_1781_cast, y = var_1782_to_fp16)[name = tensor<string, []>("tensor_119_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(877740672)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(881017536)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_115_cast = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16, x = hidden_states_115_cast)[name = tensor<string, []>("tensor_115_cast")]; |
| tensor<int32, [4]> var_1787 = const()[name = tensor<string, []>("op_1787"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1788_cast = reshape(shape = var_1787, x = tensor_115_cast)[name = tensor<string, []>("op_1788_cast")]; |
| tensor<int32, [4]> var_1789_perm_0 = const()[name = tensor<string, []>("op_1789_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(881020160)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884297024)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_117_cast = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16, x = hidden_states_115_cast)[name = tensor<string, []>("tensor_117_cast")]; |
| tensor<int32, [4]> var_1794 = const()[name = tensor<string, []>("op_1794"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1795_cast = reshape(shape = var_1794, x = tensor_117_cast)[name = tensor<string, []>("op_1795_cast")]; |
| tensor<int32, [4]> var_1796_perm_0 = const()[name = tensor<string, []>("op_1796_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1803 = const()[name = tensor<string, []>("op_1803"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1804_cast = reshape(shape = var_1803, x = tensor_119_cast)[name = tensor<string, []>("op_1804_cast")]; |
| tensor<int32, [4]> var_1805_perm_0 = const()[name = tensor<string, []>("op_1805_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1807 = const()[name = tensor<string, []>("op_1807"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_63 = transpose(perm = var_1805_perm_0, x = var_1804_cast)[name = tensor<string, []>("transpose_63")]; |
| tensor<fp16, [20, 77, 64]> query_states_39_cast = reshape(shape = var_1807, x = transpose_63)[name = tensor<string, []>("query_states_39_cast")]; |
| tensor<int32, [3]> var_1809 = const()[name = tensor<string, []>("op_1809"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_65 = transpose(perm = var_1789_perm_0, x = var_1788_cast)[name = tensor<string, []>("transpose_65")]; |
| tensor<fp16, [20, 77, 64]> key_states_79_cast = reshape(shape = var_1809, x = transpose_65)[name = tensor<string, []>("key_states_79_cast")]; |
| tensor<int32, [3]> var_1811 = const()[name = tensor<string, []>("op_1811"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_64 = transpose(perm = var_1796_perm_0, x = var_1795_cast)[name = tensor<string, []>("transpose_64")]; |
| tensor<fp16, [20, 77, 64]> value_states_79_cast = reshape(shape = var_1811, x = transpose_64)[name = tensor<string, []>("value_states_79_cast")]; |
| tensor<int32, [3]> var_1814_perm_0 = const()[name = tensor<string, []>("op_1814_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_115_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_115_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_115_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_115_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_62 = transpose(perm = var_1814_perm_0, x = key_states_79_cast)[name = tensor<string, []>("transpose_62")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_115_cast = matmul(transpose_x = attn_weights_115_transpose_x_0, transpose_y = attn_weights_115_transpose_y_0, x = query_states_39_cast, y = transpose_62)[name = tensor<string, []>("attn_weights_115_cast")]; |
| tensor<int32, [4]> var_1816 = const()[name = tensor<string, []>("op_1816"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1817_cast = reshape(shape = var_1816, x = attn_weights_115_cast)[name = tensor<string, []>("op_1817_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_117_cast = add(x = var_1817_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_117_cast")]; |
| tensor<int32, [3]> var_1822 = const()[name = tensor<string, []>("op_1822"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_309_cast = reshape(shape = var_1822, x = attn_weights_117_cast)[name = tensor<string, []>("input_309_cast")]; |
| tensor<fp16, [20, 77, 77]> input_311_cast = softmax(axis = var_5, x = input_309_cast)[name = tensor<string, []>("input_311_cast")]; |
| tensor<bool, []> attn_output_115_transpose_x_0 = const()[name = tensor<string, []>("attn_output_115_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_115_transpose_y_0 = const()[name = tensor<string, []>("attn_output_115_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_115_cast = matmul(transpose_x = attn_output_115_transpose_x_0, transpose_y = attn_output_115_transpose_y_0, x = input_311_cast, y = value_states_79_cast)[name = tensor<string, []>("attn_output_115_cast")]; |
| tensor<int32, [4]> var_1827 = const()[name = tensor<string, []>("op_1827"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_117_cast = reshape(shape = var_1827, x = attn_output_115_cast)[name = tensor<string, []>("attn_output_117_cast")]; |
| tensor<int32, [4]> attn_output_119_perm_0 = const()[name = tensor<string, []>("attn_output_119_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1830 = const()[name = tensor<string, []>("op_1830"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_61 = transpose(perm = attn_output_119_perm_0, x = attn_output_117_cast)[name = tensor<string, []>("transpose_61")]; |
| tensor<fp16, [1, 77, 1280]> input_313_cast = reshape(shape = var_1830, x = transpose_61)[name = tensor<string, []>("input_313_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(884299648)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(887576512)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_117_cast = linear(bias = text_encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16, x = input_313_cast)[name = tensor<string, []>("hidden_states_117_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_315_cast = add(x = input_307_cast, y = hidden_states_117_cast)[name = tensor<string, []>("input_315_cast")]; |
| tensor<int32, [1]> input_317_axes_0 = const()[name = tensor<string, []>("input_317_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(887579136)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(887581760)))]; |
| tensor<fp16, [1, 77, 1280]> input_317_cast = layer_norm(axes = input_317_axes_0, beta = text_encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16, x = input_315_cast)[name = tensor<string, []>("input_317_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(887584384)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900691648)))]; |
| tensor<fp16, [1, 77, 5120]> input_319_cast = linear(bias = text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16, x = input_317_cast)[name = tensor<string, []>("input_319_cast")]; |
| tensor<string, []> input_321_mode_0 = const()[name = tensor<string, []>("input_321_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_321_cast = gelu(mode = input_321_mode_0, x = input_319_cast)[name = tensor<string, []>("input_321_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(900701952)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913809216)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_119_cast = linear(bias = text_encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16, x = input_321_cast)[name = tensor<string, []>("hidden_states_119_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_323_cast = add(x = input_315_cast, y = hidden_states_119_cast)[name = tensor<string, []>("input_323_cast")]; |
| tensor<int32, [1]> hidden_states_121_axes_0 = const()[name = tensor<string, []>("hidden_states_121_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913811840)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913814464)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_121_cast = layer_norm(axes = hidden_states_121_axes_0, beta = text_encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16, x = input_323_cast)[name = tensor<string, []>("hidden_states_121_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(913817088)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917093952)))]; |
| tensor<fp16, [1, 77, 1280]> var_1868_cast = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16, x = hidden_states_121_cast)[name = tensor<string, []>("op_1868_cast")]; |
| tensor<fp16, []> var_1869_to_fp16 = const()[name = tensor<string, []>("op_1869_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_125_cast = mul(x = var_1868_cast, y = var_1869_to_fp16)[name = tensor<string, []>("tensor_125_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(917096576)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920373440)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_121_cast = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16, x = hidden_states_121_cast)[name = tensor<string, []>("tensor_121_cast")]; |
| tensor<int32, [4]> var_1874 = const()[name = tensor<string, []>("op_1874"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1875_cast = reshape(shape = var_1874, x = tensor_121_cast)[name = tensor<string, []>("op_1875_cast")]; |
| tensor<int32, [4]> var_1876_perm_0 = const()[name = tensor<string, []>("op_1876_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(920376064)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923652928)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_123_cast = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16, x = hidden_states_121_cast)[name = tensor<string, []>("tensor_123_cast")]; |
| tensor<int32, [4]> var_1881 = const()[name = tensor<string, []>("op_1881"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1882_cast = reshape(shape = var_1881, x = tensor_123_cast)[name = tensor<string, []>("op_1882_cast")]; |
| tensor<int32, [4]> var_1883_perm_0 = const()[name = tensor<string, []>("op_1883_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1890 = const()[name = tensor<string, []>("op_1890"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1891_cast = reshape(shape = var_1890, x = tensor_125_cast)[name = tensor<string, []>("op_1891_cast")]; |
| tensor<int32, [4]> var_1892_perm_0 = const()[name = tensor<string, []>("op_1892_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1894 = const()[name = tensor<string, []>("op_1894"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_58 = transpose(perm = var_1892_perm_0, x = var_1891_cast)[name = tensor<string, []>("transpose_58")]; |
| tensor<fp16, [20, 77, 64]> query_states_41_cast = reshape(shape = var_1894, x = transpose_58)[name = tensor<string, []>("query_states_41_cast")]; |
| tensor<int32, [3]> var_1896 = const()[name = tensor<string, []>("op_1896"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_60 = transpose(perm = var_1876_perm_0, x = var_1875_cast)[name = tensor<string, []>("transpose_60")]; |
| tensor<fp16, [20, 77, 64]> key_states_83_cast = reshape(shape = var_1896, x = transpose_60)[name = tensor<string, []>("key_states_83_cast")]; |
| tensor<int32, [3]> var_1898 = const()[name = tensor<string, []>("op_1898"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_59 = transpose(perm = var_1883_perm_0, x = var_1882_cast)[name = tensor<string, []>("transpose_59")]; |
| tensor<fp16, [20, 77, 64]> value_states_83_cast = reshape(shape = var_1898, x = transpose_59)[name = tensor<string, []>("value_states_83_cast")]; |
| tensor<int32, [3]> var_1901_perm_0 = const()[name = tensor<string, []>("op_1901_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_121_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_121_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_121_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_121_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_57 = transpose(perm = var_1901_perm_0, x = key_states_83_cast)[name = tensor<string, []>("transpose_57")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_121_cast = matmul(transpose_x = attn_weights_121_transpose_x_0, transpose_y = attn_weights_121_transpose_y_0, x = query_states_41_cast, y = transpose_57)[name = tensor<string, []>("attn_weights_121_cast")]; |
| tensor<int32, [4]> var_1903 = const()[name = tensor<string, []>("op_1903"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1904_cast = reshape(shape = var_1903, x = attn_weights_121_cast)[name = tensor<string, []>("op_1904_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_123_cast = add(x = var_1904_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_123_cast")]; |
| tensor<int32, [3]> var_1909 = const()[name = tensor<string, []>("op_1909"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_325_cast = reshape(shape = var_1909, x = attn_weights_123_cast)[name = tensor<string, []>("input_325_cast")]; |
| tensor<fp16, [20, 77, 77]> input_327_cast = softmax(axis = var_5, x = input_325_cast)[name = tensor<string, []>("input_327_cast")]; |
| tensor<bool, []> attn_output_121_transpose_x_0 = const()[name = tensor<string, []>("attn_output_121_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_121_transpose_y_0 = const()[name = tensor<string, []>("attn_output_121_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_121_cast = matmul(transpose_x = attn_output_121_transpose_x_0, transpose_y = attn_output_121_transpose_y_0, x = input_327_cast, y = value_states_83_cast)[name = tensor<string, []>("attn_output_121_cast")]; |
| tensor<int32, [4]> var_1914 = const()[name = tensor<string, []>("op_1914"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_123_cast = reshape(shape = var_1914, x = attn_output_121_cast)[name = tensor<string, []>("attn_output_123_cast")]; |
| tensor<int32, [4]> attn_output_125_perm_0 = const()[name = tensor<string, []>("attn_output_125_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1917 = const()[name = tensor<string, []>("op_1917"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_56 = transpose(perm = attn_output_125_perm_0, x = attn_output_123_cast)[name = tensor<string, []>("transpose_56")]; |
| tensor<fp16, [1, 77, 1280]> input_329_cast = reshape(shape = var_1917, x = transpose_56)[name = tensor<string, []>("input_329_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(923655552)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926932416)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_123_cast = linear(bias = text_encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16, x = input_329_cast)[name = tensor<string, []>("hidden_states_123_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_331_cast = add(x = input_323_cast, y = hidden_states_123_cast)[name = tensor<string, []>("input_331_cast")]; |
| tensor<int32, [1]> input_333_axes_0 = const()[name = tensor<string, []>("input_333_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926935040)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926937664)))]; |
| tensor<fp16, [1, 77, 1280]> input_333_cast = layer_norm(axes = input_333_axes_0, beta = text_encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16, x = input_331_cast)[name = tensor<string, []>("input_333_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926940288)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(940047552)))]; |
| tensor<fp16, [1, 77, 5120]> input_335_cast = linear(bias = text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16, x = input_333_cast)[name = tensor<string, []>("input_335_cast")]; |
| tensor<string, []> input_337_mode_0 = const()[name = tensor<string, []>("input_337_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_337_cast = gelu(mode = input_337_mode_0, x = input_335_cast)[name = tensor<string, []>("input_337_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(940057856)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953165120)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_125_cast = linear(bias = text_encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16, x = input_337_cast)[name = tensor<string, []>("hidden_states_125_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_339_cast = add(x = input_331_cast, y = hidden_states_125_cast)[name = tensor<string, []>("input_339_cast")]; |
| tensor<int32, [1]> hidden_states_127_axes_0 = const()[name = tensor<string, []>("hidden_states_127_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953167744)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953170368)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_127_cast = layer_norm(axes = hidden_states_127_axes_0, beta = text_encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16, x = input_339_cast)[name = tensor<string, []>("hidden_states_127_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(953172992)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956449856)))]; |
| tensor<fp16, [1, 77, 1280]> var_1955_cast = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16, x = hidden_states_127_cast)[name = tensor<string, []>("op_1955_cast")]; |
| tensor<fp16, []> var_1956_to_fp16 = const()[name = tensor<string, []>("op_1956_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_131_cast = mul(x = var_1955_cast, y = var_1956_to_fp16)[name = tensor<string, []>("tensor_131_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956452480)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959729344)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_127_cast = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16, x = hidden_states_127_cast)[name = tensor<string, []>("tensor_127_cast")]; |
| tensor<int32, [4]> var_1961 = const()[name = tensor<string, []>("op_1961"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1962_cast = reshape(shape = var_1961, x = tensor_127_cast)[name = tensor<string, []>("op_1962_cast")]; |
| tensor<int32, [4]> var_1963_perm_0 = const()[name = tensor<string, []>("op_1963_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959731968)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(963008832)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_129_cast = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16, x = hidden_states_127_cast)[name = tensor<string, []>("tensor_129_cast")]; |
| tensor<int32, [4]> var_1968 = const()[name = tensor<string, []>("op_1968"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1969_cast = reshape(shape = var_1968, x = tensor_129_cast)[name = tensor<string, []>("op_1969_cast")]; |
| tensor<int32, [4]> var_1970_perm_0 = const()[name = tensor<string, []>("op_1970_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1977 = const()[name = tensor<string, []>("op_1977"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1978_cast = reshape(shape = var_1977, x = tensor_131_cast)[name = tensor<string, []>("op_1978_cast")]; |
| tensor<int32, [4]> var_1979_perm_0 = const()[name = tensor<string, []>("op_1979_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1981 = const()[name = tensor<string, []>("op_1981"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_53 = transpose(perm = var_1979_perm_0, x = var_1978_cast)[name = tensor<string, []>("transpose_53")]; |
| tensor<fp16, [20, 77, 64]> query_states_43_cast = reshape(shape = var_1981, x = transpose_53)[name = tensor<string, []>("query_states_43_cast")]; |
| tensor<int32, [3]> var_1983 = const()[name = tensor<string, []>("op_1983"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_55 = transpose(perm = var_1963_perm_0, x = var_1962_cast)[name = tensor<string, []>("transpose_55")]; |
| tensor<fp16, [20, 77, 64]> key_states_87_cast = reshape(shape = var_1983, x = transpose_55)[name = tensor<string, []>("key_states_87_cast")]; |
| tensor<int32, [3]> var_1985 = const()[name = tensor<string, []>("op_1985"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_54 = transpose(perm = var_1970_perm_0, x = var_1969_cast)[name = tensor<string, []>("transpose_54")]; |
| tensor<fp16, [20, 77, 64]> value_states_87_cast = reshape(shape = var_1985, x = transpose_54)[name = tensor<string, []>("value_states_87_cast")]; |
| tensor<int32, [3]> var_1988_perm_0 = const()[name = tensor<string, []>("op_1988_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_127_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_127_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_127_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_127_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_52 = transpose(perm = var_1988_perm_0, x = key_states_87_cast)[name = tensor<string, []>("transpose_52")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_127_cast = matmul(transpose_x = attn_weights_127_transpose_x_0, transpose_y = attn_weights_127_transpose_y_0, x = query_states_43_cast, y = transpose_52)[name = tensor<string, []>("attn_weights_127_cast")]; |
| tensor<int32, [4]> var_1990 = const()[name = tensor<string, []>("op_1990"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_1991_cast = reshape(shape = var_1990, x = attn_weights_127_cast)[name = tensor<string, []>("op_1991_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_129_cast = add(x = var_1991_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_129_cast")]; |
| tensor<int32, [3]> var_1996 = const()[name = tensor<string, []>("op_1996"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_341_cast = reshape(shape = var_1996, x = attn_weights_129_cast)[name = tensor<string, []>("input_341_cast")]; |
| tensor<fp16, [20, 77, 77]> input_343_cast = softmax(axis = var_5, x = input_341_cast)[name = tensor<string, []>("input_343_cast")]; |
| tensor<bool, []> attn_output_127_transpose_x_0 = const()[name = tensor<string, []>("attn_output_127_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_127_transpose_y_0 = const()[name = tensor<string, []>("attn_output_127_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_127_cast = matmul(transpose_x = attn_output_127_transpose_x_0, transpose_y = attn_output_127_transpose_y_0, x = input_343_cast, y = value_states_87_cast)[name = tensor<string, []>("attn_output_127_cast")]; |
| tensor<int32, [4]> var_2001 = const()[name = tensor<string, []>("op_2001"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_129_cast = reshape(shape = var_2001, x = attn_output_127_cast)[name = tensor<string, []>("attn_output_129_cast")]; |
| tensor<int32, [4]> attn_output_131_perm_0 = const()[name = tensor<string, []>("attn_output_131_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2004 = const()[name = tensor<string, []>("op_2004"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_51 = transpose(perm = attn_output_131_perm_0, x = attn_output_129_cast)[name = tensor<string, []>("transpose_51")]; |
| tensor<fp16, [1, 77, 1280]> input_345_cast = reshape(shape = var_2004, x = transpose_51)[name = tensor<string, []>("input_345_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(963011456)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(966288320)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_129_cast = linear(bias = text_encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16, x = input_345_cast)[name = tensor<string, []>("hidden_states_129_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_347_cast = add(x = input_339_cast, y = hidden_states_129_cast)[name = tensor<string, []>("input_347_cast")]; |
| tensor<int32, [1]> input_349_axes_0 = const()[name = tensor<string, []>("input_349_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(966290944)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(966293568)))]; |
| tensor<fp16, [1, 77, 1280]> input_349_cast = layer_norm(axes = input_349_axes_0, beta = text_encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16, x = input_347_cast)[name = tensor<string, []>("input_349_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(966296192)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(979403456)))]; |
| tensor<fp16, [1, 77, 5120]> input_351_cast = linear(bias = text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16, x = input_349_cast)[name = tensor<string, []>("input_351_cast")]; |
| tensor<string, []> input_353_mode_0 = const()[name = tensor<string, []>("input_353_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_353_cast = gelu(mode = input_353_mode_0, x = input_351_cast)[name = tensor<string, []>("input_353_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(979413760)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992521024)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_131_cast = linear(bias = text_encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16, x = input_353_cast)[name = tensor<string, []>("hidden_states_131_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_355_cast = add(x = input_347_cast, y = hidden_states_131_cast)[name = tensor<string, []>("input_355_cast")]; |
| tensor<int32, [1]> hidden_states_133_axes_0 = const()[name = tensor<string, []>("hidden_states_133_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992523648)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992526272)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_133_cast = layer_norm(axes = hidden_states_133_axes_0, beta = text_encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16, x = input_355_cast)[name = tensor<string, []>("hidden_states_133_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992528896)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(995805760)))]; |
| tensor<fp16, [1, 77, 1280]> var_2042_cast = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16, x = hidden_states_133_cast)[name = tensor<string, []>("op_2042_cast")]; |
| tensor<fp16, []> var_2043_to_fp16 = const()[name = tensor<string, []>("op_2043_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_137_cast = mul(x = var_2042_cast, y = var_2043_to_fp16)[name = tensor<string, []>("tensor_137_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(995808384)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(999085248)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_133_cast = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16, x = hidden_states_133_cast)[name = tensor<string, []>("tensor_133_cast")]; |
| tensor<int32, [4]> var_2048 = const()[name = tensor<string, []>("op_2048"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2049_cast = reshape(shape = var_2048, x = tensor_133_cast)[name = tensor<string, []>("op_2049_cast")]; |
| tensor<int32, [4]> var_2050_perm_0 = const()[name = tensor<string, []>("op_2050_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(999087872)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1002364736)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_135_cast = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16, x = hidden_states_133_cast)[name = tensor<string, []>("tensor_135_cast")]; |
| tensor<int32, [4]> var_2055 = const()[name = tensor<string, []>("op_2055"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2056_cast = reshape(shape = var_2055, x = tensor_135_cast)[name = tensor<string, []>("op_2056_cast")]; |
| tensor<int32, [4]> var_2057_perm_0 = const()[name = tensor<string, []>("op_2057_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2064 = const()[name = tensor<string, []>("op_2064"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2065_cast = reshape(shape = var_2064, x = tensor_137_cast)[name = tensor<string, []>("op_2065_cast")]; |
| tensor<int32, [4]> var_2066_perm_0 = const()[name = tensor<string, []>("op_2066_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2068 = const()[name = tensor<string, []>("op_2068"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_48 = transpose(perm = var_2066_perm_0, x = var_2065_cast)[name = tensor<string, []>("transpose_48")]; |
| tensor<fp16, [20, 77, 64]> query_states_45_cast = reshape(shape = var_2068, x = transpose_48)[name = tensor<string, []>("query_states_45_cast")]; |
| tensor<int32, [3]> var_2070 = const()[name = tensor<string, []>("op_2070"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_50 = transpose(perm = var_2050_perm_0, x = var_2049_cast)[name = tensor<string, []>("transpose_50")]; |
| tensor<fp16, [20, 77, 64]> key_states_91_cast = reshape(shape = var_2070, x = transpose_50)[name = tensor<string, []>("key_states_91_cast")]; |
| tensor<int32, [3]> var_2072 = const()[name = tensor<string, []>("op_2072"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_49 = transpose(perm = var_2057_perm_0, x = var_2056_cast)[name = tensor<string, []>("transpose_49")]; |
| tensor<fp16, [20, 77, 64]> value_states_91_cast = reshape(shape = var_2072, x = transpose_49)[name = tensor<string, []>("value_states_91_cast")]; |
| tensor<int32, [3]> var_2075_perm_0 = const()[name = tensor<string, []>("op_2075_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_133_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_133_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_133_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_133_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_47 = transpose(perm = var_2075_perm_0, x = key_states_91_cast)[name = tensor<string, []>("transpose_47")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_133_cast = matmul(transpose_x = attn_weights_133_transpose_x_0, transpose_y = attn_weights_133_transpose_y_0, x = query_states_45_cast, y = transpose_47)[name = tensor<string, []>("attn_weights_133_cast")]; |
| tensor<int32, [4]> var_2077 = const()[name = tensor<string, []>("op_2077"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2078_cast = reshape(shape = var_2077, x = attn_weights_133_cast)[name = tensor<string, []>("op_2078_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_135_cast = add(x = var_2078_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_135_cast")]; |
| tensor<int32, [3]> var_2083 = const()[name = tensor<string, []>("op_2083"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_357_cast = reshape(shape = var_2083, x = attn_weights_135_cast)[name = tensor<string, []>("input_357_cast")]; |
| tensor<fp16, [20, 77, 77]> input_359_cast = softmax(axis = var_5, x = input_357_cast)[name = tensor<string, []>("input_359_cast")]; |
| tensor<bool, []> attn_output_133_transpose_x_0 = const()[name = tensor<string, []>("attn_output_133_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_133_transpose_y_0 = const()[name = tensor<string, []>("attn_output_133_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_133_cast = matmul(transpose_x = attn_output_133_transpose_x_0, transpose_y = attn_output_133_transpose_y_0, x = input_359_cast, y = value_states_91_cast)[name = tensor<string, []>("attn_output_133_cast")]; |
| tensor<int32, [4]> var_2088 = const()[name = tensor<string, []>("op_2088"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_135_cast = reshape(shape = var_2088, x = attn_output_133_cast)[name = tensor<string, []>("attn_output_135_cast")]; |
| tensor<int32, [4]> attn_output_137_perm_0 = const()[name = tensor<string, []>("attn_output_137_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2091 = const()[name = tensor<string, []>("op_2091"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_46 = transpose(perm = attn_output_137_perm_0, x = attn_output_135_cast)[name = tensor<string, []>("transpose_46")]; |
| tensor<fp16, [1, 77, 1280]> input_361_cast = reshape(shape = var_2091, x = transpose_46)[name = tensor<string, []>("input_361_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1002367360)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1005644224)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_135_cast = linear(bias = text_encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16, x = input_361_cast)[name = tensor<string, []>("hidden_states_135_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_363_cast = add(x = input_355_cast, y = hidden_states_135_cast)[name = tensor<string, []>("input_363_cast")]; |
| tensor<int32, [1]> input_365_axes_0 = const()[name = tensor<string, []>("input_365_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1005646848)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1005649472)))]; |
| tensor<fp16, [1, 77, 1280]> input_365_cast = layer_norm(axes = input_365_axes_0, beta = text_encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16, x = input_363_cast)[name = tensor<string, []>("input_365_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1005652096)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1018759360)))]; |
| tensor<fp16, [1, 77, 5120]> input_367_cast = linear(bias = text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16, x = input_365_cast)[name = tensor<string, []>("input_367_cast")]; |
| tensor<string, []> input_369_mode_0 = const()[name = tensor<string, []>("input_369_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_369_cast = gelu(mode = input_369_mode_0, x = input_367_cast)[name = tensor<string, []>("input_369_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1018769664)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031876928)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_137_cast = linear(bias = text_encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16, x = input_369_cast)[name = tensor<string, []>("hidden_states_137_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_371_cast = add(x = input_363_cast, y = hidden_states_137_cast)[name = tensor<string, []>("input_371_cast")]; |
| tensor<int32, [1]> hidden_states_139_axes_0 = const()[name = tensor<string, []>("hidden_states_139_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031879552)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031882176)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_139_cast = layer_norm(axes = hidden_states_139_axes_0, beta = text_encoder_text_model_encoder_layers_23_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_23_layer_norm1_weight_to_fp16, x = input_371_cast)[name = tensor<string, []>("hidden_states_139_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1031884800)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1035161664)))]; |
| tensor<fp16, [1, 77, 1280]> var_2129_cast = linear(bias = text_encoder_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16, x = hidden_states_139_cast)[name = tensor<string, []>("op_2129_cast")]; |
| tensor<fp16, []> var_2130_to_fp16 = const()[name = tensor<string, []>("op_2130_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_143_cast = mul(x = var_2129_cast, y = var_2130_to_fp16)[name = tensor<string, []>("tensor_143_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1035164288)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1038441152)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_139_cast = linear(bias = text_encoder_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16, x = hidden_states_139_cast)[name = tensor<string, []>("tensor_139_cast")]; |
| tensor<int32, [4]> var_2135 = const()[name = tensor<string, []>("op_2135"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2136_cast = reshape(shape = var_2135, x = tensor_139_cast)[name = tensor<string, []>("op_2136_cast")]; |
| tensor<int32, [4]> var_2137_perm_0 = const()[name = tensor<string, []>("op_2137_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1038443776)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041720640)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_141_cast = linear(bias = text_encoder_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16, x = hidden_states_139_cast)[name = tensor<string, []>("tensor_141_cast")]; |
| tensor<int32, [4]> var_2142 = const()[name = tensor<string, []>("op_2142"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2143_cast = reshape(shape = var_2142, x = tensor_141_cast)[name = tensor<string, []>("op_2143_cast")]; |
| tensor<int32, [4]> var_2144_perm_0 = const()[name = tensor<string, []>("op_2144_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2151 = const()[name = tensor<string, []>("op_2151"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2152_cast = reshape(shape = var_2151, x = tensor_143_cast)[name = tensor<string, []>("op_2152_cast")]; |
| tensor<int32, [4]> var_2153_perm_0 = const()[name = tensor<string, []>("op_2153_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2155 = const()[name = tensor<string, []>("op_2155"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_43 = transpose(perm = var_2153_perm_0, x = var_2152_cast)[name = tensor<string, []>("transpose_43")]; |
| tensor<fp16, [20, 77, 64]> query_states_47_cast = reshape(shape = var_2155, x = transpose_43)[name = tensor<string, []>("query_states_47_cast")]; |
| tensor<int32, [3]> var_2157 = const()[name = tensor<string, []>("op_2157"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_45 = transpose(perm = var_2137_perm_0, x = var_2136_cast)[name = tensor<string, []>("transpose_45")]; |
| tensor<fp16, [20, 77, 64]> key_states_95_cast = reshape(shape = var_2157, x = transpose_45)[name = tensor<string, []>("key_states_95_cast")]; |
| tensor<int32, [3]> var_2159 = const()[name = tensor<string, []>("op_2159"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_44 = transpose(perm = var_2144_perm_0, x = var_2143_cast)[name = tensor<string, []>("transpose_44")]; |
| tensor<fp16, [20, 77, 64]> value_states_95_cast = reshape(shape = var_2159, x = transpose_44)[name = tensor<string, []>("value_states_95_cast")]; |
| tensor<int32, [3]> var_2162_perm_0 = const()[name = tensor<string, []>("op_2162_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_139_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_139_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_139_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_139_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_42 = transpose(perm = var_2162_perm_0, x = key_states_95_cast)[name = tensor<string, []>("transpose_42")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_139_cast = matmul(transpose_x = attn_weights_139_transpose_x_0, transpose_y = attn_weights_139_transpose_y_0, x = query_states_47_cast, y = transpose_42)[name = tensor<string, []>("attn_weights_139_cast")]; |
| tensor<int32, [4]> var_2164 = const()[name = tensor<string, []>("op_2164"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2165_cast = reshape(shape = var_2164, x = attn_weights_139_cast)[name = tensor<string, []>("op_2165_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_141_cast = add(x = var_2165_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_141_cast")]; |
| tensor<int32, [3]> var_2170 = const()[name = tensor<string, []>("op_2170"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_373_cast = reshape(shape = var_2170, x = attn_weights_141_cast)[name = tensor<string, []>("input_373_cast")]; |
| tensor<fp16, [20, 77, 77]> input_375_cast = softmax(axis = var_5, x = input_373_cast)[name = tensor<string, []>("input_375_cast")]; |
| tensor<bool, []> attn_output_139_transpose_x_0 = const()[name = tensor<string, []>("attn_output_139_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_139_transpose_y_0 = const()[name = tensor<string, []>("attn_output_139_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_139_cast = matmul(transpose_x = attn_output_139_transpose_x_0, transpose_y = attn_output_139_transpose_y_0, x = input_375_cast, y = value_states_95_cast)[name = tensor<string, []>("attn_output_139_cast")]; |
| tensor<int32, [4]> var_2175 = const()[name = tensor<string, []>("op_2175"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_141_cast = reshape(shape = var_2175, x = attn_output_139_cast)[name = tensor<string, []>("attn_output_141_cast")]; |
| tensor<int32, [4]> attn_output_143_perm_0 = const()[name = tensor<string, []>("attn_output_143_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2178 = const()[name = tensor<string, []>("op_2178"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_41 = transpose(perm = attn_output_143_perm_0, x = attn_output_141_cast)[name = tensor<string, []>("transpose_41")]; |
| tensor<fp16, [1, 77, 1280]> input_377_cast = reshape(shape = var_2178, x = transpose_41)[name = tensor<string, []>("input_377_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041723264)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1045000128)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_141_cast = linear(bias = text_encoder_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16, x = input_377_cast)[name = tensor<string, []>("hidden_states_141_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_379_cast = add(x = input_371_cast, y = hidden_states_141_cast)[name = tensor<string, []>("input_379_cast")]; |
| tensor<int32, [1]> input_381_axes_0 = const()[name = tensor<string, []>("input_381_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1045002752)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1045005376)))]; |
| tensor<fp16, [1, 77, 1280]> input_381_cast = layer_norm(axes = input_381_axes_0, beta = text_encoder_text_model_encoder_layers_23_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_23_layer_norm2_weight_to_fp16, x = input_379_cast)[name = tensor<string, []>("input_381_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1045008000)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1058115264)))]; |
| tensor<fp16, [1, 77, 5120]> input_383_cast = linear(bias = text_encoder_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16, x = input_381_cast)[name = tensor<string, []>("input_383_cast")]; |
| tensor<string, []> input_385_mode_0 = const()[name = tensor<string, []>("input_385_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_385_cast = gelu(mode = input_385_mode_0, x = input_383_cast)[name = tensor<string, []>("input_385_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1058125568)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1071232832)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_143_cast = linear(bias = text_encoder_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16, x = input_385_cast)[name = tensor<string, []>("hidden_states_143_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_387_cast = add(x = input_379_cast, y = hidden_states_143_cast)[name = tensor<string, []>("input_387_cast")]; |
| tensor<int32, [1]> hidden_states_145_axes_0 = const()[name = tensor<string, []>("hidden_states_145_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1071235456)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1071238080)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_145_cast = layer_norm(axes = hidden_states_145_axes_0, beta = text_encoder_text_model_encoder_layers_24_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_24_layer_norm1_weight_to_fp16, x = input_387_cast)[name = tensor<string, []>("hidden_states_145_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1071240704)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1074517568)))]; |
| tensor<fp16, [1, 77, 1280]> var_2216_cast = linear(bias = text_encoder_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16, x = hidden_states_145_cast)[name = tensor<string, []>("op_2216_cast")]; |
| tensor<fp16, []> var_2217_to_fp16 = const()[name = tensor<string, []>("op_2217_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_149_cast = mul(x = var_2216_cast, y = var_2217_to_fp16)[name = tensor<string, []>("tensor_149_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1074520192)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077797056)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_145_cast = linear(bias = text_encoder_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16, x = hidden_states_145_cast)[name = tensor<string, []>("tensor_145_cast")]; |
| tensor<int32, [4]> var_2222 = const()[name = tensor<string, []>("op_2222"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2223_cast = reshape(shape = var_2222, x = tensor_145_cast)[name = tensor<string, []>("op_2223_cast")]; |
| tensor<int32, [4]> var_2224_perm_0 = const()[name = tensor<string, []>("op_2224_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1077799680)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1081076544)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_147_cast = linear(bias = text_encoder_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16, x = hidden_states_145_cast)[name = tensor<string, []>("tensor_147_cast")]; |
| tensor<int32, [4]> var_2229 = const()[name = tensor<string, []>("op_2229"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2230_cast = reshape(shape = var_2229, x = tensor_147_cast)[name = tensor<string, []>("op_2230_cast")]; |
| tensor<int32, [4]> var_2231_perm_0 = const()[name = tensor<string, []>("op_2231_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2238 = const()[name = tensor<string, []>("op_2238"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2239_cast = reshape(shape = var_2238, x = tensor_149_cast)[name = tensor<string, []>("op_2239_cast")]; |
| tensor<int32, [4]> var_2240_perm_0 = const()[name = tensor<string, []>("op_2240_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2242 = const()[name = tensor<string, []>("op_2242"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_38 = transpose(perm = var_2240_perm_0, x = var_2239_cast)[name = tensor<string, []>("transpose_38")]; |
| tensor<fp16, [20, 77, 64]> query_states_49_cast = reshape(shape = var_2242, x = transpose_38)[name = tensor<string, []>("query_states_49_cast")]; |
| tensor<int32, [3]> var_2244 = const()[name = tensor<string, []>("op_2244"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_40 = transpose(perm = var_2224_perm_0, x = var_2223_cast)[name = tensor<string, []>("transpose_40")]; |
| tensor<fp16, [20, 77, 64]> key_states_99_cast = reshape(shape = var_2244, x = transpose_40)[name = tensor<string, []>("key_states_99_cast")]; |
| tensor<int32, [3]> var_2246 = const()[name = tensor<string, []>("op_2246"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_39 = transpose(perm = var_2231_perm_0, x = var_2230_cast)[name = tensor<string, []>("transpose_39")]; |
| tensor<fp16, [20, 77, 64]> value_states_99_cast = reshape(shape = var_2246, x = transpose_39)[name = tensor<string, []>("value_states_99_cast")]; |
| tensor<int32, [3]> var_2249_perm_0 = const()[name = tensor<string, []>("op_2249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_145_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_145_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_145_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_145_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_37 = transpose(perm = var_2249_perm_0, x = key_states_99_cast)[name = tensor<string, []>("transpose_37")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_145_cast = matmul(transpose_x = attn_weights_145_transpose_x_0, transpose_y = attn_weights_145_transpose_y_0, x = query_states_49_cast, y = transpose_37)[name = tensor<string, []>("attn_weights_145_cast")]; |
| tensor<int32, [4]> var_2251 = const()[name = tensor<string, []>("op_2251"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2252_cast = reshape(shape = var_2251, x = attn_weights_145_cast)[name = tensor<string, []>("op_2252_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_147_cast = add(x = var_2252_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_147_cast")]; |
| tensor<int32, [3]> var_2257 = const()[name = tensor<string, []>("op_2257"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_389_cast = reshape(shape = var_2257, x = attn_weights_147_cast)[name = tensor<string, []>("input_389_cast")]; |
| tensor<fp16, [20, 77, 77]> input_391_cast = softmax(axis = var_5, x = input_389_cast)[name = tensor<string, []>("input_391_cast")]; |
| tensor<bool, []> attn_output_145_transpose_x_0 = const()[name = tensor<string, []>("attn_output_145_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_145_transpose_y_0 = const()[name = tensor<string, []>("attn_output_145_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_145_cast = matmul(transpose_x = attn_output_145_transpose_x_0, transpose_y = attn_output_145_transpose_y_0, x = input_391_cast, y = value_states_99_cast)[name = tensor<string, []>("attn_output_145_cast")]; |
| tensor<int32, [4]> var_2262 = const()[name = tensor<string, []>("op_2262"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_147_cast = reshape(shape = var_2262, x = attn_output_145_cast)[name = tensor<string, []>("attn_output_147_cast")]; |
| tensor<int32, [4]> attn_output_149_perm_0 = const()[name = tensor<string, []>("attn_output_149_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2265 = const()[name = tensor<string, []>("op_2265"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_36 = transpose(perm = attn_output_149_perm_0, x = attn_output_147_cast)[name = tensor<string, []>("transpose_36")]; |
| tensor<fp16, [1, 77, 1280]> input_393_cast = reshape(shape = var_2265, x = transpose_36)[name = tensor<string, []>("input_393_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1081079168)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1084356032)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_147_cast = linear(bias = text_encoder_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16, x = input_393_cast)[name = tensor<string, []>("hidden_states_147_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_395_cast = add(x = input_387_cast, y = hidden_states_147_cast)[name = tensor<string, []>("input_395_cast")]; |
| tensor<int32, [1]> input_397_axes_0 = const()[name = tensor<string, []>("input_397_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1084358656)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1084361280)))]; |
| tensor<fp16, [1, 77, 1280]> input_397_cast = layer_norm(axes = input_397_axes_0, beta = text_encoder_text_model_encoder_layers_24_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_24_layer_norm2_weight_to_fp16, x = input_395_cast)[name = tensor<string, []>("input_397_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1084363904)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1097471168)))]; |
| tensor<fp16, [1, 77, 5120]> input_399_cast = linear(bias = text_encoder_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16, x = input_397_cast)[name = tensor<string, []>("input_399_cast")]; |
| tensor<string, []> input_401_mode_0 = const()[name = tensor<string, []>("input_401_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_401_cast = gelu(mode = input_401_mode_0, x = input_399_cast)[name = tensor<string, []>("input_401_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1097481472)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1110588736)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_149_cast = linear(bias = text_encoder_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16, x = input_401_cast)[name = tensor<string, []>("hidden_states_149_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_403_cast = add(x = input_395_cast, y = hidden_states_149_cast)[name = tensor<string, []>("input_403_cast")]; |
| tensor<int32, [1]> hidden_states_151_axes_0 = const()[name = tensor<string, []>("hidden_states_151_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1110591360)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1110593984)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_151_cast = layer_norm(axes = hidden_states_151_axes_0, beta = text_encoder_text_model_encoder_layers_25_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_25_layer_norm1_weight_to_fp16, x = input_403_cast)[name = tensor<string, []>("hidden_states_151_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1110596608)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1113873472)))]; |
| tensor<fp16, [1, 77, 1280]> var_2303_cast = linear(bias = text_encoder_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16, x = hidden_states_151_cast)[name = tensor<string, []>("op_2303_cast")]; |
| tensor<fp16, []> var_2304_to_fp16 = const()[name = tensor<string, []>("op_2304_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_155_cast = mul(x = var_2303_cast, y = var_2304_to_fp16)[name = tensor<string, []>("tensor_155_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1113876096)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1117152960)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_151_cast = linear(bias = text_encoder_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16, x = hidden_states_151_cast)[name = tensor<string, []>("tensor_151_cast")]; |
| tensor<int32, [4]> var_2309 = const()[name = tensor<string, []>("op_2309"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2310_cast = reshape(shape = var_2309, x = tensor_151_cast)[name = tensor<string, []>("op_2310_cast")]; |
| tensor<int32, [4]> var_2311_perm_0 = const()[name = tensor<string, []>("op_2311_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1117155584)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120432448)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_153_cast = linear(bias = text_encoder_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16, x = hidden_states_151_cast)[name = tensor<string, []>("tensor_153_cast")]; |
| tensor<int32, [4]> var_2316 = const()[name = tensor<string, []>("op_2316"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2317_cast = reshape(shape = var_2316, x = tensor_153_cast)[name = tensor<string, []>("op_2317_cast")]; |
| tensor<int32, [4]> var_2318_perm_0 = const()[name = tensor<string, []>("op_2318_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2325 = const()[name = tensor<string, []>("op_2325"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2326_cast = reshape(shape = var_2325, x = tensor_155_cast)[name = tensor<string, []>("op_2326_cast")]; |
| tensor<int32, [4]> var_2327_perm_0 = const()[name = tensor<string, []>("op_2327_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2329 = const()[name = tensor<string, []>("op_2329"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_33 = transpose(perm = var_2327_perm_0, x = var_2326_cast)[name = tensor<string, []>("transpose_33")]; |
| tensor<fp16, [20, 77, 64]> query_states_51_cast = reshape(shape = var_2329, x = transpose_33)[name = tensor<string, []>("query_states_51_cast")]; |
| tensor<int32, [3]> var_2331 = const()[name = tensor<string, []>("op_2331"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_35 = transpose(perm = var_2311_perm_0, x = var_2310_cast)[name = tensor<string, []>("transpose_35")]; |
| tensor<fp16, [20, 77, 64]> key_states_103_cast = reshape(shape = var_2331, x = transpose_35)[name = tensor<string, []>("key_states_103_cast")]; |
| tensor<int32, [3]> var_2333 = const()[name = tensor<string, []>("op_2333"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_34 = transpose(perm = var_2318_perm_0, x = var_2317_cast)[name = tensor<string, []>("transpose_34")]; |
| tensor<fp16, [20, 77, 64]> value_states_103_cast = reshape(shape = var_2333, x = transpose_34)[name = tensor<string, []>("value_states_103_cast")]; |
| tensor<int32, [3]> var_2336_perm_0 = const()[name = tensor<string, []>("op_2336_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_151_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_151_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_151_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_151_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_32 = transpose(perm = var_2336_perm_0, x = key_states_103_cast)[name = tensor<string, []>("transpose_32")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_151_cast = matmul(transpose_x = attn_weights_151_transpose_x_0, transpose_y = attn_weights_151_transpose_y_0, x = query_states_51_cast, y = transpose_32)[name = tensor<string, []>("attn_weights_151_cast")]; |
| tensor<int32, [4]> var_2338 = const()[name = tensor<string, []>("op_2338"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2339_cast = reshape(shape = var_2338, x = attn_weights_151_cast)[name = tensor<string, []>("op_2339_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_153_cast = add(x = var_2339_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_153_cast")]; |
| tensor<int32, [3]> var_2344 = const()[name = tensor<string, []>("op_2344"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_405_cast = reshape(shape = var_2344, x = attn_weights_153_cast)[name = tensor<string, []>("input_405_cast")]; |
| tensor<fp16, [20, 77, 77]> input_407_cast = softmax(axis = var_5, x = input_405_cast)[name = tensor<string, []>("input_407_cast")]; |
| tensor<bool, []> attn_output_151_transpose_x_0 = const()[name = tensor<string, []>("attn_output_151_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_151_transpose_y_0 = const()[name = tensor<string, []>("attn_output_151_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_151_cast = matmul(transpose_x = attn_output_151_transpose_x_0, transpose_y = attn_output_151_transpose_y_0, x = input_407_cast, y = value_states_103_cast)[name = tensor<string, []>("attn_output_151_cast")]; |
| tensor<int32, [4]> var_2349 = const()[name = tensor<string, []>("op_2349"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_153_cast = reshape(shape = var_2349, x = attn_output_151_cast)[name = tensor<string, []>("attn_output_153_cast")]; |
| tensor<int32, [4]> attn_output_155_perm_0 = const()[name = tensor<string, []>("attn_output_155_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2352 = const()[name = tensor<string, []>("op_2352"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_31 = transpose(perm = attn_output_155_perm_0, x = attn_output_153_cast)[name = tensor<string, []>("transpose_31")]; |
| tensor<fp16, [1, 77, 1280]> input_409_cast = reshape(shape = var_2352, x = transpose_31)[name = tensor<string, []>("input_409_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120435072)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1123711936)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_153_cast = linear(bias = text_encoder_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16, x = input_409_cast)[name = tensor<string, []>("hidden_states_153_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_411_cast = add(x = input_403_cast, y = hidden_states_153_cast)[name = tensor<string, []>("input_411_cast")]; |
| tensor<int32, [1]> input_413_axes_0 = const()[name = tensor<string, []>("input_413_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1123714560)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1123717184)))]; |
| tensor<fp16, [1, 77, 1280]> input_413_cast = layer_norm(axes = input_413_axes_0, beta = text_encoder_text_model_encoder_layers_25_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_25_layer_norm2_weight_to_fp16, x = input_411_cast)[name = tensor<string, []>("input_413_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1123719808)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1136827072)))]; |
| tensor<fp16, [1, 77, 5120]> input_415_cast = linear(bias = text_encoder_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16, x = input_413_cast)[name = tensor<string, []>("input_415_cast")]; |
| tensor<string, []> input_417_mode_0 = const()[name = tensor<string, []>("input_417_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_417_cast = gelu(mode = input_417_mode_0, x = input_415_cast)[name = tensor<string, []>("input_417_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1136837376)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1149944640)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_155_cast = linear(bias = text_encoder_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16, x = input_417_cast)[name = tensor<string, []>("hidden_states_155_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_419_cast = add(x = input_411_cast, y = hidden_states_155_cast)[name = tensor<string, []>("input_419_cast")]; |
| tensor<int32, [1]> hidden_states_157_axes_0 = const()[name = tensor<string, []>("hidden_states_157_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1149947264)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1149949888)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_157_cast = layer_norm(axes = hidden_states_157_axes_0, beta = text_encoder_text_model_encoder_layers_26_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_26_layer_norm1_weight_to_fp16, x = input_419_cast)[name = tensor<string, []>("hidden_states_157_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1149952512)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1153229376)))]; |
| tensor<fp16, [1, 77, 1280]> var_2390_cast = linear(bias = text_encoder_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16, x = hidden_states_157_cast)[name = tensor<string, []>("op_2390_cast")]; |
| tensor<fp16, []> var_2391_to_fp16 = const()[name = tensor<string, []>("op_2391_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_161_cast = mul(x = var_2390_cast, y = var_2391_to_fp16)[name = tensor<string, []>("tensor_161_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1153232000)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1156508864)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_157_cast = linear(bias = text_encoder_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16, x = hidden_states_157_cast)[name = tensor<string, []>("tensor_157_cast")]; |
| tensor<int32, [4]> var_2396 = const()[name = tensor<string, []>("op_2396"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2397_cast = reshape(shape = var_2396, x = tensor_157_cast)[name = tensor<string, []>("op_2397_cast")]; |
| tensor<int32, [4]> var_2398_perm_0 = const()[name = tensor<string, []>("op_2398_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1156511488)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1159788352)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_159_cast = linear(bias = text_encoder_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16, x = hidden_states_157_cast)[name = tensor<string, []>("tensor_159_cast")]; |
| tensor<int32, [4]> var_2403 = const()[name = tensor<string, []>("op_2403"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2404_cast = reshape(shape = var_2403, x = tensor_159_cast)[name = tensor<string, []>("op_2404_cast")]; |
| tensor<int32, [4]> var_2405_perm_0 = const()[name = tensor<string, []>("op_2405_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2412 = const()[name = tensor<string, []>("op_2412"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2413_cast = reshape(shape = var_2412, x = tensor_161_cast)[name = tensor<string, []>("op_2413_cast")]; |
| tensor<int32, [4]> var_2414_perm_0 = const()[name = tensor<string, []>("op_2414_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2416 = const()[name = tensor<string, []>("op_2416"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_28 = transpose(perm = var_2414_perm_0, x = var_2413_cast)[name = tensor<string, []>("transpose_28")]; |
| tensor<fp16, [20, 77, 64]> query_states_53_cast = reshape(shape = var_2416, x = transpose_28)[name = tensor<string, []>("query_states_53_cast")]; |
| tensor<int32, [3]> var_2418 = const()[name = tensor<string, []>("op_2418"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_30 = transpose(perm = var_2398_perm_0, x = var_2397_cast)[name = tensor<string, []>("transpose_30")]; |
| tensor<fp16, [20, 77, 64]> key_states_107_cast = reshape(shape = var_2418, x = transpose_30)[name = tensor<string, []>("key_states_107_cast")]; |
| tensor<int32, [3]> var_2420 = const()[name = tensor<string, []>("op_2420"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_29 = transpose(perm = var_2405_perm_0, x = var_2404_cast)[name = tensor<string, []>("transpose_29")]; |
| tensor<fp16, [20, 77, 64]> value_states_107_cast = reshape(shape = var_2420, x = transpose_29)[name = tensor<string, []>("value_states_107_cast")]; |
| tensor<int32, [3]> var_2423_perm_0 = const()[name = tensor<string, []>("op_2423_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_157_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_157_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_157_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_157_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_27 = transpose(perm = var_2423_perm_0, x = key_states_107_cast)[name = tensor<string, []>("transpose_27")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_157_cast = matmul(transpose_x = attn_weights_157_transpose_x_0, transpose_y = attn_weights_157_transpose_y_0, x = query_states_53_cast, y = transpose_27)[name = tensor<string, []>("attn_weights_157_cast")]; |
| tensor<int32, [4]> var_2425 = const()[name = tensor<string, []>("op_2425"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2426_cast = reshape(shape = var_2425, x = attn_weights_157_cast)[name = tensor<string, []>("op_2426_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_159_cast = add(x = var_2426_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_159_cast")]; |
| tensor<int32, [3]> var_2431 = const()[name = tensor<string, []>("op_2431"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_421_cast = reshape(shape = var_2431, x = attn_weights_159_cast)[name = tensor<string, []>("input_421_cast")]; |
| tensor<fp16, [20, 77, 77]> input_423_cast = softmax(axis = var_5, x = input_421_cast)[name = tensor<string, []>("input_423_cast")]; |
| tensor<bool, []> attn_output_157_transpose_x_0 = const()[name = tensor<string, []>("attn_output_157_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_157_transpose_y_0 = const()[name = tensor<string, []>("attn_output_157_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_157_cast = matmul(transpose_x = attn_output_157_transpose_x_0, transpose_y = attn_output_157_transpose_y_0, x = input_423_cast, y = value_states_107_cast)[name = tensor<string, []>("attn_output_157_cast")]; |
| tensor<int32, [4]> var_2436 = const()[name = tensor<string, []>("op_2436"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_159_cast = reshape(shape = var_2436, x = attn_output_157_cast)[name = tensor<string, []>("attn_output_159_cast")]; |
| tensor<int32, [4]> attn_output_161_perm_0 = const()[name = tensor<string, []>("attn_output_161_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2439 = const()[name = tensor<string, []>("op_2439"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_26 = transpose(perm = attn_output_161_perm_0, x = attn_output_159_cast)[name = tensor<string, []>("transpose_26")]; |
| tensor<fp16, [1, 77, 1280]> input_425_cast = reshape(shape = var_2439, x = transpose_26)[name = tensor<string, []>("input_425_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1159790976)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1163067840)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_159_cast = linear(bias = text_encoder_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16, x = input_425_cast)[name = tensor<string, []>("hidden_states_159_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_427_cast = add(x = input_419_cast, y = hidden_states_159_cast)[name = tensor<string, []>("input_427_cast")]; |
| tensor<int32, [1]> input_429_axes_0 = const()[name = tensor<string, []>("input_429_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1163070464)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1163073088)))]; |
| tensor<fp16, [1, 77, 1280]> input_429_cast = layer_norm(axes = input_429_axes_0, beta = text_encoder_text_model_encoder_layers_26_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_26_layer_norm2_weight_to_fp16, x = input_427_cast)[name = tensor<string, []>("input_429_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1163075712)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1176182976)))]; |
| tensor<fp16, [1, 77, 5120]> input_431_cast = linear(bias = text_encoder_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16, x = input_429_cast)[name = tensor<string, []>("input_431_cast")]; |
| tensor<string, []> input_433_mode_0 = const()[name = tensor<string, []>("input_433_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_433_cast = gelu(mode = input_433_mode_0, x = input_431_cast)[name = tensor<string, []>("input_433_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1176193280)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1189300544)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_161_cast = linear(bias = text_encoder_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16, x = input_433_cast)[name = tensor<string, []>("hidden_states_161_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_435_cast = add(x = input_427_cast, y = hidden_states_161_cast)[name = tensor<string, []>("input_435_cast")]; |
| tensor<int32, [1]> hidden_states_163_axes_0 = const()[name = tensor<string, []>("hidden_states_163_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1189303168)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1189305792)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_163_cast = layer_norm(axes = hidden_states_163_axes_0, beta = text_encoder_text_model_encoder_layers_27_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_27_layer_norm1_weight_to_fp16, x = input_435_cast)[name = tensor<string, []>("hidden_states_163_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1189308416)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1192585280)))]; |
| tensor<fp16, [1, 77, 1280]> var_2477_cast = linear(bias = text_encoder_text_model_encoder_layers_27_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_27_self_attn_q_proj_weight_to_fp16, x = hidden_states_163_cast)[name = tensor<string, []>("op_2477_cast")]; |
| tensor<fp16, []> var_2478_to_fp16 = const()[name = tensor<string, []>("op_2478_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_167_cast = mul(x = var_2477_cast, y = var_2478_to_fp16)[name = tensor<string, []>("tensor_167_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1192587904)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195864768)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_163_cast = linear(bias = text_encoder_text_model_encoder_layers_27_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_27_self_attn_k_proj_weight_to_fp16, x = hidden_states_163_cast)[name = tensor<string, []>("tensor_163_cast")]; |
| tensor<int32, [4]> var_2483 = const()[name = tensor<string, []>("op_2483"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2484_cast = reshape(shape = var_2483, x = tensor_163_cast)[name = tensor<string, []>("op_2484_cast")]; |
| tensor<int32, [4]> var_2485_perm_0 = const()[name = tensor<string, []>("op_2485_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1195867392)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1199144256)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_165_cast = linear(bias = text_encoder_text_model_encoder_layers_27_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_27_self_attn_v_proj_weight_to_fp16, x = hidden_states_163_cast)[name = tensor<string, []>("tensor_165_cast")]; |
| tensor<int32, [4]> var_2490 = const()[name = tensor<string, []>("op_2490"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2491_cast = reshape(shape = var_2490, x = tensor_165_cast)[name = tensor<string, []>("op_2491_cast")]; |
| tensor<int32, [4]> var_2492_perm_0 = const()[name = tensor<string, []>("op_2492_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2499 = const()[name = tensor<string, []>("op_2499"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2500_cast = reshape(shape = var_2499, x = tensor_167_cast)[name = tensor<string, []>("op_2500_cast")]; |
| tensor<int32, [4]> var_2501_perm_0 = const()[name = tensor<string, []>("op_2501_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2503 = const()[name = tensor<string, []>("op_2503"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_23 = transpose(perm = var_2501_perm_0, x = var_2500_cast)[name = tensor<string, []>("transpose_23")]; |
| tensor<fp16, [20, 77, 64]> query_states_55_cast = reshape(shape = var_2503, x = transpose_23)[name = tensor<string, []>("query_states_55_cast")]; |
| tensor<int32, [3]> var_2505 = const()[name = tensor<string, []>("op_2505"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_25 = transpose(perm = var_2485_perm_0, x = var_2484_cast)[name = tensor<string, []>("transpose_25")]; |
| tensor<fp16, [20, 77, 64]> key_states_111_cast = reshape(shape = var_2505, x = transpose_25)[name = tensor<string, []>("key_states_111_cast")]; |
| tensor<int32, [3]> var_2507 = const()[name = tensor<string, []>("op_2507"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_24 = transpose(perm = var_2492_perm_0, x = var_2491_cast)[name = tensor<string, []>("transpose_24")]; |
| tensor<fp16, [20, 77, 64]> value_states_111_cast = reshape(shape = var_2507, x = transpose_24)[name = tensor<string, []>("value_states_111_cast")]; |
| tensor<int32, [3]> var_2510_perm_0 = const()[name = tensor<string, []>("op_2510_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_163_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_163_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_163_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_163_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_22 = transpose(perm = var_2510_perm_0, x = key_states_111_cast)[name = tensor<string, []>("transpose_22")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_163_cast = matmul(transpose_x = attn_weights_163_transpose_x_0, transpose_y = attn_weights_163_transpose_y_0, x = query_states_55_cast, y = transpose_22)[name = tensor<string, []>("attn_weights_163_cast")]; |
| tensor<int32, [4]> var_2512 = const()[name = tensor<string, []>("op_2512"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2513_cast = reshape(shape = var_2512, x = attn_weights_163_cast)[name = tensor<string, []>("op_2513_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_165_cast = add(x = var_2513_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_165_cast")]; |
| tensor<int32, [3]> var_2518 = const()[name = tensor<string, []>("op_2518"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_437_cast = reshape(shape = var_2518, x = attn_weights_165_cast)[name = tensor<string, []>("input_437_cast")]; |
| tensor<fp16, [20, 77, 77]> input_439_cast = softmax(axis = var_5, x = input_437_cast)[name = tensor<string, []>("input_439_cast")]; |
| tensor<bool, []> attn_output_163_transpose_x_0 = const()[name = tensor<string, []>("attn_output_163_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_163_transpose_y_0 = const()[name = tensor<string, []>("attn_output_163_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_163_cast = matmul(transpose_x = attn_output_163_transpose_x_0, transpose_y = attn_output_163_transpose_y_0, x = input_439_cast, y = value_states_111_cast)[name = tensor<string, []>("attn_output_163_cast")]; |
| tensor<int32, [4]> var_2523 = const()[name = tensor<string, []>("op_2523"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_165_cast = reshape(shape = var_2523, x = attn_output_163_cast)[name = tensor<string, []>("attn_output_165_cast")]; |
| tensor<int32, [4]> attn_output_167_perm_0 = const()[name = tensor<string, []>("attn_output_167_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2526 = const()[name = tensor<string, []>("op_2526"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_21 = transpose(perm = attn_output_167_perm_0, x = attn_output_165_cast)[name = tensor<string, []>("transpose_21")]; |
| tensor<fp16, [1, 77, 1280]> input_441_cast = reshape(shape = var_2526, x = transpose_21)[name = tensor<string, []>("input_441_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_27_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1199146880)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1202423744)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_165_cast = linear(bias = text_encoder_text_model_encoder_layers_27_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_27_self_attn_out_proj_weight_to_fp16, x = input_441_cast)[name = tensor<string, []>("hidden_states_165_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_443_cast = add(x = input_435_cast, y = hidden_states_165_cast)[name = tensor<string, []>("input_443_cast")]; |
| tensor<int32, [1]> input_445_axes_0 = const()[name = tensor<string, []>("input_445_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1202426368)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1202428992)))]; |
| tensor<fp16, [1, 77, 1280]> input_445_cast = layer_norm(axes = input_445_axes_0, beta = text_encoder_text_model_encoder_layers_27_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_27_layer_norm2_weight_to_fp16, x = input_443_cast)[name = tensor<string, []>("input_445_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_27_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1202431616)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_27_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1215538880)))]; |
| tensor<fp16, [1, 77, 5120]> input_447_cast = linear(bias = text_encoder_text_model_encoder_layers_27_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_27_mlp_fc1_weight_to_fp16, x = input_445_cast)[name = tensor<string, []>("input_447_cast")]; |
| tensor<string, []> input_449_mode_0 = const()[name = tensor<string, []>("input_449_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_449_cast = gelu(mode = input_449_mode_0, x = input_447_cast)[name = tensor<string, []>("input_449_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_27_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1215549184)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_27_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_27_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1228656448)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_167_cast = linear(bias = text_encoder_text_model_encoder_layers_27_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_27_mlp_fc2_weight_to_fp16, x = input_449_cast)[name = tensor<string, []>("hidden_states_167_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_451_cast = add(x = input_443_cast, y = hidden_states_167_cast)[name = tensor<string, []>("input_451_cast")]; |
| tensor<int32, [1]> hidden_states_169_axes_0 = const()[name = tensor<string, []>("hidden_states_169_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1228659072)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1228661696)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_169_cast = layer_norm(axes = hidden_states_169_axes_0, beta = text_encoder_text_model_encoder_layers_28_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_28_layer_norm1_weight_to_fp16, x = input_451_cast)[name = tensor<string, []>("hidden_states_169_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1228664320)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1231941184)))]; |
| tensor<fp16, [1, 77, 1280]> var_2564_cast = linear(bias = text_encoder_text_model_encoder_layers_28_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_28_self_attn_q_proj_weight_to_fp16, x = hidden_states_169_cast)[name = tensor<string, []>("op_2564_cast")]; |
| tensor<fp16, []> var_2565_to_fp16 = const()[name = tensor<string, []>("op_2565_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_173_cast = mul(x = var_2564_cast, y = var_2565_to_fp16)[name = tensor<string, []>("tensor_173_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1231943808)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1235220672)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_169_cast = linear(bias = text_encoder_text_model_encoder_layers_28_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_28_self_attn_k_proj_weight_to_fp16, x = hidden_states_169_cast)[name = tensor<string, []>("tensor_169_cast")]; |
| tensor<int32, [4]> var_2570 = const()[name = tensor<string, []>("op_2570"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2571_cast = reshape(shape = var_2570, x = tensor_169_cast)[name = tensor<string, []>("op_2571_cast")]; |
| tensor<int32, [4]> var_2572_perm_0 = const()[name = tensor<string, []>("op_2572_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1235223296)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1238500160)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_171_cast = linear(bias = text_encoder_text_model_encoder_layers_28_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_28_self_attn_v_proj_weight_to_fp16, x = hidden_states_169_cast)[name = tensor<string, []>("tensor_171_cast")]; |
| tensor<int32, [4]> var_2577 = const()[name = tensor<string, []>("op_2577"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2578_cast = reshape(shape = var_2577, x = tensor_171_cast)[name = tensor<string, []>("op_2578_cast")]; |
| tensor<int32, [4]> var_2579_perm_0 = const()[name = tensor<string, []>("op_2579_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2586 = const()[name = tensor<string, []>("op_2586"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2587_cast = reshape(shape = var_2586, x = tensor_173_cast)[name = tensor<string, []>("op_2587_cast")]; |
| tensor<int32, [4]> var_2588_perm_0 = const()[name = tensor<string, []>("op_2588_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2590 = const()[name = tensor<string, []>("op_2590"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_18 = transpose(perm = var_2588_perm_0, x = var_2587_cast)[name = tensor<string, []>("transpose_18")]; |
| tensor<fp16, [20, 77, 64]> query_states_57_cast = reshape(shape = var_2590, x = transpose_18)[name = tensor<string, []>("query_states_57_cast")]; |
| tensor<int32, [3]> var_2592 = const()[name = tensor<string, []>("op_2592"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_20 = transpose(perm = var_2572_perm_0, x = var_2571_cast)[name = tensor<string, []>("transpose_20")]; |
| tensor<fp16, [20, 77, 64]> key_states_115_cast = reshape(shape = var_2592, x = transpose_20)[name = tensor<string, []>("key_states_115_cast")]; |
| tensor<int32, [3]> var_2594 = const()[name = tensor<string, []>("op_2594"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_19 = transpose(perm = var_2579_perm_0, x = var_2578_cast)[name = tensor<string, []>("transpose_19")]; |
| tensor<fp16, [20, 77, 64]> value_states_115_cast = reshape(shape = var_2594, x = transpose_19)[name = tensor<string, []>("value_states_115_cast")]; |
| tensor<int32, [3]> var_2597_perm_0 = const()[name = tensor<string, []>("op_2597_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_169_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_169_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_169_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_169_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_17 = transpose(perm = var_2597_perm_0, x = key_states_115_cast)[name = tensor<string, []>("transpose_17")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_169_cast = matmul(transpose_x = attn_weights_169_transpose_x_0, transpose_y = attn_weights_169_transpose_y_0, x = query_states_57_cast, y = transpose_17)[name = tensor<string, []>("attn_weights_169_cast")]; |
| tensor<int32, [4]> var_2599 = const()[name = tensor<string, []>("op_2599"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2600_cast = reshape(shape = var_2599, x = attn_weights_169_cast)[name = tensor<string, []>("op_2600_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_171_cast = add(x = var_2600_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_171_cast")]; |
| tensor<int32, [3]> var_2605 = const()[name = tensor<string, []>("op_2605"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_453_cast = reshape(shape = var_2605, x = attn_weights_171_cast)[name = tensor<string, []>("input_453_cast")]; |
| tensor<fp16, [20, 77, 77]> input_455_cast = softmax(axis = var_5, x = input_453_cast)[name = tensor<string, []>("input_455_cast")]; |
| tensor<bool, []> attn_output_169_transpose_x_0 = const()[name = tensor<string, []>("attn_output_169_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_169_transpose_y_0 = const()[name = tensor<string, []>("attn_output_169_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_169_cast = matmul(transpose_x = attn_output_169_transpose_x_0, transpose_y = attn_output_169_transpose_y_0, x = input_455_cast, y = value_states_115_cast)[name = tensor<string, []>("attn_output_169_cast")]; |
| tensor<int32, [4]> var_2610 = const()[name = tensor<string, []>("op_2610"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_171_cast = reshape(shape = var_2610, x = attn_output_169_cast)[name = tensor<string, []>("attn_output_171_cast")]; |
| tensor<int32, [4]> attn_output_173_perm_0 = const()[name = tensor<string, []>("attn_output_173_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2613 = const()[name = tensor<string, []>("op_2613"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_16 = transpose(perm = attn_output_173_perm_0, x = attn_output_171_cast)[name = tensor<string, []>("transpose_16")]; |
| tensor<fp16, [1, 77, 1280]> input_457_cast = reshape(shape = var_2613, x = transpose_16)[name = tensor<string, []>("input_457_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_28_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1238502784)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1241779648)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_171_cast = linear(bias = text_encoder_text_model_encoder_layers_28_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_28_self_attn_out_proj_weight_to_fp16, x = input_457_cast)[name = tensor<string, []>("hidden_states_171_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_459_cast = add(x = input_451_cast, y = hidden_states_171_cast)[name = tensor<string, []>("input_459_cast")]; |
| tensor<int32, [1]> input_461_axes_0 = const()[name = tensor<string, []>("input_461_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1241782272)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1241784896)))]; |
| tensor<fp16, [1, 77, 1280]> input_461_cast = layer_norm(axes = input_461_axes_0, beta = text_encoder_text_model_encoder_layers_28_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_28_layer_norm2_weight_to_fp16, x = input_459_cast)[name = tensor<string, []>("input_461_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_28_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1241787520)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_28_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1254894784)))]; |
| tensor<fp16, [1, 77, 5120]> input_463_cast = linear(bias = text_encoder_text_model_encoder_layers_28_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_28_mlp_fc1_weight_to_fp16, x = input_461_cast)[name = tensor<string, []>("input_463_cast")]; |
| tensor<string, []> input_465_mode_0 = const()[name = tensor<string, []>("input_465_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_465_cast = gelu(mode = input_465_mode_0, x = input_463_cast)[name = tensor<string, []>("input_465_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_28_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1254905088)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_28_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_28_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1268012352)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_173_cast = linear(bias = text_encoder_text_model_encoder_layers_28_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_28_mlp_fc2_weight_to_fp16, x = input_465_cast)[name = tensor<string, []>("hidden_states_173_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_467_cast = add(x = input_459_cast, y = hidden_states_173_cast)[name = tensor<string, []>("input_467_cast")]; |
| tensor<int32, [1]> hidden_states_175_axes_0 = const()[name = tensor<string, []>("hidden_states_175_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1268014976)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1268017600)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_175_cast = layer_norm(axes = hidden_states_175_axes_0, beta = text_encoder_text_model_encoder_layers_29_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_29_layer_norm1_weight_to_fp16, x = input_467_cast)[name = tensor<string, []>("hidden_states_175_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1268020224)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1271297088)))]; |
| tensor<fp16, [1, 77, 1280]> var_2651_cast = linear(bias = text_encoder_text_model_encoder_layers_29_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_29_self_attn_q_proj_weight_to_fp16, x = hidden_states_175_cast)[name = tensor<string, []>("op_2651_cast")]; |
| tensor<fp16, []> var_2652_to_fp16 = const()[name = tensor<string, []>("op_2652_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_179_cast = mul(x = var_2651_cast, y = var_2652_to_fp16)[name = tensor<string, []>("tensor_179_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1271299712)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1274576576)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_175_cast = linear(bias = text_encoder_text_model_encoder_layers_29_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_29_self_attn_k_proj_weight_to_fp16, x = hidden_states_175_cast)[name = tensor<string, []>("tensor_175_cast")]; |
| tensor<int32, [4]> var_2657 = const()[name = tensor<string, []>("op_2657"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2658_cast = reshape(shape = var_2657, x = tensor_175_cast)[name = tensor<string, []>("op_2658_cast")]; |
| tensor<int32, [4]> var_2659_perm_0 = const()[name = tensor<string, []>("op_2659_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1274579200)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1277856064)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_177_cast = linear(bias = text_encoder_text_model_encoder_layers_29_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_29_self_attn_v_proj_weight_to_fp16, x = hidden_states_175_cast)[name = tensor<string, []>("tensor_177_cast")]; |
| tensor<int32, [4]> var_2664 = const()[name = tensor<string, []>("op_2664"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2665_cast = reshape(shape = var_2664, x = tensor_177_cast)[name = tensor<string, []>("op_2665_cast")]; |
| tensor<int32, [4]> var_2666_perm_0 = const()[name = tensor<string, []>("op_2666_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2673 = const()[name = tensor<string, []>("op_2673"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2674_cast = reshape(shape = var_2673, x = tensor_179_cast)[name = tensor<string, []>("op_2674_cast")]; |
| tensor<int32, [4]> var_2675_perm_0 = const()[name = tensor<string, []>("op_2675_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2677 = const()[name = tensor<string, []>("op_2677"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_13 = transpose(perm = var_2675_perm_0, x = var_2674_cast)[name = tensor<string, []>("transpose_13")]; |
| tensor<fp16, [20, 77, 64]> query_states_59_cast = reshape(shape = var_2677, x = transpose_13)[name = tensor<string, []>("query_states_59_cast")]; |
| tensor<int32, [3]> var_2679 = const()[name = tensor<string, []>("op_2679"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_15 = transpose(perm = var_2659_perm_0, x = var_2658_cast)[name = tensor<string, []>("transpose_15")]; |
| tensor<fp16, [20, 77, 64]> key_states_119_cast = reshape(shape = var_2679, x = transpose_15)[name = tensor<string, []>("key_states_119_cast")]; |
| tensor<int32, [3]> var_2681 = const()[name = tensor<string, []>("op_2681"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_14 = transpose(perm = var_2666_perm_0, x = var_2665_cast)[name = tensor<string, []>("transpose_14")]; |
| tensor<fp16, [20, 77, 64]> value_states_119_cast = reshape(shape = var_2681, x = transpose_14)[name = tensor<string, []>("value_states_119_cast")]; |
| tensor<int32, [3]> var_2684_perm_0 = const()[name = tensor<string, []>("op_2684_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_175_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_175_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_175_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_175_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_12 = transpose(perm = var_2684_perm_0, x = key_states_119_cast)[name = tensor<string, []>("transpose_12")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_175_cast = matmul(transpose_x = attn_weights_175_transpose_x_0, transpose_y = attn_weights_175_transpose_y_0, x = query_states_59_cast, y = transpose_12)[name = tensor<string, []>("attn_weights_175_cast")]; |
| tensor<int32, [4]> var_2686 = const()[name = tensor<string, []>("op_2686"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2687_cast = reshape(shape = var_2686, x = attn_weights_175_cast)[name = tensor<string, []>("op_2687_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_177_cast = add(x = var_2687_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_177_cast")]; |
| tensor<int32, [3]> var_2692 = const()[name = tensor<string, []>("op_2692"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_469_cast = reshape(shape = var_2692, x = attn_weights_177_cast)[name = tensor<string, []>("input_469_cast")]; |
| tensor<fp16, [20, 77, 77]> input_471_cast = softmax(axis = var_5, x = input_469_cast)[name = tensor<string, []>("input_471_cast")]; |
| tensor<bool, []> attn_output_175_transpose_x_0 = const()[name = tensor<string, []>("attn_output_175_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_175_transpose_y_0 = const()[name = tensor<string, []>("attn_output_175_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_175_cast = matmul(transpose_x = attn_output_175_transpose_x_0, transpose_y = attn_output_175_transpose_y_0, x = input_471_cast, y = value_states_119_cast)[name = tensor<string, []>("attn_output_175_cast")]; |
| tensor<int32, [4]> var_2697 = const()[name = tensor<string, []>("op_2697"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_177_cast = reshape(shape = var_2697, x = attn_output_175_cast)[name = tensor<string, []>("attn_output_177_cast")]; |
| tensor<int32, [4]> attn_output_179_perm_0 = const()[name = tensor<string, []>("attn_output_179_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2700 = const()[name = tensor<string, []>("op_2700"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_11 = transpose(perm = attn_output_179_perm_0, x = attn_output_177_cast)[name = tensor<string, []>("transpose_11")]; |
| tensor<fp16, [1, 77, 1280]> input_473_cast = reshape(shape = var_2700, x = transpose_11)[name = tensor<string, []>("input_473_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_29_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1277858688)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1281135552)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_177_cast = linear(bias = text_encoder_text_model_encoder_layers_29_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_29_self_attn_out_proj_weight_to_fp16, x = input_473_cast)[name = tensor<string, []>("hidden_states_177_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_475_cast = add(x = input_467_cast, y = hidden_states_177_cast)[name = tensor<string, []>("input_475_cast")]; |
| tensor<int32, [1]> input_477_axes_0 = const()[name = tensor<string, []>("input_477_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1281138176)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1281140800)))]; |
| tensor<fp16, [1, 77, 1280]> input_477_cast = layer_norm(axes = input_477_axes_0, beta = text_encoder_text_model_encoder_layers_29_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_29_layer_norm2_weight_to_fp16, x = input_475_cast)[name = tensor<string, []>("input_477_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_29_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1281143424)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_29_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1294250688)))]; |
| tensor<fp16, [1, 77, 5120]> input_479_cast = linear(bias = text_encoder_text_model_encoder_layers_29_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_29_mlp_fc1_weight_to_fp16, x = input_477_cast)[name = tensor<string, []>("input_479_cast")]; |
| tensor<string, []> input_481_mode_0 = const()[name = tensor<string, []>("input_481_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_481_cast = gelu(mode = input_481_mode_0, x = input_479_cast)[name = tensor<string, []>("input_481_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_29_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1294260992)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_29_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_29_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1307368256)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_179_cast = linear(bias = text_encoder_text_model_encoder_layers_29_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_29_mlp_fc2_weight_to_fp16, x = input_481_cast)[name = tensor<string, []>("hidden_states_179_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_483_cast = add(x = input_475_cast, y = hidden_states_179_cast)[name = tensor<string, []>("input_483_cast")]; |
| tensor<int32, [1]> hidden_states_181_axes_0 = const()[name = tensor<string, []>("hidden_states_181_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1307370880)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1307373504)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_181_cast = layer_norm(axes = hidden_states_181_axes_0, beta = text_encoder_text_model_encoder_layers_30_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_30_layer_norm1_weight_to_fp16, x = input_483_cast)[name = tensor<string, []>("hidden_states_181_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_30_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1307376128)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1310652992)))]; |
| tensor<fp16, [1, 77, 1280]> var_2738_cast = linear(bias = text_encoder_text_model_encoder_layers_30_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_30_self_attn_q_proj_weight_to_fp16, x = hidden_states_181_cast)[name = tensor<string, []>("op_2738_cast")]; |
| tensor<fp16, []> var_2739_to_fp16 = const()[name = tensor<string, []>("op_2739_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_185_cast = mul(x = var_2738_cast, y = var_2739_to_fp16)[name = tensor<string, []>("tensor_185_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_30_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1310655616)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313932480)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_181_cast = linear(bias = text_encoder_text_model_encoder_layers_30_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_30_self_attn_k_proj_weight_to_fp16, x = hidden_states_181_cast)[name = tensor<string, []>("tensor_181_cast")]; |
| tensor<int32, [4]> var_2744 = const()[name = tensor<string, []>("op_2744"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2745_cast = reshape(shape = var_2744, x = tensor_181_cast)[name = tensor<string, []>("op_2745_cast")]; |
| tensor<int32, [4]> var_2746_perm_0 = const()[name = tensor<string, []>("op_2746_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_30_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313935104)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1317211968)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_183_cast = linear(bias = text_encoder_text_model_encoder_layers_30_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_30_self_attn_v_proj_weight_to_fp16, x = hidden_states_181_cast)[name = tensor<string, []>("tensor_183_cast")]; |
| tensor<int32, [4]> var_2751 = const()[name = tensor<string, []>("op_2751"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2752_cast = reshape(shape = var_2751, x = tensor_183_cast)[name = tensor<string, []>("op_2752_cast")]; |
| tensor<int32, [4]> var_2753_perm_0 = const()[name = tensor<string, []>("op_2753_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2760 = const()[name = tensor<string, []>("op_2760"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2761_cast = reshape(shape = var_2760, x = tensor_185_cast)[name = tensor<string, []>("op_2761_cast")]; |
| tensor<int32, [4]> var_2762_perm_0 = const()[name = tensor<string, []>("op_2762_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2764 = const()[name = tensor<string, []>("op_2764"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_8 = transpose(perm = var_2762_perm_0, x = var_2761_cast)[name = tensor<string, []>("transpose_8")]; |
| tensor<fp16, [20, 77, 64]> query_states_61_cast = reshape(shape = var_2764, x = transpose_8)[name = tensor<string, []>("query_states_61_cast")]; |
| tensor<int32, [3]> var_2766 = const()[name = tensor<string, []>("op_2766"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_10 = transpose(perm = var_2746_perm_0, x = var_2745_cast)[name = tensor<string, []>("transpose_10")]; |
| tensor<fp16, [20, 77, 64]> key_states_123_cast = reshape(shape = var_2766, x = transpose_10)[name = tensor<string, []>("key_states_123_cast")]; |
| tensor<int32, [3]> var_2768 = const()[name = tensor<string, []>("op_2768"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_9 = transpose(perm = var_2753_perm_0, x = var_2752_cast)[name = tensor<string, []>("transpose_9")]; |
| tensor<fp16, [20, 77, 64]> value_states_123_cast = reshape(shape = var_2768, x = transpose_9)[name = tensor<string, []>("value_states_123_cast")]; |
| tensor<int32, [3]> var_2771_perm_0 = const()[name = tensor<string, []>("op_2771_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_181_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_181_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_181_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_181_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_7 = transpose(perm = var_2771_perm_0, x = key_states_123_cast)[name = tensor<string, []>("transpose_7")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_181_cast = matmul(transpose_x = attn_weights_181_transpose_x_0, transpose_y = attn_weights_181_transpose_y_0, x = query_states_61_cast, y = transpose_7)[name = tensor<string, []>("attn_weights_181_cast")]; |
| tensor<int32, [4]> var_2773 = const()[name = tensor<string, []>("op_2773"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2774_cast = reshape(shape = var_2773, x = attn_weights_181_cast)[name = tensor<string, []>("op_2774_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_183_cast = add(x = var_2774_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_183_cast")]; |
| tensor<int32, [3]> var_2779 = const()[name = tensor<string, []>("op_2779"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_485_cast = reshape(shape = var_2779, x = attn_weights_183_cast)[name = tensor<string, []>("input_485_cast")]; |
| tensor<fp16, [20, 77, 77]> input_487_cast = softmax(axis = var_5, x = input_485_cast)[name = tensor<string, []>("input_487_cast")]; |
| tensor<bool, []> attn_output_181_transpose_x_0 = const()[name = tensor<string, []>("attn_output_181_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_181_transpose_y_0 = const()[name = tensor<string, []>("attn_output_181_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_181_cast = matmul(transpose_x = attn_output_181_transpose_x_0, transpose_y = attn_output_181_transpose_y_0, x = input_487_cast, y = value_states_123_cast)[name = tensor<string, []>("attn_output_181_cast")]; |
| tensor<int32, [4]> var_2784 = const()[name = tensor<string, []>("op_2784"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_183_cast = reshape(shape = var_2784, x = attn_output_181_cast)[name = tensor<string, []>("attn_output_183_cast")]; |
| tensor<int32, [4]> attn_output_185_perm_0 = const()[name = tensor<string, []>("attn_output_185_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2787 = const()[name = tensor<string, []>("op_2787"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_6 = transpose(perm = attn_output_185_perm_0, x = attn_output_183_cast)[name = tensor<string, []>("transpose_6")]; |
| tensor<fp16, [1, 77, 1280]> input_489_cast = reshape(shape = var_2787, x = transpose_6)[name = tensor<string, []>("input_489_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_30_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1317214592)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1320491456)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_183_cast = linear(bias = text_encoder_text_model_encoder_layers_30_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_30_self_attn_out_proj_weight_to_fp16, x = input_489_cast)[name = tensor<string, []>("hidden_states_183_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_491_cast = add(x = input_483_cast, y = hidden_states_183_cast)[name = tensor<string, []>("input_491_cast")]; |
| tensor<int32, [1]> input_493_axes_0 = const()[name = tensor<string, []>("input_493_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1320494080)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1320496704)))]; |
| tensor<fp16, [1, 77, 1280]> input_493_cast = layer_norm(axes = input_493_axes_0, beta = text_encoder_text_model_encoder_layers_30_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_30_layer_norm2_weight_to_fp16, x = input_491_cast)[name = tensor<string, []>("input_493_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_30_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1320499328)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_30_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1333606592)))]; |
| tensor<fp16, [1, 77, 5120]> input_495_cast = linear(bias = text_encoder_text_model_encoder_layers_30_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_30_mlp_fc1_weight_to_fp16, x = input_493_cast)[name = tensor<string, []>("input_495_cast")]; |
| tensor<string, []> input_497_mode_0 = const()[name = tensor<string, []>("input_497_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_497_cast = gelu(mode = input_497_mode_0, x = input_495_cast)[name = tensor<string, []>("input_497_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_30_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1333616896)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_30_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_30_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1346724160)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_185_cast = linear(bias = text_encoder_text_model_encoder_layers_30_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_30_mlp_fc2_weight_to_fp16, x = input_497_cast)[name = tensor<string, []>("hidden_states_185_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_499_cast = add(x = input_491_cast, y = hidden_states_185_cast)[name = tensor<string, []>("input_499_cast")]; |
| tensor<string, []> input_499_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("input_499_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<int32, [1]> hidden_states_187_axes_0 = const()[name = tensor<string, []>("hidden_states_187_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1346726784)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1346729408)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_187_cast = layer_norm(axes = hidden_states_187_axes_0, beta = text_encoder_text_model_encoder_layers_31_layer_norm1_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_31_layer_norm1_weight_to_fp16, x = input_499_cast)[name = tensor<string, []>("hidden_states_187_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_31_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1346732032)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1350008896)))]; |
| tensor<fp16, [1, 77, 1280]> var_2825_cast = linear(bias = text_encoder_text_model_encoder_layers_31_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_31_self_attn_q_proj_weight_to_fp16, x = hidden_states_187_cast)[name = tensor<string, []>("op_2825_cast")]; |
| tensor<fp16, []> var_2826_to_fp16 = const()[name = tensor<string, []>("op_2826_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1280]> tensor_cast = mul(x = var_2825_cast, y = var_2826_to_fp16)[name = tensor<string, []>("tensor_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_31_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1350011520)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1353288384)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_187_cast = linear(bias = text_encoder_text_model_encoder_layers_31_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_31_self_attn_k_proj_weight_to_fp16, x = hidden_states_187_cast)[name = tensor<string, []>("tensor_187_cast")]; |
| tensor<int32, [4]> var_2831 = const()[name = tensor<string, []>("op_2831"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2832_cast = reshape(shape = var_2831, x = tensor_187_cast)[name = tensor<string, []>("op_2832_cast")]; |
| tensor<int32, [4]> var_2833_perm_0 = const()[name = tensor<string, []>("op_2833_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_31_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1353291008)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1356567872)))]; |
| tensor<fp16, [1, 77, 1280]> tensor_189_cast = linear(bias = text_encoder_text_model_encoder_layers_31_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_31_self_attn_v_proj_weight_to_fp16, x = hidden_states_187_cast)[name = tensor<string, []>("tensor_189_cast")]; |
| tensor<int32, [4]> var_2838 = const()[name = tensor<string, []>("op_2838"), val = tensor<int32, [4]>([1, -1, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2839_cast = reshape(shape = var_2838, x = tensor_189_cast)[name = tensor<string, []>("op_2839_cast")]; |
| tensor<int32, [4]> var_2840_perm_0 = const()[name = tensor<string, []>("op_2840_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2847 = const()[name = tensor<string, []>("op_2847"), val = tensor<int32, [4]>([1, 77, 20, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_2848_cast = reshape(shape = var_2847, x = tensor_cast)[name = tensor<string, []>("op_2848_cast")]; |
| tensor<int32, [4]> var_2849_perm_0 = const()[name = tensor<string, []>("op_2849_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2851 = const()[name = tensor<string, []>("op_2851"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_3 = transpose(perm = var_2849_perm_0, x = var_2848_cast)[name = tensor<string, []>("transpose_3")]; |
| tensor<fp16, [20, 77, 64]> query_states_cast = reshape(shape = var_2851, x = transpose_3)[name = tensor<string, []>("query_states_cast")]; |
| tensor<int32, [3]> var_2853 = const()[name = tensor<string, []>("op_2853"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_5 = transpose(perm = var_2833_perm_0, x = var_2832_cast)[name = tensor<string, []>("transpose_5")]; |
| tensor<fp16, [20, 77, 64]> key_states_cast = reshape(shape = var_2853, x = transpose_5)[name = tensor<string, []>("key_states_cast")]; |
| tensor<int32, [3]> var_2855 = const()[name = tensor<string, []>("op_2855"), val = tensor<int32, [3]>([20, -1, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_4 = transpose(perm = var_2840_perm_0, x = var_2839_cast)[name = tensor<string, []>("transpose_4")]; |
| tensor<fp16, [20, 77, 64]> value_states_cast = reshape(shape = var_2855, x = transpose_4)[name = tensor<string, []>("value_states_cast")]; |
| tensor<int32, [3]> var_2858_perm_0 = const()[name = tensor<string, []>("op_2858_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<bool, []> attn_weights_187_transpose_x_0 = const()[name = tensor<string, []>("attn_weights_187_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_187_transpose_y_0 = const()[name = tensor<string, []>("attn_weights_187_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 64, 77]> transpose_2 = transpose(perm = var_2858_perm_0, x = key_states_cast)[name = tensor<string, []>("transpose_2")]; |
| tensor<fp16, [20, 77, 77]> attn_weights_187_cast = matmul(transpose_x = attn_weights_187_transpose_x_0, transpose_y = attn_weights_187_transpose_y_0, x = query_states_cast, y = transpose_2)[name = tensor<string, []>("attn_weights_187_cast")]; |
| tensor<int32, [4]> var_2860 = const()[name = tensor<string, []>("op_2860"), val = tensor<int32, [4]>([1, 20, 77, 77])]; |
| tensor<fp16, [1, 20, 77, 77]> var_2861_cast = reshape(shape = var_2860, x = attn_weights_187_cast)[name = tensor<string, []>("op_2861_cast")]; |
| tensor<fp16, [1, 20, 77, 77]> attn_weights_189_cast = add(x = var_2861_cast, y = causal_attention_mask_to_fp16)[name = tensor<string, []>("attn_weights_189_cast")]; |
| tensor<int32, [3]> var_2866 = const()[name = tensor<string, []>("op_2866"), val = tensor<int32, [3]>([20, 77, 77])]; |
| tensor<fp16, [20, 77, 77]> input_501_cast = reshape(shape = var_2866, x = attn_weights_189_cast)[name = tensor<string, []>("input_501_cast")]; |
| tensor<fp16, [20, 77, 77]> input_503_cast = softmax(axis = var_5, x = input_501_cast)[name = tensor<string, []>("input_503_cast")]; |
| tensor<bool, []> attn_output_187_transpose_x_0 = const()[name = tensor<string, []>("attn_output_187_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_187_transpose_y_0 = const()[name = tensor<string, []>("attn_output_187_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [20, 77, 64]> attn_output_187_cast = matmul(transpose_x = attn_output_187_transpose_x_0, transpose_y = attn_output_187_transpose_y_0, x = input_503_cast, y = value_states_cast)[name = tensor<string, []>("attn_output_187_cast")]; |
| tensor<int32, [4]> var_2871 = const()[name = tensor<string, []>("op_2871"), val = tensor<int32, [4]>([1, 20, 77, 64])]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_189_cast = reshape(shape = var_2871, x = attn_output_187_cast)[name = tensor<string, []>("attn_output_189_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_2874 = const()[name = tensor<string, []>("op_2874"), val = tensor<int32, [3]>([1, 77, 1280])]; |
| tensor<fp16, [1, 77, 20, 64]> transpose_1 = transpose(perm = attn_output_perm_0, x = attn_output_189_cast)[name = tensor<string, []>("transpose_1")]; |
| tensor<fp16, [1, 77, 1280]> input_505_cast = reshape(shape = var_2874, x = transpose_1)[name = tensor<string, []>("input_505_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_model_encoder_layers_31_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1356570496)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1359847360)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_189_cast = linear(bias = text_encoder_text_model_encoder_layers_31_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_31_self_attn_out_proj_weight_to_fp16, x = input_505_cast)[name = tensor<string, []>("hidden_states_189_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_507_cast = add(x = input_499_cast, y = hidden_states_189_cast)[name = tensor<string, []>("input_507_cast")]; |
| tensor<int32, [1]> input_509_axes_0 = const()[name = tensor<string, []>("input_509_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1359849984)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1359852608)))]; |
| tensor<fp16, [1, 77, 1280]> input_509_cast = layer_norm(axes = input_509_axes_0, beta = text_encoder_text_model_encoder_layers_31_layer_norm2_bias_to_fp16, epsilon = var_12_to_fp16, gamma = text_encoder_text_model_encoder_layers_31_layer_norm2_weight_to_fp16, x = input_507_cast)[name = tensor<string, []>("input_509_cast")]; |
| tensor<fp16, [5120, 1280]> text_encoder_text_model_encoder_layers_31_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1359855232)))]; |
| tensor<fp16, [5120]> text_encoder_text_model_encoder_layers_31_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372962496)))]; |
| tensor<fp16, [1, 77, 5120]> input_511_cast = linear(bias = text_encoder_text_model_encoder_layers_31_mlp_fc1_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_31_mlp_fc1_weight_to_fp16, x = input_509_cast)[name = tensor<string, []>("input_511_cast")]; |
| tensor<string, []> input_513_mode_0 = const()[name = tensor<string, []>("input_513_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_513_cast = gelu(mode = input_513_mode_0, x = input_511_cast)[name = tensor<string, []>("input_513_cast")]; |
| tensor<fp16, [1280, 5120]> text_encoder_text_model_encoder_layers_31_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372972800)))]; |
| tensor<fp16, [1280]> text_encoder_text_model_encoder_layers_31_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_model_encoder_layers_31_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1386080064)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_cast = linear(bias = text_encoder_text_model_encoder_layers_31_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_31_mlp_fc2_weight_to_fp16, x = input_513_cast)[name = tensor<string, []>("hidden_states_cast")]; |
| tensor<fp16, [1, 77, 1280]> input_515_cast = add(x = input_507_cast, y = hidden_states_cast)[name = tensor<string, []>("input_515_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, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1386082688)))]; |
| tensor<fp16, [1280]> 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1386085312)))]; |
| tensor<fp16, [1, 77, 1280]> 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_12_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_515_cast)[name = tensor<string, []>("last_hidden_state_cast")]; |
| tensor<int32, [1]> var_2902 = const()[name = tensor<string, []>("op_2902"), val = tensor<int32, [1]>([0])]; |
| tensor<int32, [1]> var_2904 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_1322)[name = tensor<string, []>("op_2904")]; |
| 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_2902, var_2904))[name = tensor<string, []>("stack_0")]; |
| tensor<int32, []> input_transpose_batch_dims_0 = const()[name = tensor<string, []>("input_transpose_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp16, [1, 1280]> input_transpose_cast = gather_nd(batch_dims = input_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast)[name = tensor<string, []>("input_transpose_cast")]; |
| tensor<fp16, [1280, 1280]> text_encoder_text_projection_weight_to_fp16 = const()[name = tensor<string, []>("text_encoder_text_projection_weight_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1386087936)))]; |
| tensor<fp16, [1280]> var_2911_bias_0_to_fp16 = const()[name = tensor<string, []>("op_2911_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1389364800)))]; |
| tensor<fp16, [1, 1280]> var_2911_cast = linear(bias = var_2911_bias_0_to_fp16, weight = text_encoder_text_projection_weight_to_fp16, x = input_transpose_cast)[name = tensor<string, []>("op_2911_cast")]; |
| tensor<string, []> var_2911_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_2911_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, [1, 1280]> pooled_outputs = cast(dtype = var_2911_cast_to_fp32_dtype_0, x = var_2911_cast)[name = tensor<string, []>("cast_325")]; |
| tensor<fp32, [1, 77, 1280]> hidden_embeds = cast(dtype = input_499_cast_to_fp32_dtype_0, x = input_499_cast)[name = tensor<string, []>("cast_359")]; |
| } -> (hidden_embeds, pooled_outputs); |
| } |
