| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.1"}})] |
| { |
| 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, 1024]> 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, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| tensor<int32, [1, 77]> cast_2 = cast(dtype = cast_1_dtype_0, x = input_ids)[name = tensor<string, []>("cast_2")]; |
| tensor<fp16, [1, 77, 1024]> inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = cast_2, x = text_encoder_text_model_embeddings_token_embedding_weight_to_fp16)[name = tensor<string, []>("inputs_embeds_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> position_embeddings_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [59136]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101187712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101246912))), name = tensor<string, []>("position_embeddings_to_fp16_palettized"), shape = tensor<uint32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 1024]> input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16_palettized)[name = tensor<string, []>("input_3_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101247104)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101249216)))]; |
| tensor<fp16, []> var_15_to_fp16 = const()[name = tensor<string, []>("op_15_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101251328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102037824))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102038016)))]; |
| tensor<fp16, [1, 77, 1024]> linear_0_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; |
| tensor<fp16, []> var_129_to_fp16 = const()[name = tensor<string, []>("op_129_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_5_cast_fp16 = mul(x = linear_0_cast_fp16, y = var_129_to_fp16)[name = tensor<string, []>("tensor_5_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102040128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102826624))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102826816)))]; |
| tensor<fp16, [1, 77, 1024]> linear_1_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")]; |
| tensor<int32, [4]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_135_cast_fp16 = reshape(shape = var_134, x = linear_1_cast_fp16)[name = tensor<string, []>("op_135_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102828928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103615424))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103615616)))]; |
| tensor<fp16, [1, 77, 1024]> linear_2_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")]; |
| tensor<int32, [4]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_142_cast_fp16 = reshape(shape = var_141, x = linear_2_cast_fp16)[name = tensor<string, []>("op_142_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_151_cast_fp16 = reshape(shape = var_150, x = tensor_5_cast_fp16)[name = tensor<string, []>("op_151_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_91 = transpose(perm = var_152_perm_0, x = var_151_cast_fp16)[name = tensor<string, []>("transpose_91")]; |
| tensor<fp16, [16, 77, 64]> query_states_1_cast_fp16 = reshape(shape = var_154, x = transpose_91)[name = tensor<string, []>("query_states_1_cast_fp16")]; |
| tensor<int32, [3]> var_156 = const()[name = tensor<string, []>("op_156"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_90 = transpose(perm = var_136_perm_0, x = var_135_cast_fp16)[name = tensor<string, []>("transpose_90")]; |
| tensor<fp16, [16, 77, 64]> key_states_3_cast_fp16 = reshape(shape = var_156, x = transpose_90)[name = tensor<string, []>("key_states_3_cast_fp16")]; |
| tensor<int32, [3]> var_158 = const()[name = tensor<string, []>("op_158"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_89 = transpose(perm = var_143_perm_0, x = var_142_cast_fp16)[name = tensor<string, []>("transpose_89")]; |
| tensor<fp16, [16, 77, 64]> value_states_3_cast_fp16 = reshape(shape = var_158, x = transpose_89)[name = tensor<string, []>("value_states_3_cast_fp16")]; |
| tensor<bool, []> attn_weights_1_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_1_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_1_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_1_cast_fp16 = matmul(transpose_x = attn_weights_1_transpose_x_1, transpose_y = attn_weights_1_transpose_y_1, x = query_states_1_cast_fp16, y = key_states_3_cast_fp16)[name = tensor<string, []>("attn_weights_1_cast_fp16")]; |
| tensor<int32, [4]> var_163 = const()[name = tensor<string, []>("op_163"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_164_cast_fp16 = reshape(shape = var_163, x = attn_weights_1_cast_fp16)[name = tensor<string, []>("op_164_cast_fp16")]; |
| tensor<fp16, [1, 1, 77, 77]> var_57_to_fp16 = const()[name = tensor<string, []>("op_57_to_fp16"), val = tensor<fp16, [1, 1, 77, 77]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103617728)))]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_3_cast_fp16 = add(x = var_164_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_3_cast_fp16")]; |
| tensor<int32, [3]> var_169 = const()[name = tensor<string, []>("op_169"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_5_cast_fp16 = reshape(shape = var_169, x = attn_weights_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_7_cast_fp16 = softmax(axis = var_5, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = input_7_cast_fp16, y = value_states_3_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")]; |
| tensor<int32, [4]> var_174 = const()[name = tensor<string, []>("op_174"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_3_cast_fp16 = reshape(shape = var_174, x = attn_output_1_cast_fp16)[name = tensor<string, []>("attn_output_3_cast_fp16")]; |
| tensor<int32, [4]> attn_output_5_perm_0 = const()[name = tensor<string, []>("attn_output_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_177 = const()[name = tensor<string, []>("op_177"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_88 = transpose(perm = attn_output_5_perm_0, x = attn_output_3_cast_fp16)[name = tensor<string, []>("transpose_88")]; |
| tensor<fp16, [1, 77, 1024]> input_9_cast_fp16 = reshape(shape = var_177, x = transpose_88)[name = tensor<string, []>("input_9_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103629696))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104416192))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104416384)))]; |
| tensor<fp16, [1, 77, 1024]> linear_3_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_11_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; |
| tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104418496)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104420608)))]; |
| tensor<fp16, [1, 77, 1024]> input_13_cast_fp16 = layer_norm(axes = input_13_axes_0, beta = text_encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104422720))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107568512))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107568704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107571840))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_4_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; |
| tensor<string, []> input_17_mode_0 = const()[name = tensor<string, []>("input_17_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_17_cast_fp16 = gelu(mode = input_17_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107572032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110717824))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110718016)))]; |
| tensor<fp16, [1, 77, 1024]> linear_5_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_19_cast_fp16 = add(x = input_11_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110720128)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110722240)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110724352))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111510848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111511040)))]; |
| tensor<fp16, [1, 77, 1024]> linear_6_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")]; |
| tensor<fp16, []> var_216_to_fp16 = const()[name = tensor<string, []>("op_216_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_11_cast_fp16 = mul(x = linear_6_cast_fp16, y = var_216_to_fp16)[name = tensor<string, []>("tensor_11_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111513152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112299648))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112299840)))]; |
| tensor<fp16, [1, 77, 1024]> linear_7_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")]; |
| tensor<int32, [4]> var_221 = const()[name = tensor<string, []>("op_221"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_222_cast_fp16 = reshape(shape = var_221, x = linear_7_cast_fp16)[name = tensor<string, []>("op_222_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112301952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113088448))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113088640)))]; |
| tensor<fp16, [1, 77, 1024]> linear_8_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")]; |
| tensor<int32, [4]> var_228 = const()[name = tensor<string, []>("op_228"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_229_cast_fp16 = reshape(shape = var_228, x = linear_8_cast_fp16)[name = tensor<string, []>("op_229_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_238_cast_fp16 = reshape(shape = var_237, x = tensor_11_cast_fp16)[name = tensor<string, []>("op_238_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_87 = transpose(perm = var_239_perm_0, x = var_238_cast_fp16)[name = tensor<string, []>("transpose_87")]; |
| tensor<fp16, [16, 77, 64]> query_states_3_cast_fp16 = reshape(shape = var_241, x = transpose_87)[name = tensor<string, []>("query_states_3_cast_fp16")]; |
| tensor<int32, [3]> var_243 = const()[name = tensor<string, []>("op_243"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_86 = transpose(perm = var_223_perm_0, x = var_222_cast_fp16)[name = tensor<string, []>("transpose_86")]; |
| tensor<fp16, [16, 77, 64]> key_states_7_cast_fp16 = reshape(shape = var_243, x = transpose_86)[name = tensor<string, []>("key_states_7_cast_fp16")]; |
| tensor<int32, [3]> var_245 = const()[name = tensor<string, []>("op_245"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_85 = transpose(perm = var_230_perm_0, x = var_229_cast_fp16)[name = tensor<string, []>("transpose_85")]; |
| tensor<fp16, [16, 77, 64]> value_states_7_cast_fp16 = reshape(shape = var_245, x = transpose_85)[name = tensor<string, []>("value_states_7_cast_fp16")]; |
| tensor<bool, []> attn_weights_7_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_7_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_7_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_7_cast_fp16 = matmul(transpose_x = attn_weights_7_transpose_x_1, transpose_y = attn_weights_7_transpose_y_1, x = query_states_3_cast_fp16, y = key_states_7_cast_fp16)[name = tensor<string, []>("attn_weights_7_cast_fp16")]; |
| tensor<int32, [4]> var_250 = const()[name = tensor<string, []>("op_250"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_251_cast_fp16 = reshape(shape = var_250, x = attn_weights_7_cast_fp16)[name = tensor<string, []>("op_251_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_9_cast_fp16 = add(x = var_251_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_9_cast_fp16")]; |
| tensor<int32, [3]> var_256 = const()[name = tensor<string, []>("op_256"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_21_cast_fp16 = reshape(shape = var_256, x = attn_weights_9_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_23_cast_fp16 = softmax(axis = var_5, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; |
| tensor<bool, []> attn_output_7_transpose_x_0 = const()[name = tensor<string, []>("attn_output_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_7_transpose_y_0 = const()[name = tensor<string, []>("attn_output_7_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_7_cast_fp16 = matmul(transpose_x = attn_output_7_transpose_x_0, transpose_y = attn_output_7_transpose_y_0, x = input_23_cast_fp16, y = value_states_7_cast_fp16)[name = tensor<string, []>("attn_output_7_cast_fp16")]; |
| tensor<int32, [4]> var_261 = const()[name = tensor<string, []>("op_261"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_9_cast_fp16 = reshape(shape = var_261, x = attn_output_7_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")]; |
| tensor<int32, [4]> attn_output_11_perm_0 = const()[name = tensor<string, []>("attn_output_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_264 = const()[name = tensor<string, []>("op_264"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_84 = transpose(perm = attn_output_11_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_84")]; |
| tensor<fp16, [1, 77, 1024]> input_25_cast_fp16 = reshape(shape = var_264, x = transpose_84)[name = tensor<string, []>("input_25_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113090752))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113877248))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113877440)))]; |
| tensor<fp16, [1, 77, 1024]> linear_9_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; |
| tensor<int32, [1]> input_29_axes_0 = const()[name = tensor<string, []>("input_29_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113879552)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113881664)))]; |
| tensor<fp16, [1, 77, 1024]> input_29_cast_fp16 = layer_norm(axes = input_29_axes_0, beta = text_encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(113883776))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117029568))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117029760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117032896))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_10_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_29_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; |
| tensor<string, []> input_33_mode_0 = const()[name = tensor<string, []>("input_33_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_33_cast_fp16 = gelu(mode = input_33_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117033088))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120178880))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120179072)))]; |
| tensor<fp16, [1, 77, 1024]> linear_11_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_35_cast_fp16 = add(x = input_27_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120181184)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120183296)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120185408))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120971904))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120972096)))]; |
| tensor<fp16, [1, 77, 1024]> linear_12_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")]; |
| tensor<fp16, []> var_303_to_fp16 = const()[name = tensor<string, []>("op_303_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_17_cast_fp16 = mul(x = linear_12_cast_fp16, y = var_303_to_fp16)[name = tensor<string, []>("tensor_17_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120974208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121760704))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121760896)))]; |
| tensor<fp16, [1, 77, 1024]> linear_13_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")]; |
| tensor<int32, [4]> var_308 = const()[name = tensor<string, []>("op_308"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_309_cast_fp16 = reshape(shape = var_308, x = linear_13_cast_fp16)[name = tensor<string, []>("op_309_cast_fp16")]; |
| tensor<int32, [4]> var_310_perm_0 = const()[name = tensor<string, []>("op_310_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121763008))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122549504))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122549696)))]; |
| tensor<fp16, [1, 77, 1024]> linear_14_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")]; |
| tensor<int32, [4]> var_315 = const()[name = tensor<string, []>("op_315"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_316_cast_fp16 = reshape(shape = var_315, x = linear_14_cast_fp16)[name = tensor<string, []>("op_316_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_325_cast_fp16 = reshape(shape = var_324, x = tensor_17_cast_fp16)[name = tensor<string, []>("op_325_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_83 = transpose(perm = var_326_perm_0, x = var_325_cast_fp16)[name = tensor<string, []>("transpose_83")]; |
| tensor<fp16, [16, 77, 64]> query_states_5_cast_fp16 = reshape(shape = var_328, x = transpose_83)[name = tensor<string, []>("query_states_5_cast_fp16")]; |
| tensor<int32, [3]> var_330 = const()[name = tensor<string, []>("op_330"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_82 = transpose(perm = var_310_perm_0, x = var_309_cast_fp16)[name = tensor<string, []>("transpose_82")]; |
| tensor<fp16, [16, 77, 64]> key_states_11_cast_fp16 = reshape(shape = var_330, x = transpose_82)[name = tensor<string, []>("key_states_11_cast_fp16")]; |
| tensor<int32, [3]> var_332 = const()[name = tensor<string, []>("op_332"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_81 = transpose(perm = var_317_perm_0, x = var_316_cast_fp16)[name = tensor<string, []>("transpose_81")]; |
| tensor<fp16, [16, 77, 64]> value_states_11_cast_fp16 = reshape(shape = var_332, x = transpose_81)[name = tensor<string, []>("value_states_11_cast_fp16")]; |
| tensor<bool, []> attn_weights_13_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_13_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_13_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_13_cast_fp16 = matmul(transpose_x = attn_weights_13_transpose_x_1, transpose_y = attn_weights_13_transpose_y_1, x = query_states_5_cast_fp16, y = key_states_11_cast_fp16)[name = tensor<string, []>("attn_weights_13_cast_fp16")]; |
| tensor<int32, [4]> var_337 = const()[name = tensor<string, []>("op_337"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_338_cast_fp16 = reshape(shape = var_337, x = attn_weights_13_cast_fp16)[name = tensor<string, []>("op_338_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_15_cast_fp16 = add(x = var_338_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_15_cast_fp16")]; |
| tensor<int32, [3]> var_343 = const()[name = tensor<string, []>("op_343"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_37_cast_fp16 = reshape(shape = var_343, x = attn_weights_15_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_39_cast_fp16 = softmax(axis = var_5, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = input_39_cast_fp16, y = value_states_11_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")]; |
| tensor<int32, [4]> var_348 = const()[name = tensor<string, []>("op_348"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_15_cast_fp16 = reshape(shape = var_348, x = attn_output_13_cast_fp16)[name = tensor<string, []>("attn_output_15_cast_fp16")]; |
| tensor<int32, [4]> attn_output_17_perm_0 = const()[name = tensor<string, []>("attn_output_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_351 = const()[name = tensor<string, []>("op_351"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_80 = transpose(perm = attn_output_17_perm_0, x = attn_output_15_cast_fp16)[name = tensor<string, []>("transpose_80")]; |
| tensor<fp16, [1, 77, 1024]> input_41_cast_fp16 = reshape(shape = var_351, x = transpose_80)[name = tensor<string, []>("input_41_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(122551808))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123338304))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123338496)))]; |
| tensor<fp16, [1, 77, 1024]> linear_15_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_43_cast_fp16 = add(x = input_35_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; |
| tensor<int32, [1]> input_45_axes_0 = const()[name = tensor<string, []>("input_45_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123340608)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123342720)))]; |
| tensor<fp16, [1, 77, 1024]> input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = text_encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123344832))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126490624))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126490816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126493952))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_16_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; |
| tensor<string, []> input_49_mode_0 = const()[name = tensor<string, []>("input_49_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126494144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129639936))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129640128)))]; |
| tensor<fp16, [1, 77, 1024]> linear_17_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129642240)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129644352)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129646464))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130432960))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130433152)))]; |
| tensor<fp16, [1, 77, 1024]> linear_18_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")]; |
| tensor<fp16, []> var_390_to_fp16 = const()[name = tensor<string, []>("op_390_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_23_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_390_to_fp16)[name = tensor<string, []>("tensor_23_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130435264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131221760))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131221952)))]; |
| tensor<fp16, [1, 77, 1024]> linear_19_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")]; |
| tensor<int32, [4]> var_395 = const()[name = tensor<string, []>("op_395"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_396_cast_fp16 = reshape(shape = var_395, x = linear_19_cast_fp16)[name = tensor<string, []>("op_396_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131224064))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132010560))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132010752)))]; |
| tensor<fp16, [1, 77, 1024]> linear_20_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")]; |
| tensor<int32, [4]> var_402 = const()[name = tensor<string, []>("op_402"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_403_cast_fp16 = reshape(shape = var_402, x = linear_20_cast_fp16)[name = tensor<string, []>("op_403_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_412_cast_fp16 = reshape(shape = var_411, x = tensor_23_cast_fp16)[name = tensor<string, []>("op_412_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_79 = transpose(perm = var_413_perm_0, x = var_412_cast_fp16)[name = tensor<string, []>("transpose_79")]; |
| tensor<fp16, [16, 77, 64]> query_states_7_cast_fp16 = reshape(shape = var_415, x = transpose_79)[name = tensor<string, []>("query_states_7_cast_fp16")]; |
| tensor<int32, [3]> var_417 = const()[name = tensor<string, []>("op_417"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_78 = transpose(perm = var_397_perm_0, x = var_396_cast_fp16)[name = tensor<string, []>("transpose_78")]; |
| tensor<fp16, [16, 77, 64]> key_states_15_cast_fp16 = reshape(shape = var_417, x = transpose_78)[name = tensor<string, []>("key_states_15_cast_fp16")]; |
| tensor<int32, [3]> var_419 = const()[name = tensor<string, []>("op_419"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_77 = transpose(perm = var_404_perm_0, x = var_403_cast_fp16)[name = tensor<string, []>("transpose_77")]; |
| tensor<fp16, [16, 77, 64]> value_states_15_cast_fp16 = reshape(shape = var_419, x = transpose_77)[name = tensor<string, []>("value_states_15_cast_fp16")]; |
| tensor<bool, []> attn_weights_19_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_19_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_19_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_19_cast_fp16 = matmul(transpose_x = attn_weights_19_transpose_x_1, transpose_y = attn_weights_19_transpose_y_1, x = query_states_7_cast_fp16, y = key_states_15_cast_fp16)[name = tensor<string, []>("attn_weights_19_cast_fp16")]; |
| tensor<int32, [4]> var_424 = const()[name = tensor<string, []>("op_424"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_425_cast_fp16 = reshape(shape = var_424, x = attn_weights_19_cast_fp16)[name = tensor<string, []>("op_425_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_21_cast_fp16 = add(x = var_425_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_21_cast_fp16")]; |
| tensor<int32, [3]> var_430 = const()[name = tensor<string, []>("op_430"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_53_cast_fp16 = reshape(shape = var_430, x = attn_weights_21_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_55_cast_fp16 = softmax(axis = var_5, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; |
| tensor<bool, []> attn_output_19_transpose_x_0 = const()[name = tensor<string, []>("attn_output_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_19_transpose_y_0 = const()[name = tensor<string, []>("attn_output_19_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_19_cast_fp16 = matmul(transpose_x = attn_output_19_transpose_x_0, transpose_y = attn_output_19_transpose_y_0, x = input_55_cast_fp16, y = value_states_15_cast_fp16)[name = tensor<string, []>("attn_output_19_cast_fp16")]; |
| tensor<int32, [4]> var_435 = const()[name = tensor<string, []>("op_435"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_21_cast_fp16 = reshape(shape = var_435, x = attn_output_19_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")]; |
| tensor<int32, [4]> attn_output_23_perm_0 = const()[name = tensor<string, []>("attn_output_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_438 = const()[name = tensor<string, []>("op_438"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_76 = transpose(perm = attn_output_23_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_76")]; |
| tensor<fp16, [1, 77, 1024]> input_57_cast_fp16 = reshape(shape = var_438, x = transpose_76)[name = tensor<string, []>("input_57_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132012864))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132799360))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132799552)))]; |
| tensor<fp16, [1, 77, 1024]> linear_21_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_59_cast_fp16 = add(x = input_51_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; |
| tensor<int32, [1]> input_61_axes_0 = const()[name = tensor<string, []>("input_61_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132801664)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132803776)))]; |
| tensor<fp16, [1, 77, 1024]> input_61_cast_fp16 = layer_norm(axes = input_61_axes_0, beta = text_encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132805888))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135951680))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135951872))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135955008))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_22_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; |
| tensor<string, []> input_65_mode_0 = const()[name = tensor<string, []>("input_65_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_65_cast_fp16 = gelu(mode = input_65_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135955200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139100992))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139101184)))]; |
| tensor<fp16, [1, 77, 1024]> linear_23_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_67_cast_fp16 = add(x = input_59_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139103296)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139105408)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139107520))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139894016))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139894208)))]; |
| tensor<fp16, [1, 77, 1024]> linear_24_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")]; |
| tensor<fp16, []> var_477_to_fp16 = const()[name = tensor<string, []>("op_477_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_29_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_477_to_fp16)[name = tensor<string, []>("tensor_29_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139896320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140682816))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140683008)))]; |
| tensor<fp16, [1, 77, 1024]> linear_25_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")]; |
| tensor<int32, [4]> var_482 = const()[name = tensor<string, []>("op_482"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_483_cast_fp16 = reshape(shape = var_482, x = linear_25_cast_fp16)[name = tensor<string, []>("op_483_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140685120))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141471616))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141471808)))]; |
| tensor<fp16, [1, 77, 1024]> linear_26_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")]; |
| tensor<int32, [4]> var_489 = const()[name = tensor<string, []>("op_489"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_490_cast_fp16 = reshape(shape = var_489, x = linear_26_cast_fp16)[name = tensor<string, []>("op_490_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_499_cast_fp16 = reshape(shape = var_498, x = tensor_29_cast_fp16)[name = tensor<string, []>("op_499_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_75 = transpose(perm = var_500_perm_0, x = var_499_cast_fp16)[name = tensor<string, []>("transpose_75")]; |
| tensor<fp16, [16, 77, 64]> query_states_9_cast_fp16 = reshape(shape = var_502, x = transpose_75)[name = tensor<string, []>("query_states_9_cast_fp16")]; |
| tensor<int32, [3]> var_504 = const()[name = tensor<string, []>("op_504"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_74 = transpose(perm = var_484_perm_0, x = var_483_cast_fp16)[name = tensor<string, []>("transpose_74")]; |
| tensor<fp16, [16, 77, 64]> key_states_19_cast_fp16 = reshape(shape = var_504, x = transpose_74)[name = tensor<string, []>("key_states_19_cast_fp16")]; |
| tensor<int32, [3]> var_506 = const()[name = tensor<string, []>("op_506"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_73 = transpose(perm = var_491_perm_0, x = var_490_cast_fp16)[name = tensor<string, []>("transpose_73")]; |
| tensor<fp16, [16, 77, 64]> value_states_19_cast_fp16 = reshape(shape = var_506, x = transpose_73)[name = tensor<string, []>("value_states_19_cast_fp16")]; |
| tensor<bool, []> attn_weights_25_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_25_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_25_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_25_cast_fp16 = matmul(transpose_x = attn_weights_25_transpose_x_1, transpose_y = attn_weights_25_transpose_y_1, x = query_states_9_cast_fp16, y = key_states_19_cast_fp16)[name = tensor<string, []>("attn_weights_25_cast_fp16")]; |
| tensor<int32, [4]> var_511 = const()[name = tensor<string, []>("op_511"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_512_cast_fp16 = reshape(shape = var_511, x = attn_weights_25_cast_fp16)[name = tensor<string, []>("op_512_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_27_cast_fp16 = add(x = var_512_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_27_cast_fp16")]; |
| tensor<int32, [3]> var_517 = const()[name = tensor<string, []>("op_517"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_69_cast_fp16 = reshape(shape = var_517, x = attn_weights_27_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_71_cast_fp16 = softmax(axis = var_5, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = input_71_cast_fp16, y = value_states_19_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")]; |
| tensor<int32, [4]> var_522 = const()[name = tensor<string, []>("op_522"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_27_cast_fp16 = reshape(shape = var_522, x = attn_output_25_cast_fp16)[name = tensor<string, []>("attn_output_27_cast_fp16")]; |
| tensor<int32, [4]> attn_output_29_perm_0 = const()[name = tensor<string, []>("attn_output_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_525 = const()[name = tensor<string, []>("op_525"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_72 = transpose(perm = attn_output_29_perm_0, x = attn_output_27_cast_fp16)[name = tensor<string, []>("transpose_72")]; |
| tensor<fp16, [1, 77, 1024]> input_73_cast_fp16 = reshape(shape = var_525, x = transpose_72)[name = tensor<string, []>("input_73_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141473920))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142260416))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142260608)))]; |
| tensor<fp16, [1, 77, 1024]> linear_27_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142262720)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142264832)))]; |
| tensor<fp16, [1, 77, 1024]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = text_encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142266944))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145412736))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145412928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145416064))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_28_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")]; |
| tensor<string, []> input_81_mode_0 = const()[name = tensor<string, []>("input_81_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_81_cast_fp16 = gelu(mode = input_81_mode_0, x = linear_28_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145416256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148562048))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148562240)))]; |
| tensor<fp16, [1, 77, 1024]> linear_29_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_83_cast_fp16 = add(x = input_75_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148564352)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148566464)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("hidden_states_31_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148568576))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149355072))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149355264)))]; |
| tensor<fp16, [1, 77, 1024]> linear_30_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")]; |
| tensor<fp16, []> var_564_to_fp16 = const()[name = tensor<string, []>("op_564_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_35_cast_fp16 = mul(x = linear_30_cast_fp16, y = var_564_to_fp16)[name = tensor<string, []>("tensor_35_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149357376))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150143872))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150144064)))]; |
| tensor<fp16, [1, 77, 1024]> linear_31_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")]; |
| tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_570_cast_fp16 = reshape(shape = var_569, x = linear_31_cast_fp16)[name = tensor<string, []>("op_570_cast_fp16")]; |
| tensor<int32, [4]> var_571_perm_0 = const()[name = tensor<string, []>("op_571_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150146176))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150932672))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150932864)))]; |
| tensor<fp16, [1, 77, 1024]> linear_32_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")]; |
| tensor<int32, [4]> var_576 = const()[name = tensor<string, []>("op_576"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_577_cast_fp16 = reshape(shape = var_576, x = linear_32_cast_fp16)[name = tensor<string, []>("op_577_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_586_cast_fp16 = reshape(shape = var_585, x = tensor_35_cast_fp16)[name = tensor<string, []>("op_586_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_71 = transpose(perm = var_587_perm_0, x = var_586_cast_fp16)[name = tensor<string, []>("transpose_71")]; |
| tensor<fp16, [16, 77, 64]> query_states_11_cast_fp16 = reshape(shape = var_589, x = transpose_71)[name = tensor<string, []>("query_states_11_cast_fp16")]; |
| tensor<int32, [3]> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_70 = transpose(perm = var_571_perm_0, x = var_570_cast_fp16)[name = tensor<string, []>("transpose_70")]; |
| tensor<fp16, [16, 77, 64]> key_states_23_cast_fp16 = reshape(shape = var_591, x = transpose_70)[name = tensor<string, []>("key_states_23_cast_fp16")]; |
| tensor<int32, [3]> var_593 = const()[name = tensor<string, []>("op_593"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_69 = transpose(perm = var_578_perm_0, x = var_577_cast_fp16)[name = tensor<string, []>("transpose_69")]; |
| tensor<fp16, [16, 77, 64]> value_states_23_cast_fp16 = reshape(shape = var_593, x = transpose_69)[name = tensor<string, []>("value_states_23_cast_fp16")]; |
| tensor<bool, []> attn_weights_31_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_31_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_31_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_31_cast_fp16 = matmul(transpose_x = attn_weights_31_transpose_x_1, transpose_y = attn_weights_31_transpose_y_1, x = query_states_11_cast_fp16, y = key_states_23_cast_fp16)[name = tensor<string, []>("attn_weights_31_cast_fp16")]; |
| tensor<int32, [4]> var_598 = const()[name = tensor<string, []>("op_598"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_599_cast_fp16 = reshape(shape = var_598, x = attn_weights_31_cast_fp16)[name = tensor<string, []>("op_599_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_33_cast_fp16 = add(x = var_599_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_33_cast_fp16")]; |
| tensor<int32, [3]> var_604 = const()[name = tensor<string, []>("op_604"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_85_cast_fp16 = reshape(shape = var_604, x = attn_weights_33_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_87_cast_fp16 = softmax(axis = var_5, x = input_85_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; |
| tensor<bool, []> attn_output_31_transpose_x_0 = const()[name = tensor<string, []>("attn_output_31_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_31_transpose_y_0 = const()[name = tensor<string, []>("attn_output_31_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_31_cast_fp16 = matmul(transpose_x = attn_output_31_transpose_x_0, transpose_y = attn_output_31_transpose_y_0, x = input_87_cast_fp16, y = value_states_23_cast_fp16)[name = tensor<string, []>("attn_output_31_cast_fp16")]; |
| tensor<int32, [4]> var_609 = const()[name = tensor<string, []>("op_609"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_33_cast_fp16 = reshape(shape = var_609, x = attn_output_31_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")]; |
| tensor<int32, [4]> attn_output_35_perm_0 = const()[name = tensor<string, []>("attn_output_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_612 = const()[name = tensor<string, []>("op_612"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_68 = transpose(perm = attn_output_35_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_68")]; |
| tensor<fp16, [1, 77, 1024]> input_89_cast_fp16 = reshape(shape = var_612, x = transpose_68)[name = tensor<string, []>("input_89_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150934976))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151721472))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151721664)))]; |
| tensor<fp16, [1, 77, 1024]> linear_33_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; |
| tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151723776)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151725888)))]; |
| tensor<fp16, [1, 77, 1024]> input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = text_encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151728000))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154873792))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154873984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154877120))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_34_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")]; |
| tensor<string, []> input_97_mode_0 = const()[name = tensor<string, []>("input_97_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_97_cast_fp16 = gelu(mode = input_97_mode_0, x = linear_34_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(154877312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158023104))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158023296)))]; |
| tensor<fp16, [1, 77, 1024]> linear_35_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158025408)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158027520)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("hidden_states_37_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158029632))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158816128))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158816320)))]; |
| tensor<fp16, [1, 77, 1024]> linear_36_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")]; |
| tensor<fp16, []> var_651_to_fp16 = const()[name = tensor<string, []>("op_651_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_41_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_651_to_fp16)[name = tensor<string, []>("tensor_41_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158818432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159604928))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159605120)))]; |
| tensor<fp16, [1, 77, 1024]> linear_37_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")]; |
| tensor<int32, [4]> var_656 = const()[name = tensor<string, []>("op_656"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_657_cast_fp16 = reshape(shape = var_656, x = linear_37_cast_fp16)[name = tensor<string, []>("op_657_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159607232))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160393728))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160393920)))]; |
| tensor<fp16, [1, 77, 1024]> linear_38_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")]; |
| tensor<int32, [4]> var_663 = const()[name = tensor<string, []>("op_663"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_664_cast_fp16 = reshape(shape = var_663, x = linear_38_cast_fp16)[name = tensor<string, []>("op_664_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_673_cast_fp16 = reshape(shape = var_672, x = tensor_41_cast_fp16)[name = tensor<string, []>("op_673_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_67 = transpose(perm = var_674_perm_0, x = var_673_cast_fp16)[name = tensor<string, []>("transpose_67")]; |
| tensor<fp16, [16, 77, 64]> query_states_13_cast_fp16 = reshape(shape = var_676, x = transpose_67)[name = tensor<string, []>("query_states_13_cast_fp16")]; |
| tensor<int32, [3]> var_678 = const()[name = tensor<string, []>("op_678"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_66 = transpose(perm = var_658_perm_0, x = var_657_cast_fp16)[name = tensor<string, []>("transpose_66")]; |
| tensor<fp16, [16, 77, 64]> key_states_27_cast_fp16 = reshape(shape = var_678, x = transpose_66)[name = tensor<string, []>("key_states_27_cast_fp16")]; |
| tensor<int32, [3]> var_680 = const()[name = tensor<string, []>("op_680"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_65 = transpose(perm = var_665_perm_0, x = var_664_cast_fp16)[name = tensor<string, []>("transpose_65")]; |
| tensor<fp16, [16, 77, 64]> value_states_27_cast_fp16 = reshape(shape = var_680, x = transpose_65)[name = tensor<string, []>("value_states_27_cast_fp16")]; |
| tensor<bool, []> attn_weights_37_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_37_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_37_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_37_cast_fp16 = matmul(transpose_x = attn_weights_37_transpose_x_1, transpose_y = attn_weights_37_transpose_y_1, x = query_states_13_cast_fp16, y = key_states_27_cast_fp16)[name = tensor<string, []>("attn_weights_37_cast_fp16")]; |
| tensor<int32, [4]> var_685 = const()[name = tensor<string, []>("op_685"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_686_cast_fp16 = reshape(shape = var_685, x = attn_weights_37_cast_fp16)[name = tensor<string, []>("op_686_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_39_cast_fp16 = add(x = var_686_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_39_cast_fp16")]; |
| tensor<int32, [3]> var_691 = const()[name = tensor<string, []>("op_691"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_101_cast_fp16 = reshape(shape = var_691, x = attn_weights_39_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_103_cast_fp16 = softmax(axis = var_5, x = input_101_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = input_103_cast_fp16, y = value_states_27_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")]; |
| tensor<int32, [4]> var_696 = const()[name = tensor<string, []>("op_696"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_39_cast_fp16 = reshape(shape = var_696, x = attn_output_37_cast_fp16)[name = tensor<string, []>("attn_output_39_cast_fp16")]; |
| tensor<int32, [4]> attn_output_41_perm_0 = const()[name = tensor<string, []>("attn_output_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_699 = const()[name = tensor<string, []>("op_699"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_64 = transpose(perm = attn_output_41_perm_0, x = attn_output_39_cast_fp16)[name = tensor<string, []>("transpose_64")]; |
| tensor<fp16, [1, 77, 1024]> input_105_cast_fp16 = reshape(shape = var_699, x = transpose_64)[name = tensor<string, []>("input_105_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160396032))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161182528))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161182720)))]; |
| tensor<fp16, [1, 77, 1024]> linear_39_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_107_cast_fp16 = add(x = input_99_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; |
| tensor<int32, [1]> input_109_axes_0 = const()[name = tensor<string, []>("input_109_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161184832)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161186944)))]; |
| tensor<fp16, [1, 77, 1024]> input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = text_encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_107_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161189056))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164334848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164335040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164338176))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_40_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")]; |
| tensor<string, []> input_113_mode_0 = const()[name = tensor<string, []>("input_113_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_113_cast_fp16 = gelu(mode = input_113_mode_0, x = linear_40_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164338368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167484160))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167484352)))]; |
| tensor<fp16, [1, 77, 1024]> linear_41_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_115_cast_fp16 = add(x = input_107_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167486464)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167488576)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167490688))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168277184))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168277376)))]; |
| tensor<fp16, [1, 77, 1024]> linear_42_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")]; |
| tensor<fp16, []> var_738_to_fp16 = const()[name = tensor<string, []>("op_738_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_47_cast_fp16 = mul(x = linear_42_cast_fp16, y = var_738_to_fp16)[name = tensor<string, []>("tensor_47_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(168279488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169065984))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169066176)))]; |
| tensor<fp16, [1, 77, 1024]> linear_43_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")]; |
| tensor<int32, [4]> var_743 = const()[name = tensor<string, []>("op_743"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_744_cast_fp16 = reshape(shape = var_743, x = linear_43_cast_fp16)[name = tensor<string, []>("op_744_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169068288))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169854784))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169854976)))]; |
| tensor<fp16, [1, 77, 1024]> linear_44_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")]; |
| tensor<int32, [4]> var_750 = const()[name = tensor<string, []>("op_750"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_751_cast_fp16 = reshape(shape = var_750, x = linear_44_cast_fp16)[name = tensor<string, []>("op_751_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_760_cast_fp16 = reshape(shape = var_759, x = tensor_47_cast_fp16)[name = tensor<string, []>("op_760_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_63 = transpose(perm = var_761_perm_0, x = var_760_cast_fp16)[name = tensor<string, []>("transpose_63")]; |
| tensor<fp16, [16, 77, 64]> query_states_15_cast_fp16 = reshape(shape = var_763, x = transpose_63)[name = tensor<string, []>("query_states_15_cast_fp16")]; |
| tensor<int32, [3]> var_765 = const()[name = tensor<string, []>("op_765"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_62 = transpose(perm = var_745_perm_0, x = var_744_cast_fp16)[name = tensor<string, []>("transpose_62")]; |
| tensor<fp16, [16, 77, 64]> key_states_31_cast_fp16 = reshape(shape = var_765, x = transpose_62)[name = tensor<string, []>("key_states_31_cast_fp16")]; |
| tensor<int32, [3]> var_767 = const()[name = tensor<string, []>("op_767"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_61 = transpose(perm = var_752_perm_0, x = var_751_cast_fp16)[name = tensor<string, []>("transpose_61")]; |
| tensor<fp16, [16, 77, 64]> value_states_31_cast_fp16 = reshape(shape = var_767, x = transpose_61)[name = tensor<string, []>("value_states_31_cast_fp16")]; |
| tensor<bool, []> attn_weights_43_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_43_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_43_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_43_cast_fp16 = matmul(transpose_x = attn_weights_43_transpose_x_1, transpose_y = attn_weights_43_transpose_y_1, x = query_states_15_cast_fp16, y = key_states_31_cast_fp16)[name = tensor<string, []>("attn_weights_43_cast_fp16")]; |
| tensor<int32, [4]> var_772 = const()[name = tensor<string, []>("op_772"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_773_cast_fp16 = reshape(shape = var_772, x = attn_weights_43_cast_fp16)[name = tensor<string, []>("op_773_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_45_cast_fp16 = add(x = var_773_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_45_cast_fp16")]; |
| tensor<int32, [3]> var_778 = const()[name = tensor<string, []>("op_778"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_117_cast_fp16 = reshape(shape = var_778, x = attn_weights_45_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_119_cast_fp16 = softmax(axis = var_5, x = input_117_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")]; |
| tensor<bool, []> attn_output_43_transpose_x_0 = const()[name = tensor<string, []>("attn_output_43_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_43_transpose_y_0 = const()[name = tensor<string, []>("attn_output_43_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_43_cast_fp16 = matmul(transpose_x = attn_output_43_transpose_x_0, transpose_y = attn_output_43_transpose_y_0, x = input_119_cast_fp16, y = value_states_31_cast_fp16)[name = tensor<string, []>("attn_output_43_cast_fp16")]; |
| tensor<int32, [4]> var_783 = const()[name = tensor<string, []>("op_783"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_45_cast_fp16 = reshape(shape = var_783, x = attn_output_43_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")]; |
| tensor<int32, [4]> attn_output_47_perm_0 = const()[name = tensor<string, []>("attn_output_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_786 = const()[name = tensor<string, []>("op_786"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_60 = transpose(perm = attn_output_47_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_60")]; |
| tensor<fp16, [1, 77, 1024]> input_121_cast_fp16 = reshape(shape = var_786, x = transpose_60)[name = tensor<string, []>("input_121_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169857088))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170643584))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170643776)))]; |
| tensor<fp16, [1, 77, 1024]> linear_45_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")]; |
| tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170645888)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170648000)))]; |
| tensor<fp16, [1, 77, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = text_encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170650112))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173795904))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173796096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173799232))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_46_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")]; |
| tensor<string, []> input_129_mode_0 = const()[name = tensor<string, []>("input_129_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_129_cast_fp16 = gelu(mode = input_129_mode_0, x = linear_46_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173799424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176945216))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176945408)))]; |
| tensor<fp16, [1, 77, 1024]> linear_47_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_131_cast_fp16 = add(x = input_123_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176947520)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176949632)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_131_cast_fp16)[name = tensor<string, []>("hidden_states_49_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176951744))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177738240))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177738432)))]; |
| tensor<fp16, [1, 77, 1024]> linear_48_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")]; |
| tensor<fp16, []> var_825_to_fp16 = const()[name = tensor<string, []>("op_825_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_53_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_825_to_fp16)[name = tensor<string, []>("tensor_53_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177740544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178527040))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178527232)))]; |
| tensor<fp16, [1, 77, 1024]> linear_49_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")]; |
| tensor<int32, [4]> var_830 = const()[name = tensor<string, []>("op_830"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_831_cast_fp16 = reshape(shape = var_830, x = linear_49_cast_fp16)[name = tensor<string, []>("op_831_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178529344))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179315840))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179316032)))]; |
| tensor<fp16, [1, 77, 1024]> linear_50_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")]; |
| tensor<int32, [4]> var_837 = const()[name = tensor<string, []>("op_837"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_838_cast_fp16 = reshape(shape = var_837, x = linear_50_cast_fp16)[name = tensor<string, []>("op_838_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_847_cast_fp16 = reshape(shape = var_846, x = tensor_53_cast_fp16)[name = tensor<string, []>("op_847_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_59 = transpose(perm = var_848_perm_0, x = var_847_cast_fp16)[name = tensor<string, []>("transpose_59")]; |
| tensor<fp16, [16, 77, 64]> query_states_17_cast_fp16 = reshape(shape = var_850, x = transpose_59)[name = tensor<string, []>("query_states_17_cast_fp16")]; |
| tensor<int32, [3]> var_852 = const()[name = tensor<string, []>("op_852"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_58 = transpose(perm = var_832_perm_0, x = var_831_cast_fp16)[name = tensor<string, []>("transpose_58")]; |
| tensor<fp16, [16, 77, 64]> key_states_35_cast_fp16 = reshape(shape = var_852, x = transpose_58)[name = tensor<string, []>("key_states_35_cast_fp16")]; |
| tensor<int32, [3]> var_854 = const()[name = tensor<string, []>("op_854"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_57 = transpose(perm = var_839_perm_0, x = var_838_cast_fp16)[name = tensor<string, []>("transpose_57")]; |
| tensor<fp16, [16, 77, 64]> value_states_35_cast_fp16 = reshape(shape = var_854, x = transpose_57)[name = tensor<string, []>("value_states_35_cast_fp16")]; |
| tensor<bool, []> attn_weights_49_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_49_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_49_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_49_cast_fp16 = matmul(transpose_x = attn_weights_49_transpose_x_1, transpose_y = attn_weights_49_transpose_y_1, x = query_states_17_cast_fp16, y = key_states_35_cast_fp16)[name = tensor<string, []>("attn_weights_49_cast_fp16")]; |
| tensor<int32, [4]> var_859 = const()[name = tensor<string, []>("op_859"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_860_cast_fp16 = reshape(shape = var_859, x = attn_weights_49_cast_fp16)[name = tensor<string, []>("op_860_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_51_cast_fp16 = add(x = var_860_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_51_cast_fp16")]; |
| tensor<int32, [3]> var_865 = const()[name = tensor<string, []>("op_865"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_133_cast_fp16 = reshape(shape = var_865, x = attn_weights_51_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_135_cast_fp16 = softmax(axis = var_5, x = input_133_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")]; |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = input_135_cast_fp16, y = value_states_35_cast_fp16)[name = tensor<string, []>("attn_output_49_cast_fp16")]; |
| tensor<int32, [4]> var_870 = const()[name = tensor<string, []>("op_870"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_51_cast_fp16 = reshape(shape = var_870, x = attn_output_49_cast_fp16)[name = tensor<string, []>("attn_output_51_cast_fp16")]; |
| tensor<int32, [4]> attn_output_53_perm_0 = const()[name = tensor<string, []>("attn_output_53_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_873 = const()[name = tensor<string, []>("op_873"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_56 = transpose(perm = attn_output_53_perm_0, x = attn_output_51_cast_fp16)[name = tensor<string, []>("transpose_56")]; |
| tensor<fp16, [1, 77, 1024]> input_137_cast_fp16 = reshape(shape = var_873, x = transpose_56)[name = tensor<string, []>("input_137_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(179318144))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180104640))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180104832)))]; |
| tensor<fp16, [1, 77, 1024]> linear_51_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = input_137_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_139_cast_fp16 = add(x = input_131_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")]; |
| tensor<int32, [1]> input_141_axes_0 = const()[name = tensor<string, []>("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180106944)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180109056)))]; |
| tensor<fp16, [1, 77, 1024]> input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = text_encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("input_141_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180111168))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183256960))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183257152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183260288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_52_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")]; |
| tensor<string, []> input_145_mode_0 = const()[name = tensor<string, []>("input_145_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_145_cast_fp16 = gelu(mode = input_145_mode_0, x = linear_52_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183260480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186406272))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186406464)))]; |
| tensor<fp16, [1, 77, 1024]> linear_53_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186408576)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186410688)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("hidden_states_55_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186412800))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187199296))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187199488)))]; |
| tensor<fp16, [1, 77, 1024]> linear_54_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")]; |
| tensor<fp16, []> var_912_to_fp16 = const()[name = tensor<string, []>("op_912_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_59_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_912_to_fp16)[name = tensor<string, []>("tensor_59_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187201600))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187988096))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187988288)))]; |
| tensor<fp16, [1, 77, 1024]> linear_55_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")]; |
| tensor<int32, [4]> var_917 = const()[name = tensor<string, []>("op_917"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_918_cast_fp16 = reshape(shape = var_917, x = linear_55_cast_fp16)[name = tensor<string, []>("op_918_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187990400))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188776896))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188777088)))]; |
| tensor<fp16, [1, 77, 1024]> linear_56_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")]; |
| tensor<int32, [4]> var_924 = const()[name = tensor<string, []>("op_924"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_925_cast_fp16 = reshape(shape = var_924, x = linear_56_cast_fp16)[name = tensor<string, []>("op_925_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_934_cast_fp16 = reshape(shape = var_933, x = tensor_59_cast_fp16)[name = tensor<string, []>("op_934_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_55 = transpose(perm = var_935_perm_0, x = var_934_cast_fp16)[name = tensor<string, []>("transpose_55")]; |
| tensor<fp16, [16, 77, 64]> query_states_19_cast_fp16 = reshape(shape = var_937, x = transpose_55)[name = tensor<string, []>("query_states_19_cast_fp16")]; |
| tensor<int32, [3]> var_939 = const()[name = tensor<string, []>("op_939"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_54 = transpose(perm = var_919_perm_0, x = var_918_cast_fp16)[name = tensor<string, []>("transpose_54")]; |
| tensor<fp16, [16, 77, 64]> key_states_39_cast_fp16 = reshape(shape = var_939, x = transpose_54)[name = tensor<string, []>("key_states_39_cast_fp16")]; |
| tensor<int32, [3]> var_941 = const()[name = tensor<string, []>("op_941"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_53 = transpose(perm = var_926_perm_0, x = var_925_cast_fp16)[name = tensor<string, []>("transpose_53")]; |
| tensor<fp16, [16, 77, 64]> value_states_39_cast_fp16 = reshape(shape = var_941, x = transpose_53)[name = tensor<string, []>("value_states_39_cast_fp16")]; |
| tensor<bool, []> attn_weights_55_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_55_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_55_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_55_cast_fp16 = matmul(transpose_x = attn_weights_55_transpose_x_1, transpose_y = attn_weights_55_transpose_y_1, x = query_states_19_cast_fp16, y = key_states_39_cast_fp16)[name = tensor<string, []>("attn_weights_55_cast_fp16")]; |
| tensor<int32, [4]> var_946 = const()[name = tensor<string, []>("op_946"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_947_cast_fp16 = reshape(shape = var_946, x = attn_weights_55_cast_fp16)[name = tensor<string, []>("op_947_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_57_cast_fp16 = add(x = var_947_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_57_cast_fp16")]; |
| tensor<int32, [3]> var_952 = const()[name = tensor<string, []>("op_952"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_149_cast_fp16 = reshape(shape = var_952, x = attn_weights_57_cast_fp16)[name = tensor<string, []>("input_149_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_151_cast_fp16 = softmax(axis = var_5, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")]; |
| tensor<bool, []> attn_output_55_transpose_x_0 = const()[name = tensor<string, []>("attn_output_55_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_55_transpose_y_0 = const()[name = tensor<string, []>("attn_output_55_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_55_cast_fp16 = matmul(transpose_x = attn_output_55_transpose_x_0, transpose_y = attn_output_55_transpose_y_0, x = input_151_cast_fp16, y = value_states_39_cast_fp16)[name = tensor<string, []>("attn_output_55_cast_fp16")]; |
| tensor<int32, [4]> var_957 = const()[name = tensor<string, []>("op_957"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_57_cast_fp16 = reshape(shape = var_957, x = attn_output_55_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")]; |
| tensor<int32, [4]> attn_output_59_perm_0 = const()[name = tensor<string, []>("attn_output_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_960 = const()[name = tensor<string, []>("op_960"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_52 = transpose(perm = attn_output_59_perm_0, x = attn_output_57_cast_fp16)[name = tensor<string, []>("transpose_52")]; |
| tensor<fp16, [1, 77, 1024]> input_153_cast_fp16 = reshape(shape = var_960, x = transpose_52)[name = tensor<string, []>("input_153_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188779200))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189565696))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189565888)))]; |
| tensor<fp16, [1, 77, 1024]> linear_57_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_155_cast_fp16 = add(x = input_147_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")]; |
| tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189568000)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189570112)))]; |
| tensor<fp16, [1, 77, 1024]> input_157_cast_fp16 = layer_norm(axes = input_157_axes_0, beta = text_encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189572224))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192718016))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192718208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192721344))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_58_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")]; |
| tensor<string, []> input_161_mode_0 = const()[name = tensor<string, []>("input_161_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_161_cast_fp16 = gelu(mode = input_161_mode_0, x = linear_58_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192721536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195867328))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195867520)))]; |
| tensor<fp16, [1, 77, 1024]> linear_59_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_161_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_163_cast_fp16 = add(x = input_155_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195869632)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195871744)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("hidden_states_61_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195873856))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196660352))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196660544)))]; |
| tensor<fp16, [1, 77, 1024]> linear_60_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")]; |
| tensor<fp16, []> var_999_to_fp16 = const()[name = tensor<string, []>("op_999_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_65_cast_fp16 = mul(x = linear_60_cast_fp16, y = var_999_to_fp16)[name = tensor<string, []>("tensor_65_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196662656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197449152))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197449344)))]; |
| tensor<fp16, [1, 77, 1024]> linear_61_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")]; |
| tensor<int32, [4]> var_1004 = const()[name = tensor<string, []>("op_1004"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1005_cast_fp16 = reshape(shape = var_1004, x = linear_61_cast_fp16)[name = tensor<string, []>("op_1005_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197451456))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198237952))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198238144)))]; |
| tensor<fp16, [1, 77, 1024]> linear_62_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")]; |
| tensor<int32, [4]> var_1011 = const()[name = tensor<string, []>("op_1011"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1012_cast_fp16 = reshape(shape = var_1011, x = linear_62_cast_fp16)[name = tensor<string, []>("op_1012_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1021_cast_fp16 = reshape(shape = var_1020, x = tensor_65_cast_fp16)[name = tensor<string, []>("op_1021_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_51 = transpose(perm = var_1022_perm_0, x = var_1021_cast_fp16)[name = tensor<string, []>("transpose_51")]; |
| tensor<fp16, [16, 77, 64]> query_states_21_cast_fp16 = reshape(shape = var_1024, x = transpose_51)[name = tensor<string, []>("query_states_21_cast_fp16")]; |
| tensor<int32, [3]> var_1026 = const()[name = tensor<string, []>("op_1026"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_50 = transpose(perm = var_1006_perm_0, x = var_1005_cast_fp16)[name = tensor<string, []>("transpose_50")]; |
| tensor<fp16, [16, 77, 64]> key_states_43_cast_fp16 = reshape(shape = var_1026, x = transpose_50)[name = tensor<string, []>("key_states_43_cast_fp16")]; |
| tensor<int32, [3]> var_1028 = const()[name = tensor<string, []>("op_1028"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_49 = transpose(perm = var_1013_perm_0, x = var_1012_cast_fp16)[name = tensor<string, []>("transpose_49")]; |
| tensor<fp16, [16, 77, 64]> value_states_43_cast_fp16 = reshape(shape = var_1028, x = transpose_49)[name = tensor<string, []>("value_states_43_cast_fp16")]; |
| tensor<bool, []> attn_weights_61_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_61_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_61_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_61_cast_fp16 = matmul(transpose_x = attn_weights_61_transpose_x_1, transpose_y = attn_weights_61_transpose_y_1, x = query_states_21_cast_fp16, y = key_states_43_cast_fp16)[name = tensor<string, []>("attn_weights_61_cast_fp16")]; |
| tensor<int32, [4]> var_1033 = const()[name = tensor<string, []>("op_1033"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1034_cast_fp16 = reshape(shape = var_1033, x = attn_weights_61_cast_fp16)[name = tensor<string, []>("op_1034_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_63_cast_fp16 = add(x = var_1034_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_63_cast_fp16")]; |
| tensor<int32, [3]> var_1039 = const()[name = tensor<string, []>("op_1039"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_165_cast_fp16 = reshape(shape = var_1039, x = attn_weights_63_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_167_cast_fp16 = softmax(axis = var_5, x = input_165_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")]; |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = input_167_cast_fp16, y = value_states_43_cast_fp16)[name = tensor<string, []>("attn_output_61_cast_fp16")]; |
| tensor<int32, [4]> var_1044 = const()[name = tensor<string, []>("op_1044"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_63_cast_fp16 = reshape(shape = var_1044, x = attn_output_61_cast_fp16)[name = tensor<string, []>("attn_output_63_cast_fp16")]; |
| tensor<int32, [4]> attn_output_65_perm_0 = const()[name = tensor<string, []>("attn_output_65_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1047 = const()[name = tensor<string, []>("op_1047"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_48 = transpose(perm = attn_output_65_perm_0, x = attn_output_63_cast_fp16)[name = tensor<string, []>("transpose_48")]; |
| tensor<fp16, [1, 77, 1024]> input_169_cast_fp16 = reshape(shape = var_1047, x = transpose_48)[name = tensor<string, []>("input_169_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198240256))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199026752))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199026944)))]; |
| tensor<fp16, [1, 77, 1024]> linear_63_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")]; |
| tensor<int32, [1]> input_173_axes_0 = const()[name = tensor<string, []>("input_173_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199029056)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199031168)))]; |
| tensor<fp16, [1, 77, 1024]> input_173_cast_fp16 = layer_norm(axes = input_173_axes_0, beta = text_encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199033280))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202179072))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202179264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202182400))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_64_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")]; |
| tensor<string, []> input_177_mode_0 = const()[name = tensor<string, []>("input_177_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_177_cast_fp16 = gelu(mode = input_177_mode_0, x = linear_64_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202182592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205328384))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205328576)))]; |
| tensor<fp16, [1, 77, 1024]> linear_65_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_179_cast_fp16 = add(x = input_171_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205330688)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205332800)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_179_cast_fp16)[name = tensor<string, []>("hidden_states_67_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205334912))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206121408))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206121600)))]; |
| tensor<fp16, [1, 77, 1024]> linear_66_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")]; |
| tensor<fp16, []> var_1086_to_fp16 = const()[name = tensor<string, []>("op_1086_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_71_cast_fp16 = mul(x = linear_66_cast_fp16, y = var_1086_to_fp16)[name = tensor<string, []>("tensor_71_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206123712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206910208))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206910400)))]; |
| tensor<fp16, [1, 77, 1024]> linear_67_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")]; |
| tensor<int32, [4]> var_1091 = const()[name = tensor<string, []>("op_1091"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1092_cast_fp16 = reshape(shape = var_1091, x = linear_67_cast_fp16)[name = tensor<string, []>("op_1092_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206912512))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207699008))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207699200)))]; |
| tensor<fp16, [1, 77, 1024]> linear_68_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")]; |
| tensor<int32, [4]> var_1098 = const()[name = tensor<string, []>("op_1098"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1099_cast_fp16 = reshape(shape = var_1098, x = linear_68_cast_fp16)[name = tensor<string, []>("op_1099_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1108_cast_fp16 = reshape(shape = var_1107, x = tensor_71_cast_fp16)[name = tensor<string, []>("op_1108_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_47 = transpose(perm = var_1109_perm_0, x = var_1108_cast_fp16)[name = tensor<string, []>("transpose_47")]; |
| tensor<fp16, [16, 77, 64]> query_states_23_cast_fp16 = reshape(shape = var_1111, x = transpose_47)[name = tensor<string, []>("query_states_23_cast_fp16")]; |
| tensor<int32, [3]> var_1113 = const()[name = tensor<string, []>("op_1113"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_46 = transpose(perm = var_1093_perm_0, x = var_1092_cast_fp16)[name = tensor<string, []>("transpose_46")]; |
| tensor<fp16, [16, 77, 64]> key_states_47_cast_fp16 = reshape(shape = var_1113, x = transpose_46)[name = tensor<string, []>("key_states_47_cast_fp16")]; |
| tensor<int32, [3]> var_1115 = const()[name = tensor<string, []>("op_1115"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_45 = transpose(perm = var_1100_perm_0, x = var_1099_cast_fp16)[name = tensor<string, []>("transpose_45")]; |
| tensor<fp16, [16, 77, 64]> value_states_47_cast_fp16 = reshape(shape = var_1115, x = transpose_45)[name = tensor<string, []>("value_states_47_cast_fp16")]; |
| tensor<bool, []> attn_weights_67_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_67_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_67_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_67_cast_fp16 = matmul(transpose_x = attn_weights_67_transpose_x_1, transpose_y = attn_weights_67_transpose_y_1, x = query_states_23_cast_fp16, y = key_states_47_cast_fp16)[name = tensor<string, []>("attn_weights_67_cast_fp16")]; |
| tensor<int32, [4]> var_1120 = const()[name = tensor<string, []>("op_1120"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1121_cast_fp16 = reshape(shape = var_1120, x = attn_weights_67_cast_fp16)[name = tensor<string, []>("op_1121_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_69_cast_fp16 = add(x = var_1121_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_69_cast_fp16")]; |
| tensor<int32, [3]> var_1126 = const()[name = tensor<string, []>("op_1126"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_181_cast_fp16 = reshape(shape = var_1126, x = attn_weights_69_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_183_cast_fp16 = softmax(axis = var_5, x = input_181_cast_fp16)[name = tensor<string, []>("input_183_cast_fp16")]; |
| tensor<bool, []> attn_output_67_transpose_x_0 = const()[name = tensor<string, []>("attn_output_67_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_67_transpose_y_0 = const()[name = tensor<string, []>("attn_output_67_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [16, 77, 64]> attn_output_67_cast_fp16 = matmul(transpose_x = attn_output_67_transpose_x_0, transpose_y = attn_output_67_transpose_y_0, x = input_183_cast_fp16, y = value_states_47_cast_fp16)[name = tensor<string, []>("attn_output_67_cast_fp16")]; |
| tensor<int32, [4]> var_1131 = const()[name = tensor<string, []>("op_1131"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_69_cast_fp16 = reshape(shape = var_1131, x = attn_output_67_cast_fp16)[name = tensor<string, []>("attn_output_69_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_44 = transpose(perm = attn_output_71_perm_0, x = attn_output_69_cast_fp16)[name = tensor<string, []>("transpose_44")]; |
| tensor<fp16, [1, 77, 1024]> input_185_cast_fp16 = reshape(shape = var_1134, x = transpose_44)[name = tensor<string, []>("input_185_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207701312))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208487808))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208488000)))]; |
| tensor<fp16, [1, 77, 1024]> linear_69_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = input_185_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_187_cast_fp16 = add(x = input_179_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")]; |
| tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208490112)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208492224)))]; |
| tensor<fp16, [1, 77, 1024]> input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = text_encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208494336))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211640128))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211640320))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211643456))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_70_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")]; |
| tensor<string, []> input_193_mode_0 = const()[name = tensor<string, []>("input_193_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_193_cast_fp16 = gelu(mode = input_193_mode_0, x = linear_70_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211643648))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214789440))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214789632)))]; |
| tensor<fp16, [1, 77, 1024]> linear_71_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = text_encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_73_axes_0 = const()[name = tensor<string, []>("hidden_states_73_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214791744)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214793856)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_73_cast_fp16 = layer_norm(axes = hidden_states_73_axes_0, beta = text_encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("hidden_states_73_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214795968))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215582464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215582656)))]; |
| tensor<fp16, [1, 77, 1024]> linear_72_cast_fp16 = 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_palettized, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")]; |
| tensor<fp16, []> var_1173_to_fp16 = const()[name = tensor<string, []>("op_1173_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_77_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1173_to_fp16)[name = tensor<string, []>("tensor_77_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215584768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216371264))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216371456)))]; |
| tensor<fp16, [1, 77, 1024]> linear_73_cast_fp16 = 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_palettized, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")]; |
| tensor<int32, [4]> var_1178 = const()[name = tensor<string, []>("op_1178"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1179_cast_fp16 = reshape(shape = var_1178, x = linear_73_cast_fp16)[name = tensor<string, []>("op_1179_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216373568))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217160064))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217160256)))]; |
| tensor<fp16, [1, 77, 1024]> linear_74_cast_fp16 = 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_palettized, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")]; |
| tensor<int32, [4]> var_1185 = const()[name = tensor<string, []>("op_1185"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1186_cast_fp16 = reshape(shape = var_1185, x = linear_74_cast_fp16)[name = tensor<string, []>("op_1186_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1195_cast_fp16 = reshape(shape = var_1194, x = tensor_77_cast_fp16)[name = tensor<string, []>("op_1195_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_43 = transpose(perm = var_1196_perm_0, x = var_1195_cast_fp16)[name = tensor<string, []>("transpose_43")]; |
| tensor<fp16, [16, 77, 64]> query_states_25_cast_fp16 = reshape(shape = var_1198, x = transpose_43)[name = tensor<string, []>("query_states_25_cast_fp16")]; |
| tensor<int32, [3]> var_1200 = const()[name = tensor<string, []>("op_1200"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_42 = transpose(perm = var_1180_perm_0, x = var_1179_cast_fp16)[name = tensor<string, []>("transpose_42")]; |
| tensor<fp16, [16, 77, 64]> key_states_51_cast_fp16 = reshape(shape = var_1200, x = transpose_42)[name = tensor<string, []>("key_states_51_cast_fp16")]; |
| tensor<int32, [3]> var_1202 = const()[name = tensor<string, []>("op_1202"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_41 = transpose(perm = var_1187_perm_0, x = var_1186_cast_fp16)[name = tensor<string, []>("transpose_41")]; |
| tensor<fp16, [16, 77, 64]> value_states_51_cast_fp16 = reshape(shape = var_1202, x = transpose_41)[name = tensor<string, []>("value_states_51_cast_fp16")]; |
| tensor<bool, []> attn_weights_73_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_73_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_73_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_73_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_73_cast_fp16 = matmul(transpose_x = attn_weights_73_transpose_x_1, transpose_y = attn_weights_73_transpose_y_1, x = query_states_25_cast_fp16, y = key_states_51_cast_fp16)[name = tensor<string, []>("attn_weights_73_cast_fp16")]; |
| tensor<int32, [4]> var_1207 = const()[name = tensor<string, []>("op_1207"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1208_cast_fp16 = reshape(shape = var_1207, x = attn_weights_73_cast_fp16)[name = tensor<string, []>("op_1208_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_75_cast_fp16 = add(x = var_1208_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_75_cast_fp16")]; |
| tensor<int32, [3]> var_1213 = const()[name = tensor<string, []>("op_1213"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_197_cast_fp16 = reshape(shape = var_1213, x = attn_weights_75_cast_fp16)[name = tensor<string, []>("input_197_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_199_cast_fp16 = softmax(axis = var_5, x = input_197_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_73_cast_fp16 = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = input_199_cast_fp16, y = value_states_51_cast_fp16)[name = tensor<string, []>("attn_output_73_cast_fp16")]; |
| tensor<int32, [4]> var_1218 = const()[name = tensor<string, []>("op_1218"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_75_cast_fp16 = reshape(shape = var_1218, x = attn_output_73_cast_fp16)[name = tensor<string, []>("attn_output_75_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_40 = transpose(perm = attn_output_77_perm_0, x = attn_output_75_cast_fp16)[name = tensor<string, []>("transpose_40")]; |
| tensor<fp16, [1, 77, 1024]> input_201_cast_fp16 = reshape(shape = var_1221, x = transpose_40)[name = tensor<string, []>("input_201_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217162368))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217948864))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217949056)))]; |
| tensor<fp16, [1, 77, 1024]> linear_75_cast_fp16 = 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_palettized, x = input_201_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_203_cast_fp16 = add(x = input_195_cast_fp16, y = linear_75_cast_fp16)[name = tensor<string, []>("input_203_cast_fp16")]; |
| tensor<int32, [1]> input_205_axes_0 = const()[name = tensor<string, []>("input_205_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217951168)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217953280)))]; |
| tensor<fp16, [1, 77, 1024]> input_205_cast_fp16 = layer_norm(axes = input_205_axes_0, beta = text_encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16, x = input_203_cast_fp16)[name = tensor<string, []>("input_205_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217955392))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221101184))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221101376))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221104512))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_76_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")]; |
| tensor<string, []> input_209_mode_0 = const()[name = tensor<string, []>("input_209_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_209_cast_fp16 = gelu(mode = input_209_mode_0, x = linear_76_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221104704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224250496))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224250688)))]; |
| tensor<fp16, [1, 77, 1024]> linear_77_cast_fp16 = 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_palettized, x = input_209_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_211_cast_fp16 = add(x = input_203_cast_fp16, y = linear_77_cast_fp16)[name = tensor<string, []>("input_211_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_79_axes_0 = const()[name = tensor<string, []>("hidden_states_79_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224252800)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224254912)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_79_cast_fp16 = layer_norm(axes = hidden_states_79_axes_0, beta = text_encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16, x = input_211_cast_fp16)[name = tensor<string, []>("hidden_states_79_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224257024))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225043520))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225043712)))]; |
| tensor<fp16, [1, 77, 1024]> linear_78_cast_fp16 = 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_palettized, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")]; |
| tensor<fp16, []> var_1260_to_fp16 = const()[name = tensor<string, []>("op_1260_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_83_cast_fp16 = mul(x = linear_78_cast_fp16, y = var_1260_to_fp16)[name = tensor<string, []>("tensor_83_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225045824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225832320))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225832512)))]; |
| tensor<fp16, [1, 77, 1024]> linear_79_cast_fp16 = 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_palettized, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")]; |
| tensor<int32, [4]> var_1265 = const()[name = tensor<string, []>("op_1265"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1266_cast_fp16 = reshape(shape = var_1265, x = linear_79_cast_fp16)[name = tensor<string, []>("op_1266_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225834624))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226621120))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226621312)))]; |
| tensor<fp16, [1, 77, 1024]> linear_80_cast_fp16 = 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_palettized, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")]; |
| tensor<int32, [4]> var_1272 = const()[name = tensor<string, []>("op_1272"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1273_cast_fp16 = reshape(shape = var_1272, x = linear_80_cast_fp16)[name = tensor<string, []>("op_1273_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1282_cast_fp16 = reshape(shape = var_1281, x = tensor_83_cast_fp16)[name = tensor<string, []>("op_1282_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_39 = transpose(perm = var_1283_perm_0, x = var_1282_cast_fp16)[name = tensor<string, []>("transpose_39")]; |
| tensor<fp16, [16, 77, 64]> query_states_27_cast_fp16 = reshape(shape = var_1285, x = transpose_39)[name = tensor<string, []>("query_states_27_cast_fp16")]; |
| tensor<int32, [3]> var_1287 = const()[name = tensor<string, []>("op_1287"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_38 = transpose(perm = var_1267_perm_0, x = var_1266_cast_fp16)[name = tensor<string, []>("transpose_38")]; |
| tensor<fp16, [16, 77, 64]> key_states_55_cast_fp16 = reshape(shape = var_1287, x = transpose_38)[name = tensor<string, []>("key_states_55_cast_fp16")]; |
| tensor<int32, [3]> var_1289 = const()[name = tensor<string, []>("op_1289"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_37 = transpose(perm = var_1274_perm_0, x = var_1273_cast_fp16)[name = tensor<string, []>("transpose_37")]; |
| tensor<fp16, [16, 77, 64]> value_states_55_cast_fp16 = reshape(shape = var_1289, x = transpose_37)[name = tensor<string, []>("value_states_55_cast_fp16")]; |
| tensor<bool, []> attn_weights_79_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_79_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_79_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_79_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_79_cast_fp16 = matmul(transpose_x = attn_weights_79_transpose_x_1, transpose_y = attn_weights_79_transpose_y_1, x = query_states_27_cast_fp16, y = key_states_55_cast_fp16)[name = tensor<string, []>("attn_weights_79_cast_fp16")]; |
| tensor<int32, [4]> var_1294 = const()[name = tensor<string, []>("op_1294"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1295_cast_fp16 = reshape(shape = var_1294, x = attn_weights_79_cast_fp16)[name = tensor<string, []>("op_1295_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_81_cast_fp16 = add(x = var_1295_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_81_cast_fp16")]; |
| tensor<int32, [3]> var_1300 = const()[name = tensor<string, []>("op_1300"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_213_cast_fp16 = reshape(shape = var_1300, x = attn_weights_81_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_215_cast_fp16 = softmax(axis = var_5, x = input_213_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_79_cast_fp16 = matmul(transpose_x = attn_output_79_transpose_x_0, transpose_y = attn_output_79_transpose_y_0, x = input_215_cast_fp16, y = value_states_55_cast_fp16)[name = tensor<string, []>("attn_output_79_cast_fp16")]; |
| tensor<int32, [4]> var_1305 = const()[name = tensor<string, []>("op_1305"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_81_cast_fp16 = reshape(shape = var_1305, x = attn_output_79_cast_fp16)[name = tensor<string, []>("attn_output_81_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_36 = transpose(perm = attn_output_83_perm_0, x = attn_output_81_cast_fp16)[name = tensor<string, []>("transpose_36")]; |
| tensor<fp16, [1, 77, 1024]> input_217_cast_fp16 = reshape(shape = var_1308, x = transpose_36)[name = tensor<string, []>("input_217_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226623424))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227409920))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227410112)))]; |
| tensor<fp16, [1, 77, 1024]> linear_81_cast_fp16 = 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_palettized, x = input_217_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_219_cast_fp16 = add(x = input_211_cast_fp16, y = linear_81_cast_fp16)[name = tensor<string, []>("input_219_cast_fp16")]; |
| tensor<int32, [1]> input_221_axes_0 = const()[name = tensor<string, []>("input_221_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227412224)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227414336)))]; |
| tensor<fp16, [1, 77, 1024]> input_221_cast_fp16 = layer_norm(axes = input_221_axes_0, beta = text_encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16, x = input_219_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227416448))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(230562240))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(230562432))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(230565568))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_82_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16_palettized, x = input_221_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")]; |
| tensor<string, []> input_225_mode_0 = const()[name = tensor<string, []>("input_225_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_225_cast_fp16 = gelu(mode = input_225_mode_0, x = linear_82_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(230565760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233711552))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233711744)))]; |
| tensor<fp16, [1, 77, 1024]> linear_83_cast_fp16 = 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_palettized, x = input_225_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_227_cast_fp16 = add(x = input_219_cast_fp16, y = linear_83_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_85_axes_0 = const()[name = tensor<string, []>("hidden_states_85_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233713856)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233715968)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = text_encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16, x = input_227_cast_fp16)[name = tensor<string, []>("hidden_states_85_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233718080))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234504576))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234504768)))]; |
| tensor<fp16, [1, 77, 1024]> linear_84_cast_fp16 = 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_palettized, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")]; |
| tensor<fp16, []> var_1347_to_fp16 = const()[name = tensor<string, []>("op_1347_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_89_cast_fp16 = mul(x = linear_84_cast_fp16, y = var_1347_to_fp16)[name = tensor<string, []>("tensor_89_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(234506880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235293376))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235293568)))]; |
| tensor<fp16, [1, 77, 1024]> linear_85_cast_fp16 = 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_palettized, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")]; |
| tensor<int32, [4]> var_1352 = const()[name = tensor<string, []>("op_1352"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1353_cast_fp16 = reshape(shape = var_1352, x = linear_85_cast_fp16)[name = tensor<string, []>("op_1353_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235295680))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236082176))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236082368)))]; |
| tensor<fp16, [1, 77, 1024]> linear_86_cast_fp16 = 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_palettized, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")]; |
| tensor<int32, [4]> var_1359 = const()[name = tensor<string, []>("op_1359"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1360_cast_fp16 = reshape(shape = var_1359, x = linear_86_cast_fp16)[name = tensor<string, []>("op_1360_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1369_cast_fp16 = reshape(shape = var_1368, x = tensor_89_cast_fp16)[name = tensor<string, []>("op_1369_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_35 = transpose(perm = var_1370_perm_0, x = var_1369_cast_fp16)[name = tensor<string, []>("transpose_35")]; |
| tensor<fp16, [16, 77, 64]> query_states_29_cast_fp16 = reshape(shape = var_1372, x = transpose_35)[name = tensor<string, []>("query_states_29_cast_fp16")]; |
| tensor<int32, [3]> var_1374 = const()[name = tensor<string, []>("op_1374"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_34 = transpose(perm = var_1354_perm_0, x = var_1353_cast_fp16)[name = tensor<string, []>("transpose_34")]; |
| tensor<fp16, [16, 77, 64]> key_states_59_cast_fp16 = reshape(shape = var_1374, x = transpose_34)[name = tensor<string, []>("key_states_59_cast_fp16")]; |
| tensor<int32, [3]> var_1376 = const()[name = tensor<string, []>("op_1376"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_33 = transpose(perm = var_1361_perm_0, x = var_1360_cast_fp16)[name = tensor<string, []>("transpose_33")]; |
| tensor<fp16, [16, 77, 64]> value_states_59_cast_fp16 = reshape(shape = var_1376, x = transpose_33)[name = tensor<string, []>("value_states_59_cast_fp16")]; |
| tensor<bool, []> attn_weights_85_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_85_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_85_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_85_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_85_cast_fp16 = matmul(transpose_x = attn_weights_85_transpose_x_1, transpose_y = attn_weights_85_transpose_y_1, x = query_states_29_cast_fp16, y = key_states_59_cast_fp16)[name = tensor<string, []>("attn_weights_85_cast_fp16")]; |
| tensor<int32, [4]> var_1381 = const()[name = tensor<string, []>("op_1381"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1382_cast_fp16 = reshape(shape = var_1381, x = attn_weights_85_cast_fp16)[name = tensor<string, []>("op_1382_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_87_cast_fp16 = add(x = var_1382_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_87_cast_fp16")]; |
| tensor<int32, [3]> var_1387 = const()[name = tensor<string, []>("op_1387"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_229_cast_fp16 = reshape(shape = var_1387, x = attn_weights_87_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_231_cast_fp16 = softmax(axis = var_5, x = input_229_cast_fp16)[name = tensor<string, []>("input_231_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_85_cast_fp16 = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = input_231_cast_fp16, y = value_states_59_cast_fp16)[name = tensor<string, []>("attn_output_85_cast_fp16")]; |
| tensor<int32, [4]> var_1392 = const()[name = tensor<string, []>("op_1392"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_87_cast_fp16 = reshape(shape = var_1392, x = attn_output_85_cast_fp16)[name = tensor<string, []>("attn_output_87_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_32 = transpose(perm = attn_output_89_perm_0, x = attn_output_87_cast_fp16)[name = tensor<string, []>("transpose_32")]; |
| tensor<fp16, [1, 77, 1024]> input_233_cast_fp16 = reshape(shape = var_1395, x = transpose_32)[name = tensor<string, []>("input_233_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236084480))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236870976))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236871168)))]; |
| tensor<fp16, [1, 77, 1024]> linear_87_cast_fp16 = 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_palettized, x = input_233_cast_fp16)[name = tensor<string, []>("linear_87_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_235_cast_fp16 = add(x = input_227_cast_fp16, y = linear_87_cast_fp16)[name = tensor<string, []>("input_235_cast_fp16")]; |
| tensor<int32, [1]> input_237_axes_0 = const()[name = tensor<string, []>("input_237_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236873280)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236875392)))]; |
| tensor<fp16, [1, 77, 1024]> input_237_cast_fp16 = layer_norm(axes = input_237_axes_0, beta = text_encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16, x = input_235_cast_fp16)[name = tensor<string, []>("input_237_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(236877504))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(240023296))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(240023488))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(240026624))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_88_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16_palettized, x = input_237_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")]; |
| tensor<string, []> input_241_mode_0 = const()[name = tensor<string, []>("input_241_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_241_cast_fp16 = gelu(mode = input_241_mode_0, x = linear_88_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(240026816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243172608))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243172800)))]; |
| tensor<fp16, [1, 77, 1024]> linear_89_cast_fp16 = 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_palettized, x = input_241_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_243_cast_fp16 = add(x = input_235_cast_fp16, y = linear_89_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_91_axes_0 = const()[name = tensor<string, []>("hidden_states_91_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243174912)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243177024)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_91_cast_fp16 = layer_norm(axes = hidden_states_91_axes_0, beta = text_encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16, x = input_243_cast_fp16)[name = tensor<string, []>("hidden_states_91_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243179136))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243965632))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243965824)))]; |
| tensor<fp16, [1, 77, 1024]> linear_90_cast_fp16 = 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_palettized, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")]; |
| tensor<fp16, []> var_1434_to_fp16 = const()[name = tensor<string, []>("op_1434_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_95_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_1434_to_fp16)[name = tensor<string, []>("tensor_95_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243967936))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244754432))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244754624)))]; |
| tensor<fp16, [1, 77, 1024]> linear_91_cast_fp16 = 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_palettized, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")]; |
| tensor<int32, [4]> var_1439 = const()[name = tensor<string, []>("op_1439"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1440_cast_fp16 = reshape(shape = var_1439, x = linear_91_cast_fp16)[name = tensor<string, []>("op_1440_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244756736))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245543232))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245543424)))]; |
| tensor<fp16, [1, 77, 1024]> linear_92_cast_fp16 = 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_palettized, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")]; |
| tensor<int32, [4]> var_1446 = const()[name = tensor<string, []>("op_1446"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1447_cast_fp16 = reshape(shape = var_1446, x = linear_92_cast_fp16)[name = tensor<string, []>("op_1447_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1456_cast_fp16 = reshape(shape = var_1455, x = tensor_95_cast_fp16)[name = tensor<string, []>("op_1456_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_31 = transpose(perm = var_1457_perm_0, x = var_1456_cast_fp16)[name = tensor<string, []>("transpose_31")]; |
| tensor<fp16, [16, 77, 64]> query_states_31_cast_fp16 = reshape(shape = var_1459, x = transpose_31)[name = tensor<string, []>("query_states_31_cast_fp16")]; |
| tensor<int32, [3]> var_1461 = const()[name = tensor<string, []>("op_1461"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_30 = transpose(perm = var_1441_perm_0, x = var_1440_cast_fp16)[name = tensor<string, []>("transpose_30")]; |
| tensor<fp16, [16, 77, 64]> key_states_63_cast_fp16 = reshape(shape = var_1461, x = transpose_30)[name = tensor<string, []>("key_states_63_cast_fp16")]; |
| tensor<int32, [3]> var_1463 = const()[name = tensor<string, []>("op_1463"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_29 = transpose(perm = var_1448_perm_0, x = var_1447_cast_fp16)[name = tensor<string, []>("transpose_29")]; |
| tensor<fp16, [16, 77, 64]> value_states_63_cast_fp16 = reshape(shape = var_1463, x = transpose_29)[name = tensor<string, []>("value_states_63_cast_fp16")]; |
| tensor<bool, []> attn_weights_91_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_91_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_91_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_91_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_91_cast_fp16 = matmul(transpose_x = attn_weights_91_transpose_x_1, transpose_y = attn_weights_91_transpose_y_1, x = query_states_31_cast_fp16, y = key_states_63_cast_fp16)[name = tensor<string, []>("attn_weights_91_cast_fp16")]; |
| tensor<int32, [4]> var_1468 = const()[name = tensor<string, []>("op_1468"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1469_cast_fp16 = reshape(shape = var_1468, x = attn_weights_91_cast_fp16)[name = tensor<string, []>("op_1469_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_93_cast_fp16 = add(x = var_1469_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_93_cast_fp16")]; |
| tensor<int32, [3]> var_1474 = const()[name = tensor<string, []>("op_1474"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_245_cast_fp16 = reshape(shape = var_1474, x = attn_weights_93_cast_fp16)[name = tensor<string, []>("input_245_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_247_cast_fp16 = softmax(axis = var_5, x = input_245_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_91_cast_fp16 = matmul(transpose_x = attn_output_91_transpose_x_0, transpose_y = attn_output_91_transpose_y_0, x = input_247_cast_fp16, y = value_states_63_cast_fp16)[name = tensor<string, []>("attn_output_91_cast_fp16")]; |
| tensor<int32, [4]> var_1479 = const()[name = tensor<string, []>("op_1479"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_93_cast_fp16 = reshape(shape = var_1479, x = attn_output_91_cast_fp16)[name = tensor<string, []>("attn_output_93_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_28 = transpose(perm = attn_output_95_perm_0, x = attn_output_93_cast_fp16)[name = tensor<string, []>("transpose_28")]; |
| tensor<fp16, [1, 77, 1024]> input_249_cast_fp16 = reshape(shape = var_1482, x = transpose_28)[name = tensor<string, []>("input_249_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245545536))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246332032))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246332224)))]; |
| tensor<fp16, [1, 77, 1024]> linear_93_cast_fp16 = 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_palettized, x = input_249_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_251_cast_fp16 = add(x = input_243_cast_fp16, y = linear_93_cast_fp16)[name = tensor<string, []>("input_251_cast_fp16")]; |
| tensor<int32, [1]> input_253_axes_0 = const()[name = tensor<string, []>("input_253_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246334336)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246336448)))]; |
| tensor<fp16, [1, 77, 1024]> input_253_cast_fp16 = layer_norm(axes = input_253_axes_0, beta = text_encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16, x = input_251_cast_fp16)[name = tensor<string, []>("input_253_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246338560))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249484352))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249484544))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249487680))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_94_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16_palettized, x = input_253_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")]; |
| tensor<string, []> input_257_mode_0 = const()[name = tensor<string, []>("input_257_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_257_cast_fp16 = gelu(mode = input_257_mode_0, x = linear_94_cast_fp16)[name = tensor<string, []>("input_257_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249487872))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252633664))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252633856)))]; |
| tensor<fp16, [1, 77, 1024]> linear_95_cast_fp16 = 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_palettized, x = input_257_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_259_cast_fp16 = add(x = input_251_cast_fp16, y = linear_95_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_97_axes_0 = const()[name = tensor<string, []>("hidden_states_97_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252635968)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252638080)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_97_cast_fp16 = layer_norm(axes = hidden_states_97_axes_0, beta = text_encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16, x = input_259_cast_fp16)[name = tensor<string, []>("hidden_states_97_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252640192))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253426688))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253426880)))]; |
| tensor<fp16, [1, 77, 1024]> linear_96_cast_fp16 = 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_palettized, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")]; |
| tensor<fp16, []> var_1521_to_fp16 = const()[name = tensor<string, []>("op_1521_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_101_cast_fp16 = mul(x = linear_96_cast_fp16, y = var_1521_to_fp16)[name = tensor<string, []>("tensor_101_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253428992))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254215488))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254215680)))]; |
| tensor<fp16, [1, 77, 1024]> linear_97_cast_fp16 = 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_palettized, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")]; |
| tensor<int32, [4]> var_1526 = const()[name = tensor<string, []>("op_1526"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1527_cast_fp16 = reshape(shape = var_1526, x = linear_97_cast_fp16)[name = tensor<string, []>("op_1527_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254217792))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255004288))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255004480)))]; |
| tensor<fp16, [1, 77, 1024]> linear_98_cast_fp16 = 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_palettized, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")]; |
| tensor<int32, [4]> var_1533 = const()[name = tensor<string, []>("op_1533"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1534_cast_fp16 = reshape(shape = var_1533, x = linear_98_cast_fp16)[name = tensor<string, []>("op_1534_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1543_cast_fp16 = reshape(shape = var_1542, x = tensor_101_cast_fp16)[name = tensor<string, []>("op_1543_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_27 = transpose(perm = var_1544_perm_0, x = var_1543_cast_fp16)[name = tensor<string, []>("transpose_27")]; |
| tensor<fp16, [16, 77, 64]> query_states_33_cast_fp16 = reshape(shape = var_1546, x = transpose_27)[name = tensor<string, []>("query_states_33_cast_fp16")]; |
| tensor<int32, [3]> var_1548 = const()[name = tensor<string, []>("op_1548"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_26 = transpose(perm = var_1528_perm_0, x = var_1527_cast_fp16)[name = tensor<string, []>("transpose_26")]; |
| tensor<fp16, [16, 77, 64]> key_states_67_cast_fp16 = reshape(shape = var_1548, x = transpose_26)[name = tensor<string, []>("key_states_67_cast_fp16")]; |
| tensor<int32, [3]> var_1550 = const()[name = tensor<string, []>("op_1550"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_25 = transpose(perm = var_1535_perm_0, x = var_1534_cast_fp16)[name = tensor<string, []>("transpose_25")]; |
| tensor<fp16, [16, 77, 64]> value_states_67_cast_fp16 = reshape(shape = var_1550, x = transpose_25)[name = tensor<string, []>("value_states_67_cast_fp16")]; |
| tensor<bool, []> attn_weights_97_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_97_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_97_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_97_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_97_cast_fp16 = matmul(transpose_x = attn_weights_97_transpose_x_1, transpose_y = attn_weights_97_transpose_y_1, x = query_states_33_cast_fp16, y = key_states_67_cast_fp16)[name = tensor<string, []>("attn_weights_97_cast_fp16")]; |
| tensor<int32, [4]> var_1555 = const()[name = tensor<string, []>("op_1555"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1556_cast_fp16 = reshape(shape = var_1555, x = attn_weights_97_cast_fp16)[name = tensor<string, []>("op_1556_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_99_cast_fp16 = add(x = var_1556_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_99_cast_fp16")]; |
| tensor<int32, [3]> var_1561 = const()[name = tensor<string, []>("op_1561"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_261_cast_fp16 = reshape(shape = var_1561, x = attn_weights_99_cast_fp16)[name = tensor<string, []>("input_261_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_263_cast_fp16 = softmax(axis = var_5, x = input_261_cast_fp16)[name = tensor<string, []>("input_263_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_97_cast_fp16 = matmul(transpose_x = attn_output_97_transpose_x_0, transpose_y = attn_output_97_transpose_y_0, x = input_263_cast_fp16, y = value_states_67_cast_fp16)[name = tensor<string, []>("attn_output_97_cast_fp16")]; |
| tensor<int32, [4]> var_1566 = const()[name = tensor<string, []>("op_1566"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_99_cast_fp16 = reshape(shape = var_1566, x = attn_output_97_cast_fp16)[name = tensor<string, []>("attn_output_99_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_24 = transpose(perm = attn_output_101_perm_0, x = attn_output_99_cast_fp16)[name = tensor<string, []>("transpose_24")]; |
| tensor<fp16, [1, 77, 1024]> input_265_cast_fp16 = reshape(shape = var_1569, x = transpose_24)[name = tensor<string, []>("input_265_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255006592))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255793088))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255793280)))]; |
| tensor<fp16, [1, 77, 1024]> linear_99_cast_fp16 = 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_palettized, x = input_265_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_267_cast_fp16 = add(x = input_259_cast_fp16, y = linear_99_cast_fp16)[name = tensor<string, []>("input_267_cast_fp16")]; |
| tensor<int32, [1]> input_269_axes_0 = const()[name = tensor<string, []>("input_269_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255795392)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255797504)))]; |
| tensor<fp16, [1, 77, 1024]> input_269_cast_fp16 = layer_norm(axes = input_269_axes_0, beta = text_encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16, x = input_267_cast_fp16)[name = tensor<string, []>("input_269_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255799616))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(258945408))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(258945600))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(258948736))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_100_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16_palettized, x = input_269_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")]; |
| tensor<string, []> input_273_mode_0 = const()[name = tensor<string, []>("input_273_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_273_cast_fp16 = gelu(mode = input_273_mode_0, x = linear_100_cast_fp16)[name = tensor<string, []>("input_273_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(258948928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262094720))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262094912)))]; |
| tensor<fp16, [1, 77, 1024]> linear_101_cast_fp16 = 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_palettized, x = input_273_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_275_cast_fp16 = add(x = input_267_cast_fp16, y = linear_101_cast_fp16)[name = tensor<string, []>("input_275_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_103_axes_0 = const()[name = tensor<string, []>("hidden_states_103_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262097024)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262099136)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_103_cast_fp16 = layer_norm(axes = hidden_states_103_axes_0, beta = text_encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16, x = input_275_cast_fp16)[name = tensor<string, []>("hidden_states_103_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262101248))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262887744))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262887936)))]; |
| tensor<fp16, [1, 77, 1024]> linear_102_cast_fp16 = 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_palettized, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")]; |
| tensor<fp16, []> var_1608_to_fp16 = const()[name = tensor<string, []>("op_1608_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_107_cast_fp16 = mul(x = linear_102_cast_fp16, y = var_1608_to_fp16)[name = tensor<string, []>("tensor_107_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(262890048))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263676544))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263676736)))]; |
| tensor<fp16, [1, 77, 1024]> linear_103_cast_fp16 = 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_palettized, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")]; |
| tensor<int32, [4]> var_1613 = const()[name = tensor<string, []>("op_1613"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1614_cast_fp16 = reshape(shape = var_1613, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1614_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263678848))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264465344))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264465536)))]; |
| tensor<fp16, [1, 77, 1024]> linear_104_cast_fp16 = 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_palettized, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")]; |
| tensor<int32, [4]> var_1620 = const()[name = tensor<string, []>("op_1620"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1621_cast_fp16 = reshape(shape = var_1620, x = linear_104_cast_fp16)[name = tensor<string, []>("op_1621_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1630_cast_fp16 = reshape(shape = var_1629, x = tensor_107_cast_fp16)[name = tensor<string, []>("op_1630_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_23 = transpose(perm = var_1631_perm_0, x = var_1630_cast_fp16)[name = tensor<string, []>("transpose_23")]; |
| tensor<fp16, [16, 77, 64]> query_states_35_cast_fp16 = reshape(shape = var_1633, x = transpose_23)[name = tensor<string, []>("query_states_35_cast_fp16")]; |
| tensor<int32, [3]> var_1635 = const()[name = tensor<string, []>("op_1635"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_22 = transpose(perm = var_1615_perm_0, x = var_1614_cast_fp16)[name = tensor<string, []>("transpose_22")]; |
| tensor<fp16, [16, 77, 64]> key_states_71_cast_fp16 = reshape(shape = var_1635, x = transpose_22)[name = tensor<string, []>("key_states_71_cast_fp16")]; |
| tensor<int32, [3]> var_1637 = const()[name = tensor<string, []>("op_1637"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_21 = transpose(perm = var_1622_perm_0, x = var_1621_cast_fp16)[name = tensor<string, []>("transpose_21")]; |
| tensor<fp16, [16, 77, 64]> value_states_71_cast_fp16 = reshape(shape = var_1637, x = transpose_21)[name = tensor<string, []>("value_states_71_cast_fp16")]; |
| tensor<bool, []> attn_weights_103_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_103_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_103_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_103_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_103_cast_fp16 = matmul(transpose_x = attn_weights_103_transpose_x_1, transpose_y = attn_weights_103_transpose_y_1, x = query_states_35_cast_fp16, y = key_states_71_cast_fp16)[name = tensor<string, []>("attn_weights_103_cast_fp16")]; |
| tensor<int32, [4]> var_1642 = const()[name = tensor<string, []>("op_1642"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1643_cast_fp16 = reshape(shape = var_1642, x = attn_weights_103_cast_fp16)[name = tensor<string, []>("op_1643_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_105_cast_fp16 = add(x = var_1643_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_105_cast_fp16")]; |
| tensor<int32, [3]> var_1648 = const()[name = tensor<string, []>("op_1648"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_277_cast_fp16 = reshape(shape = var_1648, x = attn_weights_105_cast_fp16)[name = tensor<string, []>("input_277_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_279_cast_fp16 = softmax(axis = var_5, x = input_277_cast_fp16)[name = tensor<string, []>("input_279_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_103_cast_fp16 = matmul(transpose_x = attn_output_103_transpose_x_0, transpose_y = attn_output_103_transpose_y_0, x = input_279_cast_fp16, y = value_states_71_cast_fp16)[name = tensor<string, []>("attn_output_103_cast_fp16")]; |
| tensor<int32, [4]> var_1653 = const()[name = tensor<string, []>("op_1653"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_105_cast_fp16 = reshape(shape = var_1653, x = attn_output_103_cast_fp16)[name = tensor<string, []>("attn_output_105_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_20 = transpose(perm = attn_output_107_perm_0, x = attn_output_105_cast_fp16)[name = tensor<string, []>("transpose_20")]; |
| tensor<fp16, [1, 77, 1024]> input_281_cast_fp16 = reshape(shape = var_1656, x = transpose_20)[name = tensor<string, []>("input_281_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264467648))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265254144))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265254336)))]; |
| tensor<fp16, [1, 77, 1024]> linear_105_cast_fp16 = 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_palettized, x = input_281_cast_fp16)[name = tensor<string, []>("linear_105_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_283_cast_fp16 = add(x = input_275_cast_fp16, y = linear_105_cast_fp16)[name = tensor<string, []>("input_283_cast_fp16")]; |
| tensor<int32, [1]> input_285_axes_0 = const()[name = tensor<string, []>("input_285_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265256448)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265258560)))]; |
| tensor<fp16, [1, 77, 1024]> input_285_cast_fp16 = layer_norm(axes = input_285_axes_0, beta = text_encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16, x = input_283_cast_fp16)[name = tensor<string, []>("input_285_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(265260672))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268406464))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268406656))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268409792))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_106_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")]; |
| tensor<string, []> input_289_mode_0 = const()[name = tensor<string, []>("input_289_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_289_cast_fp16 = gelu(mode = input_289_mode_0, x = linear_106_cast_fp16)[name = tensor<string, []>("input_289_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(268409984))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271555776))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271555968)))]; |
| tensor<fp16, [1, 77, 1024]> linear_107_cast_fp16 = 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_palettized, x = input_289_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_291_cast_fp16 = add(x = input_283_cast_fp16, y = linear_107_cast_fp16)[name = tensor<string, []>("input_291_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_109_axes_0 = const()[name = tensor<string, []>("hidden_states_109_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271558080)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271560192)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_109_cast_fp16 = layer_norm(axes = hidden_states_109_axes_0, beta = text_encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16, x = input_291_cast_fp16)[name = tensor<string, []>("hidden_states_109_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271562304))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(272348800))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(272348992)))]; |
| tensor<fp16, [1, 77, 1024]> linear_108_cast_fp16 = 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_palettized, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")]; |
| tensor<fp16, []> var_1695_to_fp16 = const()[name = tensor<string, []>("op_1695_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_113_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_1695_to_fp16)[name = tensor<string, []>("tensor_113_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(272351104))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(273137600))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(273137792)))]; |
| tensor<fp16, [1, 77, 1024]> linear_109_cast_fp16 = 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_palettized, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")]; |
| tensor<int32, [4]> var_1700 = const()[name = tensor<string, []>("op_1700"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1701_cast_fp16 = reshape(shape = var_1700, x = linear_109_cast_fp16)[name = tensor<string, []>("op_1701_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(273139904))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(273926400))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(273926592)))]; |
| tensor<fp16, [1, 77, 1024]> linear_110_cast_fp16 = 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_palettized, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")]; |
| tensor<int32, [4]> var_1707 = const()[name = tensor<string, []>("op_1707"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1708_cast_fp16 = reshape(shape = var_1707, x = linear_110_cast_fp16)[name = tensor<string, []>("op_1708_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1717_cast_fp16 = reshape(shape = var_1716, x = tensor_113_cast_fp16)[name = tensor<string, []>("op_1717_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_19 = transpose(perm = var_1718_perm_0, x = var_1717_cast_fp16)[name = tensor<string, []>("transpose_19")]; |
| tensor<fp16, [16, 77, 64]> query_states_37_cast_fp16 = reshape(shape = var_1720, x = transpose_19)[name = tensor<string, []>("query_states_37_cast_fp16")]; |
| tensor<int32, [3]> var_1722 = const()[name = tensor<string, []>("op_1722"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_18 = transpose(perm = var_1702_perm_0, x = var_1701_cast_fp16)[name = tensor<string, []>("transpose_18")]; |
| tensor<fp16, [16, 77, 64]> key_states_75_cast_fp16 = reshape(shape = var_1722, x = transpose_18)[name = tensor<string, []>("key_states_75_cast_fp16")]; |
| tensor<int32, [3]> var_1724 = const()[name = tensor<string, []>("op_1724"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_17 = transpose(perm = var_1709_perm_0, x = var_1708_cast_fp16)[name = tensor<string, []>("transpose_17")]; |
| tensor<fp16, [16, 77, 64]> value_states_75_cast_fp16 = reshape(shape = var_1724, x = transpose_17)[name = tensor<string, []>("value_states_75_cast_fp16")]; |
| tensor<bool, []> attn_weights_109_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_109_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_109_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_109_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_109_cast_fp16 = matmul(transpose_x = attn_weights_109_transpose_x_1, transpose_y = attn_weights_109_transpose_y_1, x = query_states_37_cast_fp16, y = key_states_75_cast_fp16)[name = tensor<string, []>("attn_weights_109_cast_fp16")]; |
| tensor<int32, [4]> var_1729 = const()[name = tensor<string, []>("op_1729"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1730_cast_fp16 = reshape(shape = var_1729, x = attn_weights_109_cast_fp16)[name = tensor<string, []>("op_1730_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_111_cast_fp16 = add(x = var_1730_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_111_cast_fp16")]; |
| tensor<int32, [3]> var_1735 = const()[name = tensor<string, []>("op_1735"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_293_cast_fp16 = reshape(shape = var_1735, x = attn_weights_111_cast_fp16)[name = tensor<string, []>("input_293_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_295_cast_fp16 = softmax(axis = var_5, x = input_293_cast_fp16)[name = tensor<string, []>("input_295_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_109_cast_fp16 = matmul(transpose_x = attn_output_109_transpose_x_0, transpose_y = attn_output_109_transpose_y_0, x = input_295_cast_fp16, y = value_states_75_cast_fp16)[name = tensor<string, []>("attn_output_109_cast_fp16")]; |
| tensor<int32, [4]> var_1740 = const()[name = tensor<string, []>("op_1740"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_111_cast_fp16 = reshape(shape = var_1740, x = attn_output_109_cast_fp16)[name = tensor<string, []>("attn_output_111_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_16 = transpose(perm = attn_output_113_perm_0, x = attn_output_111_cast_fp16)[name = tensor<string, []>("transpose_16")]; |
| tensor<fp16, [1, 77, 1024]> input_297_cast_fp16 = reshape(shape = var_1743, x = transpose_16)[name = tensor<string, []>("input_297_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(273928704))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274715200))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274715392)))]; |
| tensor<fp16, [1, 77, 1024]> linear_111_cast_fp16 = 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_palettized, x = input_297_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_111_cast_fp16)[name = tensor<string, []>("input_299_cast_fp16")]; |
| tensor<int32, [1]> input_301_axes_0 = const()[name = tensor<string, []>("input_301_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274717504)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274719616)))]; |
| tensor<fp16, [1, 77, 1024]> input_301_cast_fp16 = layer_norm(axes = input_301_axes_0, beta = text_encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16, x = input_299_cast_fp16)[name = tensor<string, []>("input_301_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274721728))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277867520))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277867712))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277870848))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_112_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16_palettized, x = input_301_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")]; |
| tensor<string, []> input_305_mode_0 = const()[name = tensor<string, []>("input_305_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_305_cast_fp16 = gelu(mode = input_305_mode_0, x = linear_112_cast_fp16)[name = tensor<string, []>("input_305_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277871040))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281016832))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281017024)))]; |
| tensor<fp16, [1, 77, 1024]> linear_113_cast_fp16 = 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_palettized, x = input_305_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_307_cast_fp16 = add(x = input_299_cast_fp16, y = linear_113_cast_fp16)[name = tensor<string, []>("input_307_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_115_axes_0 = const()[name = tensor<string, []>("hidden_states_115_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281019136)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281021248)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_115_cast_fp16 = layer_norm(axes = hidden_states_115_axes_0, beta = text_encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16, x = input_307_cast_fp16)[name = tensor<string, []>("hidden_states_115_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281023360))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281809856))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281810048)))]; |
| tensor<fp16, [1, 77, 1024]> linear_114_cast_fp16 = 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_palettized, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")]; |
| tensor<fp16, []> var_1782_to_fp16 = const()[name = tensor<string, []>("op_1782_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_119_cast_fp16 = mul(x = linear_114_cast_fp16, y = var_1782_to_fp16)[name = tensor<string, []>("tensor_119_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(281812160))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282598656))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282598848)))]; |
| tensor<fp16, [1, 77, 1024]> linear_115_cast_fp16 = 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_palettized, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")]; |
| tensor<int32, [4]> var_1787 = const()[name = tensor<string, []>("op_1787"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1788_cast_fp16 = reshape(shape = var_1787, x = linear_115_cast_fp16)[name = tensor<string, []>("op_1788_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282600960))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283387456))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283387648)))]; |
| tensor<fp16, [1, 77, 1024]> linear_116_cast_fp16 = 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_palettized, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")]; |
| tensor<int32, [4]> var_1794 = const()[name = tensor<string, []>("op_1794"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1795_cast_fp16 = reshape(shape = var_1794, x = linear_116_cast_fp16)[name = tensor<string, []>("op_1795_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1804_cast_fp16 = reshape(shape = var_1803, x = tensor_119_cast_fp16)[name = tensor<string, []>("op_1804_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_15 = transpose(perm = var_1805_perm_0, x = var_1804_cast_fp16)[name = tensor<string, []>("transpose_15")]; |
| tensor<fp16, [16, 77, 64]> query_states_39_cast_fp16 = reshape(shape = var_1807, x = transpose_15)[name = tensor<string, []>("query_states_39_cast_fp16")]; |
| tensor<int32, [3]> var_1809 = const()[name = tensor<string, []>("op_1809"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_14 = transpose(perm = var_1789_perm_0, x = var_1788_cast_fp16)[name = tensor<string, []>("transpose_14")]; |
| tensor<fp16, [16, 77, 64]> key_states_79_cast_fp16 = reshape(shape = var_1809, x = transpose_14)[name = tensor<string, []>("key_states_79_cast_fp16")]; |
| tensor<int32, [3]> var_1811 = const()[name = tensor<string, []>("op_1811"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_13 = transpose(perm = var_1796_perm_0, x = var_1795_cast_fp16)[name = tensor<string, []>("transpose_13")]; |
| tensor<fp16, [16, 77, 64]> value_states_79_cast_fp16 = reshape(shape = var_1811, x = transpose_13)[name = tensor<string, []>("value_states_79_cast_fp16")]; |
| tensor<bool, []> attn_weights_115_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_115_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_115_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_115_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_115_cast_fp16 = matmul(transpose_x = attn_weights_115_transpose_x_1, transpose_y = attn_weights_115_transpose_y_1, x = query_states_39_cast_fp16, y = key_states_79_cast_fp16)[name = tensor<string, []>("attn_weights_115_cast_fp16")]; |
| tensor<int32, [4]> var_1816 = const()[name = tensor<string, []>("op_1816"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1817_cast_fp16 = reshape(shape = var_1816, x = attn_weights_115_cast_fp16)[name = tensor<string, []>("op_1817_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_117_cast_fp16 = add(x = var_1817_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_117_cast_fp16")]; |
| tensor<int32, [3]> var_1822 = const()[name = tensor<string, []>("op_1822"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_309_cast_fp16 = reshape(shape = var_1822, x = attn_weights_117_cast_fp16)[name = tensor<string, []>("input_309_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_311_cast_fp16 = softmax(axis = var_5, x = input_309_cast_fp16)[name = tensor<string, []>("input_311_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_115_cast_fp16 = matmul(transpose_x = attn_output_115_transpose_x_0, transpose_y = attn_output_115_transpose_y_0, x = input_311_cast_fp16, y = value_states_79_cast_fp16)[name = tensor<string, []>("attn_output_115_cast_fp16")]; |
| tensor<int32, [4]> var_1827 = const()[name = tensor<string, []>("op_1827"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_117_cast_fp16 = reshape(shape = var_1827, x = attn_output_115_cast_fp16)[name = tensor<string, []>("attn_output_117_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_12 = transpose(perm = attn_output_119_perm_0, x = attn_output_117_cast_fp16)[name = tensor<string, []>("transpose_12")]; |
| tensor<fp16, [1, 77, 1024]> input_313_cast_fp16 = reshape(shape = var_1830, x = transpose_12)[name = tensor<string, []>("input_313_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(283389760))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284176256))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284176448)))]; |
| tensor<fp16, [1, 77, 1024]> linear_117_cast_fp16 = 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_palettized, x = input_313_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_315_cast_fp16 = add(x = input_307_cast_fp16, y = linear_117_cast_fp16)[name = tensor<string, []>("input_315_cast_fp16")]; |
| tensor<int32, [1]> input_317_axes_0 = const()[name = tensor<string, []>("input_317_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284178560)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284180672)))]; |
| tensor<fp16, [1, 77, 1024]> input_317_cast_fp16 = layer_norm(axes = input_317_axes_0, beta = text_encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16, x = input_315_cast_fp16)[name = tensor<string, []>("input_317_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284182784))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287328576))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287328768))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287331904))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_118_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16_palettized, x = input_317_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")]; |
| tensor<string, []> input_321_mode_0 = const()[name = tensor<string, []>("input_321_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_321_cast_fp16 = gelu(mode = input_321_mode_0, x = linear_118_cast_fp16)[name = tensor<string, []>("input_321_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(287332096))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290477888))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290478080)))]; |
| tensor<fp16, [1, 77, 1024]> linear_119_cast_fp16 = 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_palettized, x = input_321_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_323_cast_fp16 = add(x = input_315_cast_fp16, y = linear_119_cast_fp16)[name = tensor<string, []>("input_323_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_121_axes_0 = const()[name = tensor<string, []>("hidden_states_121_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290480192)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290482304)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = text_encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16, x = input_323_cast_fp16)[name = tensor<string, []>("hidden_states_121_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(290484416))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291270912))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291271104)))]; |
| tensor<fp16, [1, 77, 1024]> linear_120_cast_fp16 = 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_palettized, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")]; |
| tensor<fp16, []> var_1869_to_fp16 = const()[name = tensor<string, []>("op_1869_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_125_cast_fp16 = mul(x = linear_120_cast_fp16, y = var_1869_to_fp16)[name = tensor<string, []>("tensor_125_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291273216))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292059712))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292059904)))]; |
| tensor<fp16, [1, 77, 1024]> linear_121_cast_fp16 = 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_palettized, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")]; |
| tensor<int32, [4]> var_1874 = const()[name = tensor<string, []>("op_1874"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1875_cast_fp16 = reshape(shape = var_1874, x = linear_121_cast_fp16)[name = tensor<string, []>("op_1875_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292062016))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292848512))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292848704)))]; |
| tensor<fp16, [1, 77, 1024]> linear_122_cast_fp16 = 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_palettized, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")]; |
| tensor<int32, [4]> var_1881 = const()[name = tensor<string, []>("op_1881"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1882_cast_fp16 = reshape(shape = var_1881, x = linear_122_cast_fp16)[name = tensor<string, []>("op_1882_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1891_cast_fp16 = reshape(shape = var_1890, x = tensor_125_cast_fp16)[name = tensor<string, []>("op_1891_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_11 = transpose(perm = var_1892_perm_0, x = var_1891_cast_fp16)[name = tensor<string, []>("transpose_11")]; |
| tensor<fp16, [16, 77, 64]> query_states_41_cast_fp16 = reshape(shape = var_1894, x = transpose_11)[name = tensor<string, []>("query_states_41_cast_fp16")]; |
| tensor<int32, [3]> var_1896 = const()[name = tensor<string, []>("op_1896"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_10 = transpose(perm = var_1876_perm_0, x = var_1875_cast_fp16)[name = tensor<string, []>("transpose_10")]; |
| tensor<fp16, [16, 77, 64]> key_states_83_cast_fp16 = reshape(shape = var_1896, x = transpose_10)[name = tensor<string, []>("key_states_83_cast_fp16")]; |
| tensor<int32, [3]> var_1898 = const()[name = tensor<string, []>("op_1898"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_9 = transpose(perm = var_1883_perm_0, x = var_1882_cast_fp16)[name = tensor<string, []>("transpose_9")]; |
| tensor<fp16, [16, 77, 64]> value_states_83_cast_fp16 = reshape(shape = var_1898, x = transpose_9)[name = tensor<string, []>("value_states_83_cast_fp16")]; |
| tensor<bool, []> attn_weights_121_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_121_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_121_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_121_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_121_cast_fp16 = matmul(transpose_x = attn_weights_121_transpose_x_1, transpose_y = attn_weights_121_transpose_y_1, x = query_states_41_cast_fp16, y = key_states_83_cast_fp16)[name = tensor<string, []>("attn_weights_121_cast_fp16")]; |
| tensor<int32, [4]> var_1903 = const()[name = tensor<string, []>("op_1903"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1904_cast_fp16 = reshape(shape = var_1903, x = attn_weights_121_cast_fp16)[name = tensor<string, []>("op_1904_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_123_cast_fp16 = add(x = var_1904_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_123_cast_fp16")]; |
| tensor<int32, [3]> var_1909 = const()[name = tensor<string, []>("op_1909"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_325_cast_fp16 = reshape(shape = var_1909, x = attn_weights_123_cast_fp16)[name = tensor<string, []>("input_325_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_327_cast_fp16 = softmax(axis = var_5, x = input_325_cast_fp16)[name = tensor<string, []>("input_327_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_121_cast_fp16 = matmul(transpose_x = attn_output_121_transpose_x_0, transpose_y = attn_output_121_transpose_y_0, x = input_327_cast_fp16, y = value_states_83_cast_fp16)[name = tensor<string, []>("attn_output_121_cast_fp16")]; |
| tensor<int32, [4]> var_1914 = const()[name = tensor<string, []>("op_1914"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_123_cast_fp16 = reshape(shape = var_1914, x = attn_output_121_cast_fp16)[name = tensor<string, []>("attn_output_123_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_8 = transpose(perm = attn_output_125_perm_0, x = attn_output_123_cast_fp16)[name = tensor<string, []>("transpose_8")]; |
| tensor<fp16, [1, 77, 1024]> input_329_cast_fp16 = reshape(shape = var_1917, x = transpose_8)[name = tensor<string, []>("input_329_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(292850816))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293637312))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293637504)))]; |
| tensor<fp16, [1, 77, 1024]> linear_123_cast_fp16 = 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_palettized, x = input_329_cast_fp16)[name = tensor<string, []>("linear_123_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_331_cast_fp16 = add(x = input_323_cast_fp16, y = linear_123_cast_fp16)[name = tensor<string, []>("input_331_cast_fp16")]; |
| tensor<int32, [1]> input_333_axes_0 = const()[name = tensor<string, []>("input_333_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293639616)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293641728)))]; |
| tensor<fp16, [1, 77, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = text_encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16, x = input_331_cast_fp16)[name = tensor<string, []>("input_333_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293643840))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296789632))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296789824))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296792960))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_124_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16_palettized, x = input_333_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")]; |
| tensor<string, []> input_337_mode_0 = const()[name = tensor<string, []>("input_337_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_337_cast_fp16 = gelu(mode = input_337_mode_0, x = linear_124_cast_fp16)[name = tensor<string, []>("input_337_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296793152))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299938944))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299939136)))]; |
| tensor<fp16, [1, 77, 1024]> linear_125_cast_fp16 = 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_palettized, x = input_337_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_339_cast_fp16 = add(x = input_331_cast_fp16, y = linear_125_cast_fp16)[name = tensor<string, []>("input_339_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_127_axes_0 = const()[name = tensor<string, []>("hidden_states_127_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299941248)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299943360)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_127_cast_fp16 = layer_norm(axes = hidden_states_127_axes_0, beta = text_encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16, x = input_339_cast_fp16)[name = tensor<string, []>("hidden_states_127_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299945472))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(300731968))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(300732160)))]; |
| tensor<fp16, [1, 77, 1024]> linear_126_cast_fp16 = 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_palettized, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")]; |
| tensor<fp16, []> var_1956_to_fp16 = const()[name = tensor<string, []>("op_1956_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_131_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_1956_to_fp16)[name = tensor<string, []>("tensor_131_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(300734272))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301520768))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301520960)))]; |
| tensor<fp16, [1, 77, 1024]> linear_127_cast_fp16 = 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_palettized, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")]; |
| tensor<int32, [4]> var_1961 = const()[name = tensor<string, []>("op_1961"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1962_cast_fp16 = reshape(shape = var_1961, x = linear_127_cast_fp16)[name = tensor<string, []>("op_1962_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301523072))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302309568))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302309760)))]; |
| tensor<fp16, [1, 77, 1024]> linear_128_cast_fp16 = 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_palettized, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")]; |
| tensor<int32, [4]> var_1968 = const()[name = tensor<string, []>("op_1968"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1969_cast_fp16 = reshape(shape = var_1968, x = linear_128_cast_fp16)[name = tensor<string, []>("op_1969_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_1978_cast_fp16 = reshape(shape = var_1977, x = tensor_131_cast_fp16)[name = tensor<string, []>("op_1978_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_7 = transpose(perm = var_1979_perm_0, x = var_1978_cast_fp16)[name = tensor<string, []>("transpose_7")]; |
| tensor<fp16, [16, 77, 64]> query_states_43_cast_fp16 = reshape(shape = var_1981, x = transpose_7)[name = tensor<string, []>("query_states_43_cast_fp16")]; |
| tensor<int32, [3]> var_1983 = const()[name = tensor<string, []>("op_1983"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_6 = transpose(perm = var_1963_perm_0, x = var_1962_cast_fp16)[name = tensor<string, []>("transpose_6")]; |
| tensor<fp16, [16, 77, 64]> key_states_87_cast_fp16 = reshape(shape = var_1983, x = transpose_6)[name = tensor<string, []>("key_states_87_cast_fp16")]; |
| tensor<int32, [3]> var_1985 = const()[name = tensor<string, []>("op_1985"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_5 = transpose(perm = var_1970_perm_0, x = var_1969_cast_fp16)[name = tensor<string, []>("transpose_5")]; |
| tensor<fp16, [16, 77, 64]> value_states_87_cast_fp16 = reshape(shape = var_1985, x = transpose_5)[name = tensor<string, []>("value_states_87_cast_fp16")]; |
| tensor<bool, []> attn_weights_127_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_127_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_127_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_127_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_127_cast_fp16 = matmul(transpose_x = attn_weights_127_transpose_x_1, transpose_y = attn_weights_127_transpose_y_1, x = query_states_43_cast_fp16, y = key_states_87_cast_fp16)[name = tensor<string, []>("attn_weights_127_cast_fp16")]; |
| tensor<int32, [4]> var_1990 = const()[name = tensor<string, []>("op_1990"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_1991_cast_fp16 = reshape(shape = var_1990, x = attn_weights_127_cast_fp16)[name = tensor<string, []>("op_1991_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_129_cast_fp16 = add(x = var_1991_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_129_cast_fp16")]; |
| tensor<int32, [3]> var_1996 = const()[name = tensor<string, []>("op_1996"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_341_cast_fp16 = reshape(shape = var_1996, x = attn_weights_129_cast_fp16)[name = tensor<string, []>("input_341_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_343_cast_fp16 = softmax(axis = var_5, x = input_341_cast_fp16)[name = tensor<string, []>("input_343_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_127_cast_fp16 = matmul(transpose_x = attn_output_127_transpose_x_0, transpose_y = attn_output_127_transpose_y_0, x = input_343_cast_fp16, y = value_states_87_cast_fp16)[name = tensor<string, []>("attn_output_127_cast_fp16")]; |
| tensor<int32, [4]> var_2001 = const()[name = tensor<string, []>("op_2001"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_129_cast_fp16 = reshape(shape = var_2001, x = attn_output_127_cast_fp16)[name = tensor<string, []>("attn_output_129_cast_fp16")]; |
| 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, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_4 = transpose(perm = attn_output_131_perm_0, x = attn_output_129_cast_fp16)[name = tensor<string, []>("transpose_4")]; |
| tensor<fp16, [1, 77, 1024]> input_345_cast_fp16 = reshape(shape = var_2004, x = transpose_4)[name = tensor<string, []>("input_345_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302311872))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303098368))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303098560)))]; |
| tensor<fp16, [1, 77, 1024]> linear_129_cast_fp16 = 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_palettized, x = input_345_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_347_cast_fp16 = add(x = input_339_cast_fp16, y = linear_129_cast_fp16)[name = tensor<string, []>("input_347_cast_fp16")]; |
| tensor<int32, [1]> input_349_axes_0 = const()[name = tensor<string, []>("input_349_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303100672)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303102784)))]; |
| tensor<fp16, [1, 77, 1024]> input_349_cast_fp16 = layer_norm(axes = input_349_axes_0, beta = text_encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16, x = input_347_cast_fp16)[name = tensor<string, []>("input_349_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(303104896))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(306250688))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(306250880))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(306254016))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_130_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16_palettized, x = input_349_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")]; |
| tensor<string, []> input_353_mode_0 = const()[name = tensor<string, []>("input_353_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_353_cast_fp16 = gelu(mode = input_353_mode_0, x = linear_130_cast_fp16)[name = tensor<string, []>("input_353_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(306254208))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(309400000))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(309400192)))]; |
| tensor<fp16, [1, 77, 1024]> linear_131_cast_fp16 = 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_palettized, x = input_353_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_355_cast_fp16 = add(x = input_347_cast_fp16, y = linear_131_cast_fp16)[name = tensor<string, []>("input_355_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_133_axes_0 = const()[name = tensor<string, []>("hidden_states_133_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(309402304)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(309404416)))]; |
| tensor<fp16, [1, 77, 1024]> hidden_states_133_cast_fp16 = layer_norm(axes = hidden_states_133_axes_0, beta = text_encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16, x = input_355_cast_fp16)[name = tensor<string, []>("hidden_states_133_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(309406528))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310193024))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310193216)))]; |
| tensor<fp16, [1, 77, 1024]> linear_132_cast_fp16 = 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_palettized, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_132_cast_fp16")]; |
| tensor<fp16, []> var_2043_to_fp16 = const()[name = tensor<string, []>("op_2043_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| tensor<fp16, [1, 77, 1024]> tensor_cast_fp16 = mul(x = linear_132_cast_fp16, y = var_2043_to_fp16)[name = tensor<string, []>("tensor_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310195328))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310981824))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310982016)))]; |
| tensor<fp16, [1, 77, 1024]> linear_133_cast_fp16 = 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_palettized, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")]; |
| tensor<int32, [4]> var_2048 = const()[name = tensor<string, []>("op_2048"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_2049_cast_fp16 = reshape(shape = var_2048, x = linear_133_cast_fp16)[name = tensor<string, []>("op_2049_cast_fp16")]; |
| 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, [1024, 1024]> text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310984128))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(311770624))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(311770816)))]; |
| tensor<fp16, [1, 77, 1024]> linear_134_cast_fp16 = 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_palettized, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")]; |
| tensor<int32, [4]> var_2055 = const()[name = tensor<string, []>("op_2055"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_2056_cast_fp16 = reshape(shape = var_2055, x = linear_134_cast_fp16)[name = tensor<string, []>("op_2056_cast_fp16")]; |
| 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, 16, 64])]; |
| tensor<fp16, [1, 77, 16, 64]> var_2065_cast_fp16 = reshape(shape = var_2064, x = tensor_cast_fp16)[name = tensor<string, []>("op_2065_cast_fp16")]; |
| 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]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_3 = transpose(perm = var_2066_perm_0, x = var_2065_cast_fp16)[name = tensor<string, []>("transpose_3")]; |
| tensor<fp16, [16, 77, 64]> query_states_cast_fp16 = reshape(shape = var_2068, x = transpose_3)[name = tensor<string, []>("query_states_cast_fp16")]; |
| tensor<int32, [3]> var_2070 = const()[name = tensor<string, []>("op_2070"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_2 = transpose(perm = var_2050_perm_0, x = var_2049_cast_fp16)[name = tensor<string, []>("transpose_2")]; |
| tensor<fp16, [16, 77, 64]> key_states_cast_fp16 = reshape(shape = var_2070, x = transpose_2)[name = tensor<string, []>("key_states_cast_fp16")]; |
| tensor<int32, [3]> var_2072 = const()[name = tensor<string, []>("op_2072"), val = tensor<int32, [3]>([16, -1, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> transpose_1 = transpose(perm = var_2057_perm_0, x = var_2056_cast_fp16)[name = tensor<string, []>("transpose_1")]; |
| tensor<fp16, [16, 77, 64]> value_states_cast_fp16 = reshape(shape = var_2072, x = transpose_1)[name = tensor<string, []>("value_states_cast_fp16")]; |
| tensor<bool, []> attn_weights_133_transpose_x_1 = const()[name = tensor<string, []>("attn_weights_133_transpose_x_1"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_weights_133_transpose_y_1 = const()[name = tensor<string, []>("attn_weights_133_transpose_y_1"), val = tensor<bool, []>(true)]; |
| tensor<fp16, [16, 77, 77]> attn_weights_133_cast_fp16 = matmul(transpose_x = attn_weights_133_transpose_x_1, transpose_y = attn_weights_133_transpose_y_1, x = query_states_cast_fp16, y = key_states_cast_fp16)[name = tensor<string, []>("attn_weights_133_cast_fp16")]; |
| tensor<int32, [4]> var_2077 = const()[name = tensor<string, []>("op_2077"), val = tensor<int32, [4]>([1, 16, 77, 77])]; |
| tensor<fp16, [1, 16, 77, 77]> var_2078_cast_fp16 = reshape(shape = var_2077, x = attn_weights_133_cast_fp16)[name = tensor<string, []>("op_2078_cast_fp16")]; |
| tensor<fp16, [1, 16, 77, 77]> attn_weights_135_cast_fp16 = add(x = var_2078_cast_fp16, y = var_57_to_fp16)[name = tensor<string, []>("attn_weights_135_cast_fp16")]; |
| tensor<int32, [3]> var_2083 = const()[name = tensor<string, []>("op_2083"), val = tensor<int32, [3]>([16, 77, 77])]; |
| tensor<fp16, [16, 77, 77]> input_357_cast_fp16 = reshape(shape = var_2083, x = attn_weights_135_cast_fp16)[name = tensor<string, []>("input_357_cast_fp16")]; |
| tensor<fp16, [16, 77, 77]> input_359_cast_fp16 = softmax(axis = var_5, x = input_357_cast_fp16)[name = tensor<string, []>("input_359_cast_fp16")]; |
| 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, [16, 77, 64]> attn_output_133_cast_fp16 = matmul(transpose_x = attn_output_133_transpose_x_0, transpose_y = attn_output_133_transpose_y_0, x = input_359_cast_fp16, y = value_states_cast_fp16)[name = tensor<string, []>("attn_output_133_cast_fp16")]; |
| tensor<int32, [4]> var_2088 = const()[name = tensor<string, []>("op_2088"), val = tensor<int32, [4]>([1, 16, 77, 64])]; |
| tensor<fp16, [1, 16, 77, 64]> attn_output_135_cast_fp16 = reshape(shape = var_2088, x = attn_output_133_cast_fp16)[name = tensor<string, []>("attn_output_135_cast_fp16")]; |
| tensor<int32, [4]> attn_output_perm_0 = const()[name = tensor<string, []>("attn_output_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2091 = const()[name = tensor<string, []>("op_2091"), val = tensor<int32, [3]>([1, 77, 1024])]; |
| tensor<fp16, [1, 77, 16, 64]> transpose_0 = transpose(perm = attn_output_perm_0, x = attn_output_135_cast_fp16)[name = tensor<string, []>("transpose_0")]; |
| tensor<fp16, [1, 77, 1024]> input_361_cast_fp16 = reshape(shape = var_2091, x = transpose_0)[name = tensor<string, []>("input_361_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [786432]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(311772928))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312559424))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 1024])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312559616)))]; |
| tensor<fp16, [1, 77, 1024]> linear_135_cast_fp16 = 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_palettized, x = input_361_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_363_cast_fp16 = add(x = input_355_cast_fp16, y = linear_135_cast_fp16)[name = tensor<string, []>("input_363_cast_fp16")]; |
| tensor<int32, [1]> input_365_axes_0 = const()[name = tensor<string, []>("input_365_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312561728)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312563840)))]; |
| tensor<fp16, [1, 77, 1024]> input_365_cast_fp16 = layer_norm(axes = input_365_axes_0, beta = text_encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16, x = input_363_cast_fp16)[name = tensor<string, []>("input_365_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312565952))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315711744))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([4096, 1024])]; |
| tensor<fp16, [4096]> text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315711936))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315715072))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16_palettized"), shape = tensor<uint32, [1]>([4096])]; |
| tensor<fp16, [1, 77, 4096]> linear_136_cast_fp16 = linear(bias = text_encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16_palettized, weight = text_encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16_palettized, x = input_365_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")]; |
| tensor<string, []> input_369_mode_0 = const()[name = tensor<string, []>("input_369_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 77, 4096]> input_369_cast_fp16 = gelu(mode = input_369_mode_0, x = linear_136_cast_fp16)[name = tensor<string, []>("input_369_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [3145728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315715264))), lut = tensor<fp16, [64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318861056))), name = tensor<string, []>("text_encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([1024, 4096])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318861248)))]; |
| tensor<fp16, [1, 77, 1024]> linear_137_cast_fp16 = 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_palettized, x = input_369_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")]; |
| tensor<fp16, [1, 77, 1024]> input_cast_fp16 = add(x = input_363_cast_fp16, y = linear_137_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; |
| tensor<int32, [1]> last_hidden_state_axes_0 = const()[name = tensor<string, []>("last_hidden_state_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318863360)))]; |
| tensor<fp16, [1024]> 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, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318865472)))]; |
| tensor<fp16, [1, 77, 1024]> last_hidden_state_cast_fp16 = layer_norm(axes = last_hidden_state_axes_0, beta = text_encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_15_to_fp16, gamma = text_encoder_text_model_final_layer_norm_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("last_hidden_state_cast_fp16")]; |
| tensor<string, []> last_hidden_state_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("last_hidden_state_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<int32, [1]> var_2116 = const()[name = tensor<string, []>("op_2116"), val = tensor<int32, [1]>([0])]; |
| tensor<int32, [1]> var_2118 = reduce_argmax(axis = var_5, keep_dims = var_6, x = cast_2)[name = tensor<string, []>("op_2118")]; |
| 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_2116, var_2118))[name = tensor<string, []>("stack_0")]; |
| tensor<int32, []> var_2120_transpose_batch_dims_0 = const()[name = tensor<string, []>("op_2120_transpose_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp16, [1, 1024]> var_2120_transpose_cast_fp16 = gather_nd(batch_dims = var_2120_transpose_batch_dims_0, indices = stack_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("op_2120_transpose_cast_fp16")]; |
| tensor<string, []> var_2120_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_2120_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, [1, 77, 1024]> last_hidden_state = cast(dtype = last_hidden_state_cast_fp16_to_fp32_dtype_0, x = last_hidden_state_cast_fp16)[name = tensor<string, []>("cast_0")]; |
| tensor<fp32, [1, 1024]> pooled_outputs = cast(dtype = var_2120_cast_fp16_to_fp32_dtype_0, x = var_2120_transpose_cast_fp16)[name = tensor<string, []>("cast_1")]; |
| } -> (last_hidden_state, pooled_outputs); |
| } |
