| program(1.3) |
| [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})] |
| { |
| func main<ios18>(tensor<int32, [1, 77]> input_ids) { |
| int32 inputs_embeds_batch_dims_0 = const()[name = string("inputs_embeds_batch_dims_0"), val = int32(0)]; |
| bool inputs_embeds_validate_indices_0 = const()[name = string("inputs_embeds_validate_indices_0"), val = bool(false)]; |
| tensor<fp16, [49408, 1280]> encoder_text_model_embeddings_token_embedding_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [49408, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [3088, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47431808))))[name = string("encoder_text_model_embeddings_token_embedding_weight_to_fp16_palettized")]; |
| int32 greater_equal_0_y_0 = const()[name = string("greater_equal_0_y_0"), val = int32(0)]; |
| tensor<bool, [1, 77]> greater_equal_0 = greater_equal(x = input_ids, y = greater_equal_0_y_0)[name = string("greater_equal_0")]; |
| int32 slice_by_index_0 = const()[name = string("slice_by_index_0"), val = int32(49408)]; |
| tensor<int32, [1, 77]> add_32 = add(x = input_ids, y = slice_by_index_0)[name = string("add_32")]; |
| tensor<int32, [1, 77]> select_0 = select(a = input_ids, b = add_32, cond = greater_equal_0)[name = string("select_0")]; |
| int32 greater_equal_0_y_0_1 = const()[name = string("greater_equal_0_y_0_1"), val = int32(0)]; |
| tensor<bool, [1, 77]> greater_equal_0_1 = greater_equal(x = select_0, y = greater_equal_0_y_0_1)[name = string("greater_equal_0_1")]; |
| int32 slice_by_index_0_1 = const()[name = string("slice_by_index_0_1"), val = int32(49408)]; |
| tensor<int32, [1, 77]> add_0 = add(x = select_0, y = slice_by_index_0_1)[name = string("add_0")]; |
| tensor<int32, [1, 77]> select_0_1 = select(a = select_0, b = add_0, cond = greater_equal_0_1)[name = string("select_0_1")]; |
| int32 inputs_embeds_cast_fp16_axis_0 = const()[name = string("inputs_embeds_cast_fp16_axis_0"), val = int32(0)]; |
| tensor<fp16, [1, 77, 1280]> inputs_embeds_cast_fp16 = gather(axis = inputs_embeds_cast_fp16_axis_0, batch_dims = inputs_embeds_batch_dims_0, indices = select_0_1, validate_indices = inputs_embeds_validate_indices_0, x = encoder_text_model_embeddings_token_embedding_weight_to_fp16_palettized)[name = string("inputs_embeds_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> position_embeddings_to_fp16 = const()[name = string("position_embeddings_to_fp16"), val = tensor<fp16, [1, 77, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47827136)))]; |
| tensor<fp16, [1, 77, 1280]> input_3_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = position_embeddings_to_fp16)[name = string("input_3_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = string("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48024320)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48026944)))]; |
| fp16 var_9_to_fp16 = const()[name = string("op_9_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = encoder_text_model_encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_0_layer_norm1_weight_to_fp16, x = input_3_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48029568))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49258432))))[name = string("encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49268736)))]; |
| tensor<fp16, [1, 77, 1280]> linear_0_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = string("linear_0_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49271360))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50500224))))[name = string("encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50510528)))]; |
| tensor<fp16, [1, 77, 1280]> linear_1_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = string("linear_1_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50513152))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51742016))))[name = string("encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51752320)))]; |
| tensor<fp16, [1, 77, 1280]> linear_2_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = string("linear_2_cast_fp16")]; |
| tensor<int32, [4]> var_125 = const()[name = string("op_125"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_126_cast_fp16 = reshape(shape = var_125, x = linear_0_cast_fp16)[name = string("op_126_cast_fp16")]; |
| tensor<int32, [4]> var_128 = const()[name = string("op_128"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_129_cast_fp16 = reshape(shape = var_128, x = linear_1_cast_fp16)[name = string("op_129_cast_fp16")]; |
| tensor<int32, [4]> var_131 = const()[name = string("op_131"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_132_cast_fp16 = reshape(shape = var_131, x = linear_2_cast_fp16)[name = string("op_132_cast_fp16")]; |
| tensor<int32, [4]> value_1_perm_0 = const()[name = string("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| fp16 var_11_to_fp16 = const()[name = string("op_11_to_fp16"), val = fp16(0x1p-3)]; |
| tensor<fp16, [1, 77, 20, 64]> mul_1_cast_fp16 = mul(x = var_126_cast_fp16, y = var_11_to_fp16)[name = string("mul_1_cast_fp16")]; |
| bool matmul_0_transpose_y_0 = const()[name = string("matmul_0_transpose_y_0"), val = bool(true)]; |
| bool matmul_0_transpose_x_0 = const()[name = string("matmul_0_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_129 = transpose(perm = transpose_129_perm_0, x = var_129_cast_fp16)[name = string("transpose_318")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_128 = transpose(perm = transpose_128_perm_0, x = mul_1_cast_fp16)[name = string("transpose_319")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("matmul_0_cast_fp16")]; |
| tensor<fp16, [77, 77]> mul_0_to_fp16 = const()[name = string("mul_0_to_fp16"), val = tensor<fp16, [77, 77]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51754944)))]; |
| tensor<fp16, [1, 20, 77, 77]> add_0_cast_fp16 = add(x = matmul_0_cast_fp16, y = mul_0_to_fp16)[name = string("add_0_cast_fp16")]; |
| int32 softmax_0_axis_0 = const()[name = string("softmax_0_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_0_cast_fp16 = softmax(axis = softmax_0_axis_0, x = add_0_cast_fp16)[name = string("softmax_0_cast_fp16")]; |
| bool attn_output_1_transpose_x_0 = const()[name = string("attn_output_1_transpose_x_0"), val = bool(false)]; |
| bool attn_output_1_transpose_y_0 = const()[name = string("attn_output_1_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_1_cast_fp16 = transpose(perm = value_1_perm_0, x = var_132_cast_fp16)[name = string("transpose_317")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0_cast_fp16, y = value_1_cast_fp16)[name = string("attn_output_1_cast_fp16")]; |
| tensor<int32, [4]> var_135_perm_0 = const()[name = string("op_135_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_137 = const()[name = string("op_137"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_135_cast_fp16 = transpose(perm = var_135_perm_0, x = attn_output_1_cast_fp16)[name = string("transpose_316")]; |
| tensor<fp16, [1, 77, 1280]> var_138_cast_fp16 = reshape(shape = var_137, x = var_135_cast_fp16)[name = string("op_138_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51766912))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52995776))))[name = string("encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53006080)))]; |
| tensor<fp16, [1, 77, 1280]> linear_3_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_0_self_attn_out_proj_weight_to_fp16_palettized, x = var_138_cast_fp16)[name = string("linear_3_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_7_cast_fp16 = add(x = input_3_cast_fp16, y = linear_3_cast_fp16)[name = string("input_7_cast_fp16")]; |
| tensor<int32, [1]> input_9_axes_0 = const()[name = string("input_9_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53008704)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53011328)))]; |
| tensor<fp16, [1, 77, 1280]> input_9_cast_fp16 = layer_norm(axes = input_9_axes_0, beta = encoder_text_model_encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_0_layer_norm2_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53013952))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57929216))))[name = string("encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57970240)))]; |
| tensor<fp16, [1, 77, 5120]> linear_4_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_0_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_0_mlp_fc1_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = string("linear_4_cast_fp16")]; |
| string input_13_mode_0 = const()[name = string("input_13_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_13_cast_fp16 = gelu(mode = input_13_mode_0, x = linear_4_cast_fp16)[name = string("input_13_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57980544))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62895808))))[name = string("encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62906112)))]; |
| tensor<fp16, [1, 77, 1280]> linear_5_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_0_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_0_mlp_fc2_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = string("linear_5_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_15_cast_fp16 = add(x = input_7_cast_fp16, y = linear_5_cast_fp16)[name = string("input_15_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = string("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62908736)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62911360)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = encoder_text_model_encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_1_layer_norm1_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62913984))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64142848))))[name = string("encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64153152)))]; |
| tensor<fp16, [1, 77, 1280]> linear_6_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_6_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64155776))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65384640))))[name = string("encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65394944)))]; |
| tensor<fp16, [1, 77, 1280]> linear_7_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_7_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65397568))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66626432))))[name = string("encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66636736)))]; |
| tensor<fp16, [1, 77, 1280]> linear_8_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_7_cast_fp16)[name = string("linear_8_cast_fp16")]; |
| tensor<int32, [4]> var_181 = const()[name = string("op_181"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_182_cast_fp16 = reshape(shape = var_181, x = linear_6_cast_fp16)[name = string("op_182_cast_fp16")]; |
| tensor<int32, [4]> var_184 = const()[name = string("op_184"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_185_cast_fp16 = reshape(shape = var_184, x = linear_7_cast_fp16)[name = string("op_185_cast_fp16")]; |
| tensor<int32, [4]> var_187 = const()[name = string("op_187"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_188_cast_fp16 = reshape(shape = var_187, x = linear_8_cast_fp16)[name = string("op_188_cast_fp16")]; |
| tensor<int32, [4]> value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_3_cast_fp16 = mul(x = var_182_cast_fp16, y = var_11_to_fp16)[name = string("mul_3_cast_fp16")]; |
| bool matmul_1_transpose_y_0 = const()[name = string("matmul_1_transpose_y_0"), val = bool(true)]; |
| bool matmul_1_transpose_x_0 = const()[name = string("matmul_1_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_131 = transpose(perm = transpose_131_perm_0, x = var_185_cast_fp16)[name = string("transpose_314")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_130 = transpose(perm = transpose_130_perm_0, x = mul_3_cast_fp16)[name = string("transpose_315")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("matmul_1_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_1_cast_fp16 = add(x = matmul_1_cast_fp16, y = mul_0_to_fp16)[name = string("add_1_cast_fp16")]; |
| int32 softmax_1_axis_0 = const()[name = string("softmax_1_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_1_cast_fp16 = softmax(axis = softmax_1_axis_0, x = add_1_cast_fp16)[name = string("softmax_1_cast_fp16")]; |
| bool attn_output_5_transpose_x_0 = const()[name = string("attn_output_5_transpose_x_0"), val = bool(false)]; |
| bool attn_output_5_transpose_y_0 = const()[name = string("attn_output_5_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = var_188_cast_fp16)[name = string("transpose_313")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_1_cast_fp16, y = value_3_cast_fp16)[name = string("attn_output_5_cast_fp16")]; |
| tensor<int32, [4]> var_191_perm_0 = const()[name = string("op_191_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_193 = const()[name = string("op_193"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_191_cast_fp16 = transpose(perm = var_191_perm_0, x = attn_output_5_cast_fp16)[name = string("transpose_312")]; |
| tensor<fp16, [1, 77, 1280]> var_194_cast_fp16 = reshape(shape = var_193, x = var_191_cast_fp16)[name = string("op_194_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66639360))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67868224))))[name = string("encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67878528)))]; |
| tensor<fp16, [1, 77, 1280]> linear_9_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_1_self_attn_out_proj_weight_to_fp16_palettized, x = var_194_cast_fp16)[name = string("linear_9_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_19_cast_fp16 = add(x = input_15_cast_fp16, y = linear_9_cast_fp16)[name = string("input_19_cast_fp16")]; |
| tensor<int32, [1]> input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67881152)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67883776)))]; |
| tensor<fp16, [1, 77, 1280]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = encoder_text_model_encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_1_layer_norm2_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67886400))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72801664))))[name = string("encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72842688)))]; |
| tensor<fp16, [1, 77, 5120]> linear_10_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_1_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_1_mlp_fc1_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = string("linear_10_cast_fp16")]; |
| string input_25_mode_0 = const()[name = string("input_25_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_25_cast_fp16 = gelu(mode = input_25_mode_0, x = linear_10_cast_fp16)[name = string("input_25_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72852992))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77768256))))[name = string("encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77778560)))]; |
| tensor<fp16, [1, 77, 1280]> linear_11_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_1_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_1_mlp_fc2_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = string("linear_11_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_11_cast_fp16)[name = string("input_27_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = string("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77781184)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77783808)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = encoder_text_model_encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_2_layer_norm1_weight_to_fp16, x = input_27_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77786432))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79015296))))[name = string("encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79025600)))]; |
| tensor<fp16, [1, 77, 1280]> linear_12_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_12_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79028224))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80257088))))[name = string("encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80267392)))]; |
| tensor<fp16, [1, 77, 1280]> linear_13_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_13_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80270016))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81498880))))[name = string("encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81509184)))]; |
| tensor<fp16, [1, 77, 1280]> linear_14_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_13_cast_fp16)[name = string("linear_14_cast_fp16")]; |
| tensor<int32, [4]> var_237 = const()[name = string("op_237"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_238_cast_fp16 = reshape(shape = var_237, x = linear_12_cast_fp16)[name = string("op_238_cast_fp16")]; |
| tensor<int32, [4]> var_240 = const()[name = string("op_240"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_241_cast_fp16 = reshape(shape = var_240, x = linear_13_cast_fp16)[name = string("op_241_cast_fp16")]; |
| tensor<int32, [4]> var_243 = const()[name = string("op_243"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_244_cast_fp16 = reshape(shape = var_243, x = linear_14_cast_fp16)[name = string("op_244_cast_fp16")]; |
| tensor<int32, [4]> value_5_perm_0 = const()[name = string("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_5_cast_fp16 = mul(x = var_238_cast_fp16, y = var_11_to_fp16)[name = string("mul_5_cast_fp16")]; |
| bool matmul_2_transpose_y_0 = const()[name = string("matmul_2_transpose_y_0"), val = bool(true)]; |
| bool matmul_2_transpose_x_0 = const()[name = string("matmul_2_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_133 = transpose(perm = transpose_133_perm_0, x = var_241_cast_fp16)[name = string("transpose_310")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_132 = transpose(perm = transpose_132_perm_0, x = mul_5_cast_fp16)[name = string("transpose_311")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("matmul_2_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_2_cast_fp16 = add(x = matmul_2_cast_fp16, y = mul_0_to_fp16)[name = string("add_2_cast_fp16")]; |
| int32 softmax_2_axis_0 = const()[name = string("softmax_2_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_2_cast_fp16 = softmax(axis = softmax_2_axis_0, x = add_2_cast_fp16)[name = string("softmax_2_cast_fp16")]; |
| bool attn_output_9_transpose_x_0 = const()[name = string("attn_output_9_transpose_x_0"), val = bool(false)]; |
| bool attn_output_9_transpose_y_0 = const()[name = string("attn_output_9_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = var_244_cast_fp16)[name = string("transpose_309")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_2_cast_fp16, y = value_5_cast_fp16)[name = string("attn_output_9_cast_fp16")]; |
| tensor<int32, [4]> var_247_perm_0 = const()[name = string("op_247_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_249 = const()[name = string("op_249"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_247_cast_fp16 = transpose(perm = var_247_perm_0, x = attn_output_9_cast_fp16)[name = string("transpose_308")]; |
| tensor<fp16, [1, 77, 1280]> var_250_cast_fp16 = reshape(shape = var_249, x = var_247_cast_fp16)[name = string("op_250_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81511808))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82740672))))[name = string("encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82750976)))]; |
| tensor<fp16, [1, 77, 1280]> linear_15_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_2_self_attn_out_proj_weight_to_fp16_palettized, x = var_250_cast_fp16)[name = string("linear_15_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_31_cast_fp16 = add(x = input_27_cast_fp16, y = linear_15_cast_fp16)[name = string("input_31_cast_fp16")]; |
| tensor<int32, [1]> input_33_axes_0 = const()[name = string("input_33_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82753600)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82756224)))]; |
| tensor<fp16, [1, 77, 1280]> input_33_cast_fp16 = layer_norm(axes = input_33_axes_0, beta = encoder_text_model_encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_2_layer_norm2_weight_to_fp16, x = input_31_cast_fp16)[name = string("input_33_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82758848))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87674112))))[name = string("encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87715136)))]; |
| tensor<fp16, [1, 77, 5120]> linear_16_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_2_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_2_mlp_fc1_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = string("linear_16_cast_fp16")]; |
| string input_37_mode_0 = const()[name = string("input_37_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_37_cast_fp16 = gelu(mode = input_37_mode_0, x = linear_16_cast_fp16)[name = string("input_37_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87725440))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92640704))))[name = string("encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92651008)))]; |
| tensor<fp16, [1, 77, 1280]> linear_17_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_2_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_2_mlp_fc2_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = string("linear_17_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_17_cast_fp16)[name = string("input_39_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = string("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92653632)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92656256)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = encoder_text_model_encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_3_layer_norm1_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92658880))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93887744))))[name = string("encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93898048)))]; |
| tensor<fp16, [1, 77, 1280]> linear_18_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_18_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93900672))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95129536))))[name = string("encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95139840)))]; |
| tensor<fp16, [1, 77, 1280]> linear_19_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_19_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95142464))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96371328))))[name = string("encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96381632)))]; |
| tensor<fp16, [1, 77, 1280]> linear_20_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_19_cast_fp16)[name = string("linear_20_cast_fp16")]; |
| tensor<int32, [4]> var_293 = const()[name = string("op_293"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_294_cast_fp16 = reshape(shape = var_293, x = linear_18_cast_fp16)[name = string("op_294_cast_fp16")]; |
| tensor<int32, [4]> var_296 = const()[name = string("op_296"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_297_cast_fp16 = reshape(shape = var_296, x = linear_19_cast_fp16)[name = string("op_297_cast_fp16")]; |
| tensor<int32, [4]> var_299 = const()[name = string("op_299"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_300_cast_fp16 = reshape(shape = var_299, x = linear_20_cast_fp16)[name = string("op_300_cast_fp16")]; |
| tensor<int32, [4]> value_7_perm_0 = const()[name = string("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_7_cast_fp16 = mul(x = var_294_cast_fp16, y = var_11_to_fp16)[name = string("mul_7_cast_fp16")]; |
| bool matmul_3_transpose_y_0 = const()[name = string("matmul_3_transpose_y_0"), val = bool(true)]; |
| bool matmul_3_transpose_x_0 = const()[name = string("matmul_3_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_135 = transpose(perm = transpose_135_perm_0, x = var_297_cast_fp16)[name = string("transpose_306")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_134 = transpose(perm = transpose_134_perm_0, x = mul_7_cast_fp16)[name = string("transpose_307")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("matmul_3_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_3_cast_fp16 = add(x = matmul_3_cast_fp16, y = mul_0_to_fp16)[name = string("add_3_cast_fp16")]; |
| int32 softmax_3_axis_0 = const()[name = string("softmax_3_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_3_cast_fp16 = softmax(axis = softmax_3_axis_0, x = add_3_cast_fp16)[name = string("softmax_3_cast_fp16")]; |
| bool attn_output_13_transpose_x_0 = const()[name = string("attn_output_13_transpose_x_0"), val = bool(false)]; |
| bool attn_output_13_transpose_y_0 = const()[name = string("attn_output_13_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = var_300_cast_fp16)[name = string("transpose_305")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_3_cast_fp16, y = value_7_cast_fp16)[name = string("attn_output_13_cast_fp16")]; |
| tensor<int32, [4]> var_303_perm_0 = const()[name = string("op_303_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_305 = const()[name = string("op_305"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_303_cast_fp16 = transpose(perm = var_303_perm_0, x = attn_output_13_cast_fp16)[name = string("transpose_304")]; |
| tensor<fp16, [1, 77, 1280]> var_306_cast_fp16 = reshape(shape = var_305, x = var_303_cast_fp16)[name = string("op_306_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96384256))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97613120))))[name = string("encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97623424)))]; |
| tensor<fp16, [1, 77, 1280]> linear_21_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_3_self_attn_out_proj_weight_to_fp16_palettized, x = var_306_cast_fp16)[name = string("linear_21_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_43_cast_fp16 = add(x = input_39_cast_fp16, y = linear_21_cast_fp16)[name = string("input_43_cast_fp16")]; |
| tensor<int32, [1]> input_45_axes_0 = const()[name = string("input_45_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97626048)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97628672)))]; |
| tensor<fp16, [1, 77, 1280]> input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = encoder_text_model_encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_3_layer_norm2_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97631296))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102546560))))[name = string("encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102587584)))]; |
| tensor<fp16, [1, 77, 5120]> linear_22_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_3_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_3_mlp_fc1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = string("linear_22_cast_fp16")]; |
| string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = linear_22_cast_fp16)[name = string("input_49_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102597888))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107513152))))[name = string("encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107523456)))]; |
| tensor<fp16, [1, 77, 1280]> linear_23_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_3_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_3_mlp_fc2_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = string("linear_23_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_23_cast_fp16)[name = string("input_51_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = string("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107526080)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107528704)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = encoder_text_model_encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_4_layer_norm1_weight_to_fp16, x = input_51_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107531328))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108760192))))[name = string("encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108770496)))]; |
| tensor<fp16, [1, 77, 1280]> linear_24_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_24_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108773120))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110001984))))[name = string("encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110012288)))]; |
| tensor<fp16, [1, 77, 1280]> linear_25_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_25_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110014912))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111243776))))[name = string("encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111254080)))]; |
| tensor<fp16, [1, 77, 1280]> linear_26_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_25_cast_fp16)[name = string("linear_26_cast_fp16")]; |
| tensor<int32, [4]> var_349 = const()[name = string("op_349"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_350_cast_fp16 = reshape(shape = var_349, x = linear_24_cast_fp16)[name = string("op_350_cast_fp16")]; |
| tensor<int32, [4]> var_352 = const()[name = string("op_352"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_353_cast_fp16 = reshape(shape = var_352, x = linear_25_cast_fp16)[name = string("op_353_cast_fp16")]; |
| tensor<int32, [4]> var_355 = const()[name = string("op_355"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_356_cast_fp16 = reshape(shape = var_355, x = linear_26_cast_fp16)[name = string("op_356_cast_fp16")]; |
| tensor<int32, [4]> value_9_perm_0 = const()[name = string("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_9_cast_fp16 = mul(x = var_350_cast_fp16, y = var_11_to_fp16)[name = string("mul_9_cast_fp16")]; |
| bool matmul_4_transpose_y_0 = const()[name = string("matmul_4_transpose_y_0"), val = bool(true)]; |
| bool matmul_4_transpose_x_0 = const()[name = string("matmul_4_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_137 = transpose(perm = transpose_137_perm_0, x = var_353_cast_fp16)[name = string("transpose_302")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_136 = transpose(perm = transpose_136_perm_0, x = mul_9_cast_fp16)[name = string("transpose_303")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("matmul_4_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_4_cast_fp16 = add(x = matmul_4_cast_fp16, y = mul_0_to_fp16)[name = string("add_4_cast_fp16")]; |
| int32 softmax_4_axis_0 = const()[name = string("softmax_4_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_4_cast_fp16 = softmax(axis = softmax_4_axis_0, x = add_4_cast_fp16)[name = string("softmax_4_cast_fp16")]; |
| bool attn_output_17_transpose_x_0 = const()[name = string("attn_output_17_transpose_x_0"), val = bool(false)]; |
| bool attn_output_17_transpose_y_0 = const()[name = string("attn_output_17_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = var_356_cast_fp16)[name = string("transpose_301")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = softmax_4_cast_fp16, y = value_9_cast_fp16)[name = string("attn_output_17_cast_fp16")]; |
| tensor<int32, [4]> var_359_perm_0 = const()[name = string("op_359_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_361 = const()[name = string("op_361"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_359_cast_fp16 = transpose(perm = var_359_perm_0, x = attn_output_17_cast_fp16)[name = string("transpose_300")]; |
| tensor<fp16, [1, 77, 1280]> var_362_cast_fp16 = reshape(shape = var_361, x = var_359_cast_fp16)[name = string("op_362_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111256704))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112485568))))[name = string("encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112495872)))]; |
| tensor<fp16, [1, 77, 1280]> linear_27_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_4_self_attn_out_proj_weight_to_fp16_palettized, x = var_362_cast_fp16)[name = string("linear_27_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_55_cast_fp16 = add(x = input_51_cast_fp16, y = linear_27_cast_fp16)[name = string("input_55_cast_fp16")]; |
| tensor<int32, [1]> input_57_axes_0 = const()[name = string("input_57_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112498496)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112501120)))]; |
| tensor<fp16, [1, 77, 1280]> input_57_cast_fp16 = layer_norm(axes = input_57_axes_0, beta = encoder_text_model_encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_4_layer_norm2_weight_to_fp16, x = input_55_cast_fp16)[name = string("input_57_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112503744))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117419008))))[name = string("encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117460032)))]; |
| tensor<fp16, [1, 77, 5120]> linear_28_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_4_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_4_mlp_fc1_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = string("linear_28_cast_fp16")]; |
| string input_61_mode_0 = const()[name = string("input_61_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_61_cast_fp16 = gelu(mode = input_61_mode_0, x = linear_28_cast_fp16)[name = string("input_61_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117470336))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122385600))))[name = string("encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122395904)))]; |
| tensor<fp16, [1, 77, 1280]> linear_29_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_4_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_4_mlp_fc2_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = string("linear_29_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_63_cast_fp16 = add(x = input_55_cast_fp16, y = linear_29_cast_fp16)[name = string("input_63_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = string("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122398528)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122401152)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = encoder_text_model_encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_5_layer_norm1_weight_to_fp16, x = input_63_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122403776))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123632640))))[name = string("encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123642944)))]; |
| tensor<fp16, [1, 77, 1280]> linear_30_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = string("linear_30_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123645568))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124874432))))[name = string("encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124884736)))]; |
| tensor<fp16, [1, 77, 1280]> linear_31_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = string("linear_31_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124887360))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126116224))))[name = string("encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126126528)))]; |
| tensor<fp16, [1, 77, 1280]> linear_32_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_31_cast_fp16)[name = string("linear_32_cast_fp16")]; |
| tensor<int32, [4]> var_405 = const()[name = string("op_405"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_406_cast_fp16 = reshape(shape = var_405, x = linear_30_cast_fp16)[name = string("op_406_cast_fp16")]; |
| tensor<int32, [4]> var_408 = const()[name = string("op_408"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_409_cast_fp16 = reshape(shape = var_408, x = linear_31_cast_fp16)[name = string("op_409_cast_fp16")]; |
| tensor<int32, [4]> var_411 = const()[name = string("op_411"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_412_cast_fp16 = reshape(shape = var_411, x = linear_32_cast_fp16)[name = string("op_412_cast_fp16")]; |
| tensor<int32, [4]> value_11_perm_0 = const()[name = string("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_11_cast_fp16 = mul(x = var_406_cast_fp16, y = var_11_to_fp16)[name = string("mul_11_cast_fp16")]; |
| bool matmul_5_transpose_y_0 = const()[name = string("matmul_5_transpose_y_0"), val = bool(true)]; |
| bool matmul_5_transpose_x_0 = const()[name = string("matmul_5_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_139 = transpose(perm = transpose_139_perm_0, x = var_409_cast_fp16)[name = string("transpose_298")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_138 = transpose(perm = transpose_138_perm_0, x = mul_11_cast_fp16)[name = string("transpose_299")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("matmul_5_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_5_cast_fp16 = add(x = matmul_5_cast_fp16, y = mul_0_to_fp16)[name = string("add_5_cast_fp16")]; |
| int32 softmax_5_axis_0 = const()[name = string("softmax_5_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_5_cast_fp16 = softmax(axis = softmax_5_axis_0, x = add_5_cast_fp16)[name = string("softmax_5_cast_fp16")]; |
| bool attn_output_21_transpose_x_0 = const()[name = string("attn_output_21_transpose_x_0"), val = bool(false)]; |
| bool attn_output_21_transpose_y_0 = const()[name = string("attn_output_21_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = var_412_cast_fp16)[name = string("transpose_297")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = softmax_5_cast_fp16, y = value_11_cast_fp16)[name = string("attn_output_21_cast_fp16")]; |
| tensor<int32, [4]> var_415_perm_0 = const()[name = string("op_415_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_417 = const()[name = string("op_417"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_415_cast_fp16 = transpose(perm = var_415_perm_0, x = attn_output_21_cast_fp16)[name = string("transpose_296")]; |
| tensor<fp16, [1, 77, 1280]> var_418_cast_fp16 = reshape(shape = var_417, x = var_415_cast_fp16)[name = string("op_418_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126129152))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127358016))))[name = string("encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127368320)))]; |
| tensor<fp16, [1, 77, 1280]> linear_33_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_5_self_attn_out_proj_weight_to_fp16_palettized, x = var_418_cast_fp16)[name = string("linear_33_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_67_cast_fp16 = add(x = input_63_cast_fp16, y = linear_33_cast_fp16)[name = string("input_67_cast_fp16")]; |
| tensor<int32, [1]> input_69_axes_0 = const()[name = string("input_69_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127370944)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127373568)))]; |
| tensor<fp16, [1, 77, 1280]> input_69_cast_fp16 = layer_norm(axes = input_69_axes_0, beta = encoder_text_model_encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_5_layer_norm2_weight_to_fp16, x = input_67_cast_fp16)[name = string("input_69_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127376192))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132291456))))[name = string("encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132332480)))]; |
| tensor<fp16, [1, 77, 5120]> linear_34_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_5_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_5_mlp_fc1_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = string("linear_34_cast_fp16")]; |
| string input_73_mode_0 = const()[name = string("input_73_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_73_cast_fp16 = gelu(mode = input_73_mode_0, x = linear_34_cast_fp16)[name = string("input_73_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132342784))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137258048))))[name = string("encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137268352)))]; |
| tensor<fp16, [1, 77, 1280]> linear_35_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_5_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_5_mlp_fc2_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = string("linear_35_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_75_cast_fp16 = add(x = input_67_cast_fp16, y = linear_35_cast_fp16)[name = string("input_75_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = string("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137270976)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137273600)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = encoder_text_model_encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_6_layer_norm1_weight_to_fp16, x = input_75_cast_fp16)[name = string("hidden_states_37_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137276224))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138505088))))[name = string("encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138515392)))]; |
| tensor<fp16, [1, 77, 1280]> linear_36_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = string("linear_36_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138518016))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139746880))))[name = string("encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139757184)))]; |
| tensor<fp16, [1, 77, 1280]> linear_37_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = string("linear_37_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139759808))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140988672))))[name = string("encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140998976)))]; |
| tensor<fp16, [1, 77, 1280]> linear_38_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_37_cast_fp16)[name = string("linear_38_cast_fp16")]; |
| tensor<int32, [4]> var_461 = const()[name = string("op_461"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_462_cast_fp16 = reshape(shape = var_461, x = linear_36_cast_fp16)[name = string("op_462_cast_fp16")]; |
| tensor<int32, [4]> var_464 = const()[name = string("op_464"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_465_cast_fp16 = reshape(shape = var_464, x = linear_37_cast_fp16)[name = string("op_465_cast_fp16")]; |
| tensor<int32, [4]> var_467 = const()[name = string("op_467"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_468_cast_fp16 = reshape(shape = var_467, x = linear_38_cast_fp16)[name = string("op_468_cast_fp16")]; |
| tensor<int32, [4]> value_13_perm_0 = const()[name = string("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_13_cast_fp16 = mul(x = var_462_cast_fp16, y = var_11_to_fp16)[name = string("mul_13_cast_fp16")]; |
| bool matmul_6_transpose_y_0 = const()[name = string("matmul_6_transpose_y_0"), val = bool(true)]; |
| bool matmul_6_transpose_x_0 = const()[name = string("matmul_6_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_141 = transpose(perm = transpose_141_perm_0, x = var_465_cast_fp16)[name = string("transpose_294")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_140 = transpose(perm = transpose_140_perm_0, x = mul_13_cast_fp16)[name = string("transpose_295")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("matmul_6_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_6_cast_fp16 = add(x = matmul_6_cast_fp16, y = mul_0_to_fp16)[name = string("add_6_cast_fp16")]; |
| int32 softmax_6_axis_0 = const()[name = string("softmax_6_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_6_cast_fp16 = softmax(axis = softmax_6_axis_0, x = add_6_cast_fp16)[name = string("softmax_6_cast_fp16")]; |
| bool attn_output_25_transpose_x_0 = const()[name = string("attn_output_25_transpose_x_0"), val = bool(false)]; |
| bool attn_output_25_transpose_y_0 = const()[name = string("attn_output_25_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = var_468_cast_fp16)[name = string("transpose_293")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = softmax_6_cast_fp16, y = value_13_cast_fp16)[name = string("attn_output_25_cast_fp16")]; |
| tensor<int32, [4]> var_471_perm_0 = const()[name = string("op_471_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_473 = const()[name = string("op_473"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_471_cast_fp16 = transpose(perm = var_471_perm_0, x = attn_output_25_cast_fp16)[name = string("transpose_292")]; |
| tensor<fp16, [1, 77, 1280]> var_474_cast_fp16 = reshape(shape = var_473, x = var_471_cast_fp16)[name = string("op_474_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141001600))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142230464))))[name = string("encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142240768)))]; |
| tensor<fp16, [1, 77, 1280]> linear_39_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_6_self_attn_out_proj_weight_to_fp16_palettized, x = var_474_cast_fp16)[name = string("linear_39_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_79_cast_fp16 = add(x = input_75_cast_fp16, y = linear_39_cast_fp16)[name = string("input_79_cast_fp16")]; |
| tensor<int32, [1]> input_81_axes_0 = const()[name = string("input_81_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142243392)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142246016)))]; |
| tensor<fp16, [1, 77, 1280]> input_81_cast_fp16 = layer_norm(axes = input_81_axes_0, beta = encoder_text_model_encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_6_layer_norm2_weight_to_fp16, x = input_79_cast_fp16)[name = string("input_81_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142248640))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147163904))))[name = string("encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147204928)))]; |
| tensor<fp16, [1, 77, 5120]> linear_40_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_6_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_6_mlp_fc1_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = string("linear_40_cast_fp16")]; |
| string input_85_mode_0 = const()[name = string("input_85_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_85_cast_fp16 = gelu(mode = input_85_mode_0, x = linear_40_cast_fp16)[name = string("input_85_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147215232))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152130496))))[name = string("encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152140800)))]; |
| tensor<fp16, [1, 77, 1280]> linear_41_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_6_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_6_mlp_fc2_weight_to_fp16_palettized, x = input_85_cast_fp16)[name = string("linear_41_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_87_cast_fp16 = add(x = input_79_cast_fp16, y = linear_41_cast_fp16)[name = string("input_87_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = string("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152143424)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152146048)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = encoder_text_model_encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_7_layer_norm1_weight_to_fp16, x = input_87_cast_fp16)[name = string("hidden_states_43_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152148672))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153377536))))[name = string("encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153387840)))]; |
| tensor<fp16, [1, 77, 1280]> linear_42_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = string("linear_42_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153390464))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154619328))))[name = string("encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154629632)))]; |
| tensor<fp16, [1, 77, 1280]> linear_43_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = string("linear_43_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154632256))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155861120))))[name = string("encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155871424)))]; |
| tensor<fp16, [1, 77, 1280]> linear_44_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_43_cast_fp16)[name = string("linear_44_cast_fp16")]; |
| tensor<int32, [4]> var_517 = const()[name = string("op_517"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_518_cast_fp16 = reshape(shape = var_517, x = linear_42_cast_fp16)[name = string("op_518_cast_fp16")]; |
| tensor<int32, [4]> var_520 = const()[name = string("op_520"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_521_cast_fp16 = reshape(shape = var_520, x = linear_43_cast_fp16)[name = string("op_521_cast_fp16")]; |
| tensor<int32, [4]> var_523 = const()[name = string("op_523"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_524_cast_fp16 = reshape(shape = var_523, x = linear_44_cast_fp16)[name = string("op_524_cast_fp16")]; |
| tensor<int32, [4]> value_15_perm_0 = const()[name = string("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_15_cast_fp16 = mul(x = var_518_cast_fp16, y = var_11_to_fp16)[name = string("mul_15_cast_fp16")]; |
| bool matmul_7_transpose_y_0 = const()[name = string("matmul_7_transpose_y_0"), val = bool(true)]; |
| bool matmul_7_transpose_x_0 = const()[name = string("matmul_7_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_143 = transpose(perm = transpose_143_perm_0, x = var_521_cast_fp16)[name = string("transpose_290")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_142 = transpose(perm = transpose_142_perm_0, x = mul_15_cast_fp16)[name = string("transpose_291")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("matmul_7_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_7_cast_fp16 = add(x = matmul_7_cast_fp16, y = mul_0_to_fp16)[name = string("add_7_cast_fp16")]; |
| int32 softmax_7_axis_0 = const()[name = string("softmax_7_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_7_cast_fp16 = softmax(axis = softmax_7_axis_0, x = add_7_cast_fp16)[name = string("softmax_7_cast_fp16")]; |
| bool attn_output_29_transpose_x_0 = const()[name = string("attn_output_29_transpose_x_0"), val = bool(false)]; |
| bool attn_output_29_transpose_y_0 = const()[name = string("attn_output_29_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = var_524_cast_fp16)[name = string("transpose_289")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = softmax_7_cast_fp16, y = value_15_cast_fp16)[name = string("attn_output_29_cast_fp16")]; |
| tensor<int32, [4]> var_527_perm_0 = const()[name = string("op_527_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_529 = const()[name = string("op_529"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_527_cast_fp16 = transpose(perm = var_527_perm_0, x = attn_output_29_cast_fp16)[name = string("transpose_288")]; |
| tensor<fp16, [1, 77, 1280]> var_530_cast_fp16 = reshape(shape = var_529, x = var_527_cast_fp16)[name = string("op_530_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155874048))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157102912))))[name = string("encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157113216)))]; |
| tensor<fp16, [1, 77, 1280]> linear_45_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_7_self_attn_out_proj_weight_to_fp16_palettized, x = var_530_cast_fp16)[name = string("linear_45_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_91_cast_fp16 = add(x = input_87_cast_fp16, y = linear_45_cast_fp16)[name = string("input_91_cast_fp16")]; |
| tensor<int32, [1]> input_93_axes_0 = const()[name = string("input_93_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157115840)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157118464)))]; |
| tensor<fp16, [1, 77, 1280]> input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = encoder_text_model_encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_7_layer_norm2_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157121088))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162036352))))[name = string("encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162077376)))]; |
| tensor<fp16, [1, 77, 5120]> linear_46_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_7_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_7_mlp_fc1_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = string("linear_46_cast_fp16")]; |
| string input_97_mode_0 = const()[name = string("input_97_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_97_cast_fp16 = gelu(mode = input_97_mode_0, x = linear_46_cast_fp16)[name = string("input_97_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162087680))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167002944))))[name = string("encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167013248)))]; |
| tensor<fp16, [1, 77, 1280]> linear_47_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_7_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_7_mlp_fc2_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = string("linear_47_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_99_cast_fp16 = add(x = input_91_cast_fp16, y = linear_47_cast_fp16)[name = string("input_99_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = string("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167015872)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167018496)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = encoder_text_model_encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_8_layer_norm1_weight_to_fp16, x = input_99_cast_fp16)[name = string("hidden_states_49_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167021120))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168249984))))[name = string("encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168260288)))]; |
| tensor<fp16, [1, 77, 1280]> linear_48_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = string("linear_48_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168262912))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169491776))))[name = string("encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169502080)))]; |
| tensor<fp16, [1, 77, 1280]> linear_49_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = string("linear_49_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169504704))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170733568))))[name = string("encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170743872)))]; |
| tensor<fp16, [1, 77, 1280]> linear_50_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_49_cast_fp16)[name = string("linear_50_cast_fp16")]; |
| tensor<int32, [4]> var_573 = const()[name = string("op_573"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_574_cast_fp16 = reshape(shape = var_573, x = linear_48_cast_fp16)[name = string("op_574_cast_fp16")]; |
| tensor<int32, [4]> var_576 = const()[name = string("op_576"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_577_cast_fp16 = reshape(shape = var_576, x = linear_49_cast_fp16)[name = string("op_577_cast_fp16")]; |
| tensor<int32, [4]> var_579 = const()[name = string("op_579"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_580_cast_fp16 = reshape(shape = var_579, x = linear_50_cast_fp16)[name = string("op_580_cast_fp16")]; |
| tensor<int32, [4]> value_17_perm_0 = const()[name = string("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_17_cast_fp16 = mul(x = var_574_cast_fp16, y = var_11_to_fp16)[name = string("mul_17_cast_fp16")]; |
| bool matmul_8_transpose_y_0 = const()[name = string("matmul_8_transpose_y_0"), val = bool(true)]; |
| bool matmul_8_transpose_x_0 = const()[name = string("matmul_8_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_144_perm_0 = const()[name = string("transpose_144_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_145_perm_0 = const()[name = string("transpose_145_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_145 = transpose(perm = transpose_145_perm_0, x = var_577_cast_fp16)[name = string("transpose_286")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_144 = transpose(perm = transpose_144_perm_0, x = mul_17_cast_fp16)[name = string("transpose_287")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_144, y = transpose_145)[name = string("matmul_8_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_8_cast_fp16 = add(x = matmul_8_cast_fp16, y = mul_0_to_fp16)[name = string("add_8_cast_fp16")]; |
| int32 softmax_8_axis_0 = const()[name = string("softmax_8_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_8_cast_fp16 = softmax(axis = softmax_8_axis_0, x = add_8_cast_fp16)[name = string("softmax_8_cast_fp16")]; |
| bool attn_output_33_transpose_x_0 = const()[name = string("attn_output_33_transpose_x_0"), val = bool(false)]; |
| bool attn_output_33_transpose_y_0 = const()[name = string("attn_output_33_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = var_580_cast_fp16)[name = string("transpose_285")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_33_cast_fp16 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = softmax_8_cast_fp16, y = value_17_cast_fp16)[name = string("attn_output_33_cast_fp16")]; |
| tensor<int32, [4]> var_583_perm_0 = const()[name = string("op_583_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_585 = const()[name = string("op_585"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_583_cast_fp16 = transpose(perm = var_583_perm_0, x = attn_output_33_cast_fp16)[name = string("transpose_284")]; |
| tensor<fp16, [1, 77, 1280]> var_586_cast_fp16 = reshape(shape = var_585, x = var_583_cast_fp16)[name = string("op_586_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170746496))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171975360))))[name = string("encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171985664)))]; |
| tensor<fp16, [1, 77, 1280]> linear_51_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_8_self_attn_out_proj_weight_to_fp16_palettized, x = var_586_cast_fp16)[name = string("linear_51_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_103_cast_fp16 = add(x = input_99_cast_fp16, y = linear_51_cast_fp16)[name = string("input_103_cast_fp16")]; |
| tensor<int32, [1]> input_105_axes_0 = const()[name = string("input_105_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171988288)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171990912)))]; |
| tensor<fp16, [1, 77, 1280]> input_105_cast_fp16 = layer_norm(axes = input_105_axes_0, beta = encoder_text_model_encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_8_layer_norm2_weight_to_fp16, x = input_103_cast_fp16)[name = string("input_105_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171993536))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176908800))))[name = string("encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176949824)))]; |
| tensor<fp16, [1, 77, 5120]> linear_52_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_8_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_8_mlp_fc1_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("linear_52_cast_fp16")]; |
| string input_109_mode_0 = const()[name = string("input_109_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = linear_52_cast_fp16)[name = string("input_109_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176960128))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181875392))))[name = string("encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181885696)))]; |
| tensor<fp16, [1, 77, 1280]> linear_53_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_8_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_8_mlp_fc2_weight_to_fp16_palettized, x = input_109_cast_fp16)[name = string("linear_53_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_111_cast_fp16 = add(x = input_103_cast_fp16, y = linear_53_cast_fp16)[name = string("input_111_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = string("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181888320)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181890944)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = encoder_text_model_encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_9_layer_norm1_weight_to_fp16, x = input_111_cast_fp16)[name = string("hidden_states_55_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181893568))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183122432))))[name = string("encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183132736)))]; |
| tensor<fp16, [1, 77, 1280]> linear_54_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = string("linear_54_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183135360))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184364224))))[name = string("encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184374528)))]; |
| tensor<fp16, [1, 77, 1280]> linear_55_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = string("linear_55_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184377152))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185606016))))[name = string("encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185616320)))]; |
| tensor<fp16, [1, 77, 1280]> linear_56_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_55_cast_fp16)[name = string("linear_56_cast_fp16")]; |
| tensor<int32, [4]> var_629 = const()[name = string("op_629"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_630_cast_fp16 = reshape(shape = var_629, x = linear_54_cast_fp16)[name = string("op_630_cast_fp16")]; |
| tensor<int32, [4]> var_632 = const()[name = string("op_632"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_633_cast_fp16 = reshape(shape = var_632, x = linear_55_cast_fp16)[name = string("op_633_cast_fp16")]; |
| tensor<int32, [4]> var_635 = const()[name = string("op_635"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_636_cast_fp16 = reshape(shape = var_635, x = linear_56_cast_fp16)[name = string("op_636_cast_fp16")]; |
| tensor<int32, [4]> value_19_perm_0 = const()[name = string("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_19_cast_fp16 = mul(x = var_630_cast_fp16, y = var_11_to_fp16)[name = string("mul_19_cast_fp16")]; |
| bool matmul_9_transpose_y_0 = const()[name = string("matmul_9_transpose_y_0"), val = bool(true)]; |
| bool matmul_9_transpose_x_0 = const()[name = string("matmul_9_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_146_perm_0 = const()[name = string("transpose_146_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_147_perm_0 = const()[name = string("transpose_147_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_147 = transpose(perm = transpose_147_perm_0, x = var_633_cast_fp16)[name = string("transpose_282")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_146 = transpose(perm = transpose_146_perm_0, x = mul_19_cast_fp16)[name = string("transpose_283")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = transpose_146, y = transpose_147)[name = string("matmul_9_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_9_cast_fp16 = add(x = matmul_9_cast_fp16, y = mul_0_to_fp16)[name = string("add_9_cast_fp16")]; |
| int32 softmax_9_axis_0 = const()[name = string("softmax_9_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_9_cast_fp16 = softmax(axis = softmax_9_axis_0, x = add_9_cast_fp16)[name = string("softmax_9_cast_fp16")]; |
| bool attn_output_37_transpose_x_0 = const()[name = string("attn_output_37_transpose_x_0"), val = bool(false)]; |
| bool attn_output_37_transpose_y_0 = const()[name = string("attn_output_37_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = var_636_cast_fp16)[name = string("transpose_281")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = softmax_9_cast_fp16, y = value_19_cast_fp16)[name = string("attn_output_37_cast_fp16")]; |
| tensor<int32, [4]> var_639_perm_0 = const()[name = string("op_639_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_641 = const()[name = string("op_641"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_639_cast_fp16 = transpose(perm = var_639_perm_0, x = attn_output_37_cast_fp16)[name = string("transpose_280")]; |
| tensor<fp16, [1, 77, 1280]> var_642_cast_fp16 = reshape(shape = var_641, x = var_639_cast_fp16)[name = string("op_642_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185618944))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186847808))))[name = string("encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186858112)))]; |
| tensor<fp16, [1, 77, 1280]> linear_57_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_9_self_attn_out_proj_weight_to_fp16_palettized, x = var_642_cast_fp16)[name = string("linear_57_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_115_cast_fp16 = add(x = input_111_cast_fp16, y = linear_57_cast_fp16)[name = string("input_115_cast_fp16")]; |
| tensor<int32, [1]> input_117_axes_0 = const()[name = string("input_117_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186860736)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186863360)))]; |
| tensor<fp16, [1, 77, 1280]> input_117_cast_fp16 = layer_norm(axes = input_117_axes_0, beta = encoder_text_model_encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_9_layer_norm2_weight_to_fp16, x = input_115_cast_fp16)[name = string("input_117_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186865984))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191781248))))[name = string("encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191822272)))]; |
| tensor<fp16, [1, 77, 5120]> linear_58_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_9_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_9_mlp_fc1_weight_to_fp16_palettized, x = input_117_cast_fp16)[name = string("linear_58_cast_fp16")]; |
| string input_121_mode_0 = const()[name = string("input_121_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_121_cast_fp16 = gelu(mode = input_121_mode_0, x = linear_58_cast_fp16)[name = string("input_121_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191832576))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196747840))))[name = string("encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196758144)))]; |
| tensor<fp16, [1, 77, 1280]> linear_59_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_9_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_9_mlp_fc2_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = string("linear_59_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_123_cast_fp16 = add(x = input_115_cast_fp16, y = linear_59_cast_fp16)[name = string("input_123_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = string("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196760768)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196763392)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = encoder_text_model_encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_10_layer_norm1_weight_to_fp16, x = input_123_cast_fp16)[name = string("hidden_states_61_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196766016))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197994880))))[name = string("encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198005184)))]; |
| tensor<fp16, [1, 77, 1280]> linear_60_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = string("linear_60_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198007808))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199236672))))[name = string("encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199246976)))]; |
| tensor<fp16, [1, 77, 1280]> linear_61_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = string("linear_61_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199249600))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200478464))))[name = string("encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200488768)))]; |
| tensor<fp16, [1, 77, 1280]> linear_62_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_61_cast_fp16)[name = string("linear_62_cast_fp16")]; |
| tensor<int32, [4]> var_685 = const()[name = string("op_685"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_686_cast_fp16 = reshape(shape = var_685, x = linear_60_cast_fp16)[name = string("op_686_cast_fp16")]; |
| tensor<int32, [4]> var_688 = const()[name = string("op_688"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_689_cast_fp16 = reshape(shape = var_688, x = linear_61_cast_fp16)[name = string("op_689_cast_fp16")]; |
| tensor<int32, [4]> var_691 = const()[name = string("op_691"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_692_cast_fp16 = reshape(shape = var_691, x = linear_62_cast_fp16)[name = string("op_692_cast_fp16")]; |
| tensor<int32, [4]> value_21_perm_0 = const()[name = string("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_21_cast_fp16 = mul(x = var_686_cast_fp16, y = var_11_to_fp16)[name = string("mul_21_cast_fp16")]; |
| bool matmul_10_transpose_y_0 = const()[name = string("matmul_10_transpose_y_0"), val = bool(true)]; |
| bool matmul_10_transpose_x_0 = const()[name = string("matmul_10_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_148_perm_0 = const()[name = string("transpose_148_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_149_perm_0 = const()[name = string("transpose_149_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_149 = transpose(perm = transpose_149_perm_0, x = var_689_cast_fp16)[name = string("transpose_278")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_148 = transpose(perm = transpose_148_perm_0, x = mul_21_cast_fp16)[name = string("transpose_279")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_148, y = transpose_149)[name = string("matmul_10_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_10_cast_fp16 = add(x = matmul_10_cast_fp16, y = mul_0_to_fp16)[name = string("add_10_cast_fp16")]; |
| int32 softmax_10_axis_0 = const()[name = string("softmax_10_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_10_cast_fp16 = softmax(axis = softmax_10_axis_0, x = add_10_cast_fp16)[name = string("softmax_10_cast_fp16")]; |
| bool attn_output_41_transpose_x_0 = const()[name = string("attn_output_41_transpose_x_0"), val = bool(false)]; |
| bool attn_output_41_transpose_y_0 = const()[name = string("attn_output_41_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = var_692_cast_fp16)[name = string("transpose_277")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_41_cast_fp16 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = softmax_10_cast_fp16, y = value_21_cast_fp16)[name = string("attn_output_41_cast_fp16")]; |
| tensor<int32, [4]> var_695_perm_0 = const()[name = string("op_695_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_697 = const()[name = string("op_697"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_695_cast_fp16 = transpose(perm = var_695_perm_0, x = attn_output_41_cast_fp16)[name = string("transpose_276")]; |
| tensor<fp16, [1, 77, 1280]> var_698_cast_fp16 = reshape(shape = var_697, x = var_695_cast_fp16)[name = string("op_698_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200491392))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201720256))))[name = string("encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201730560)))]; |
| tensor<fp16, [1, 77, 1280]> linear_63_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_10_self_attn_out_proj_weight_to_fp16_palettized, x = var_698_cast_fp16)[name = string("linear_63_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_127_cast_fp16 = add(x = input_123_cast_fp16, y = linear_63_cast_fp16)[name = string("input_127_cast_fp16")]; |
| tensor<int32, [1]> input_129_axes_0 = const()[name = string("input_129_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201733184)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201735808)))]; |
| tensor<fp16, [1, 77, 1280]> input_129_cast_fp16 = layer_norm(axes = input_129_axes_0, beta = encoder_text_model_encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_10_layer_norm2_weight_to_fp16, x = input_127_cast_fp16)[name = string("input_129_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201738432))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206653696))))[name = string("encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206694720)))]; |
| tensor<fp16, [1, 77, 5120]> linear_64_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_10_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_10_mlp_fc1_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = string("linear_64_cast_fp16")]; |
| string input_133_mode_0 = const()[name = string("input_133_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_133_cast_fp16 = gelu(mode = input_133_mode_0, x = linear_64_cast_fp16)[name = string("input_133_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206705024))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211620288))))[name = string("encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211630592)))]; |
| tensor<fp16, [1, 77, 1280]> linear_65_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_10_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_10_mlp_fc2_weight_to_fp16_palettized, x = input_133_cast_fp16)[name = string("linear_65_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_135_cast_fp16 = add(x = input_127_cast_fp16, y = linear_65_cast_fp16)[name = string("input_135_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = string("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211633216)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211635840)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = encoder_text_model_encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_11_layer_norm1_weight_to_fp16, x = input_135_cast_fp16)[name = string("hidden_states_67_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211638464))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212867328))))[name = string("encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212877632)))]; |
| tensor<fp16, [1, 77, 1280]> linear_66_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = string("linear_66_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212880256))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214109120))))[name = string("encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214119424)))]; |
| tensor<fp16, [1, 77, 1280]> linear_67_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = string("linear_67_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214122048))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215350912))))[name = string("encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215361216)))]; |
| tensor<fp16, [1, 77, 1280]> linear_68_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_67_cast_fp16)[name = string("linear_68_cast_fp16")]; |
| tensor<int32, [4]> var_741 = const()[name = string("op_741"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_742_cast_fp16 = reshape(shape = var_741, x = linear_66_cast_fp16)[name = string("op_742_cast_fp16")]; |
| tensor<int32, [4]> var_744 = const()[name = string("op_744"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_745_cast_fp16 = reshape(shape = var_744, x = linear_67_cast_fp16)[name = string("op_745_cast_fp16")]; |
| tensor<int32, [4]> var_747 = const()[name = string("op_747"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_748_cast_fp16 = reshape(shape = var_747, x = linear_68_cast_fp16)[name = string("op_748_cast_fp16")]; |
| tensor<int32, [4]> value_23_perm_0 = const()[name = string("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_23_cast_fp16 = mul(x = var_742_cast_fp16, y = var_11_to_fp16)[name = string("mul_23_cast_fp16")]; |
| bool matmul_11_transpose_y_0 = const()[name = string("matmul_11_transpose_y_0"), val = bool(true)]; |
| bool matmul_11_transpose_x_0 = const()[name = string("matmul_11_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_150_perm_0 = const()[name = string("transpose_150_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_151_perm_0 = const()[name = string("transpose_151_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_151 = transpose(perm = transpose_151_perm_0, x = var_745_cast_fp16)[name = string("transpose_274")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_150 = transpose(perm = transpose_150_perm_0, x = mul_23_cast_fp16)[name = string("transpose_275")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = transpose_150, y = transpose_151)[name = string("matmul_11_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_11_cast_fp16 = add(x = matmul_11_cast_fp16, y = mul_0_to_fp16)[name = string("add_11_cast_fp16")]; |
| int32 softmax_11_axis_0 = const()[name = string("softmax_11_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_11_cast_fp16 = softmax(axis = softmax_11_axis_0, x = add_11_cast_fp16)[name = string("softmax_11_cast_fp16")]; |
| bool attn_output_45_transpose_x_0 = const()[name = string("attn_output_45_transpose_x_0"), val = bool(false)]; |
| bool attn_output_45_transpose_y_0 = const()[name = string("attn_output_45_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = var_748_cast_fp16)[name = string("transpose_273")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_45_cast_fp16 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = softmax_11_cast_fp16, y = value_23_cast_fp16)[name = string("attn_output_45_cast_fp16")]; |
| tensor<int32, [4]> var_751_perm_0 = const()[name = string("op_751_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_753 = const()[name = string("op_753"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_751_cast_fp16 = transpose(perm = var_751_perm_0, x = attn_output_45_cast_fp16)[name = string("transpose_272")]; |
| tensor<fp16, [1, 77, 1280]> var_754_cast_fp16 = reshape(shape = var_753, x = var_751_cast_fp16)[name = string("op_754_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215363840))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216592704))))[name = string("encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216603008)))]; |
| tensor<fp16, [1, 77, 1280]> linear_69_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_11_self_attn_out_proj_weight_to_fp16_palettized, x = var_754_cast_fp16)[name = string("linear_69_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_139_cast_fp16 = add(x = input_135_cast_fp16, y = linear_69_cast_fp16)[name = string("input_139_cast_fp16")]; |
| tensor<int32, [1]> input_141_axes_0 = const()[name = string("input_141_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216605632)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216608256)))]; |
| tensor<fp16, [1, 77, 1280]> input_141_cast_fp16 = layer_norm(axes = input_141_axes_0, beta = encoder_text_model_encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_11_layer_norm2_weight_to_fp16, x = input_139_cast_fp16)[name = string("input_141_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216610880))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221526144))))[name = string("encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221567168)))]; |
| tensor<fp16, [1, 77, 5120]> linear_70_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_11_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_11_mlp_fc1_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = string("linear_70_cast_fp16")]; |
| string input_145_mode_0 = const()[name = string("input_145_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_145_cast_fp16 = gelu(mode = input_145_mode_0, x = linear_70_cast_fp16)[name = string("input_145_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221577472))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226492736))))[name = string("encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226503040)))]; |
| tensor<fp16, [1, 77, 1280]> linear_71_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_11_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_11_mlp_fc2_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = string("linear_71_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_147_cast_fp16 = add(x = input_139_cast_fp16, y = linear_71_cast_fp16)[name = string("input_147_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_73_axes_0 = const()[name = string("hidden_states_73_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226505664)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226508288)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_73_cast_fp16 = layer_norm(axes = hidden_states_73_axes_0, beta = encoder_text_model_encoder_layers_12_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_12_layer_norm1_weight_to_fp16, x = input_147_cast_fp16)[name = string("hidden_states_73_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226510912))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227739776))))[name = string("encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227750080)))]; |
| tensor<fp16, [1, 77, 1280]> linear_72_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_12_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_12_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_73_cast_fp16)[name = string("linear_72_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227752704))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228981568))))[name = string("encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228991872)))]; |
| tensor<fp16, [1, 77, 1280]> linear_73_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_12_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_12_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_73_cast_fp16)[name = string("linear_73_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228994496))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230223360))))[name = string("encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230233664)))]; |
| tensor<fp16, [1, 77, 1280]> linear_74_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_12_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_12_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_73_cast_fp16)[name = string("linear_74_cast_fp16")]; |
| tensor<int32, [4]> var_797 = const()[name = string("op_797"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_798_cast_fp16 = reshape(shape = var_797, x = linear_72_cast_fp16)[name = string("op_798_cast_fp16")]; |
| tensor<int32, [4]> var_800 = const()[name = string("op_800"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_801_cast_fp16 = reshape(shape = var_800, x = linear_73_cast_fp16)[name = string("op_801_cast_fp16")]; |
| tensor<int32, [4]> var_803 = const()[name = string("op_803"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_804_cast_fp16 = reshape(shape = var_803, x = linear_74_cast_fp16)[name = string("op_804_cast_fp16")]; |
| tensor<int32, [4]> value_25_perm_0 = const()[name = string("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_25_cast_fp16 = mul(x = var_798_cast_fp16, y = var_11_to_fp16)[name = string("mul_25_cast_fp16")]; |
| bool matmul_12_transpose_y_0 = const()[name = string("matmul_12_transpose_y_0"), val = bool(true)]; |
| bool matmul_12_transpose_x_0 = const()[name = string("matmul_12_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_152_perm_0 = const()[name = string("transpose_152_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_153_perm_0 = const()[name = string("transpose_153_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_153 = transpose(perm = transpose_153_perm_0, x = var_801_cast_fp16)[name = string("transpose_270")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_152 = transpose(perm = transpose_152_perm_0, x = mul_25_cast_fp16)[name = string("transpose_271")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_12_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = transpose_152, y = transpose_153)[name = string("matmul_12_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_12_cast_fp16 = add(x = matmul_12_cast_fp16, y = mul_0_to_fp16)[name = string("add_12_cast_fp16")]; |
| int32 softmax_12_axis_0 = const()[name = string("softmax_12_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_12_cast_fp16 = softmax(axis = softmax_12_axis_0, x = add_12_cast_fp16)[name = string("softmax_12_cast_fp16")]; |
| bool attn_output_49_transpose_x_0 = const()[name = string("attn_output_49_transpose_x_0"), val = bool(false)]; |
| bool attn_output_49_transpose_y_0 = const()[name = string("attn_output_49_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = var_804_cast_fp16)[name = string("transpose_269")]; |
| tensor<fp16, [1, 20, 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 = softmax_12_cast_fp16, y = value_25_cast_fp16)[name = string("attn_output_49_cast_fp16")]; |
| tensor<int32, [4]> var_807_perm_0 = const()[name = string("op_807_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_809 = const()[name = string("op_809"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_807_cast_fp16 = transpose(perm = var_807_perm_0, x = attn_output_49_cast_fp16)[name = string("transpose_268")]; |
| tensor<fp16, [1, 77, 1280]> var_810_cast_fp16 = reshape(shape = var_809, x = var_807_cast_fp16)[name = string("op_810_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230236288))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231465152))))[name = string("encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231475456)))]; |
| tensor<fp16, [1, 77, 1280]> linear_75_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_12_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_12_self_attn_out_proj_weight_to_fp16_palettized, x = var_810_cast_fp16)[name = string("linear_75_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_151_cast_fp16 = add(x = input_147_cast_fp16, y = linear_75_cast_fp16)[name = string("input_151_cast_fp16")]; |
| tensor<int32, [1]> input_153_axes_0 = const()[name = string("input_153_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231478080)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231480704)))]; |
| tensor<fp16, [1, 77, 1280]> input_153_cast_fp16 = layer_norm(axes = input_153_axes_0, beta = encoder_text_model_encoder_layers_12_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_12_layer_norm2_weight_to_fp16, x = input_151_cast_fp16)[name = string("input_153_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231483328))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236398592))))[name = string("encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236439616)))]; |
| tensor<fp16, [1, 77, 5120]> linear_76_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_12_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_12_mlp_fc1_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = string("linear_76_cast_fp16")]; |
| string input_157_mode_0 = const()[name = string("input_157_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_157_cast_fp16 = gelu(mode = input_157_mode_0, x = linear_76_cast_fp16)[name = string("input_157_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236449920))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241365184))))[name = string("encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241375488)))]; |
| tensor<fp16, [1, 77, 1280]> linear_77_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_12_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_12_mlp_fc2_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = string("linear_77_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_159_cast_fp16 = add(x = input_151_cast_fp16, y = linear_77_cast_fp16)[name = string("input_159_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_79_axes_0 = const()[name = string("hidden_states_79_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241378112)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241380736)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_79_cast_fp16 = layer_norm(axes = hidden_states_79_axes_0, beta = encoder_text_model_encoder_layers_13_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_13_layer_norm1_weight_to_fp16, x = input_159_cast_fp16)[name = string("hidden_states_79_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241383360))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242612224))))[name = string("encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242622528)))]; |
| tensor<fp16, [1, 77, 1280]> linear_78_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_13_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_13_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_79_cast_fp16)[name = string("linear_78_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242625152))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243854016))))[name = string("encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243864320)))]; |
| tensor<fp16, [1, 77, 1280]> linear_79_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_13_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_13_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_79_cast_fp16)[name = string("linear_79_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243866944))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245095808))))[name = string("encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245106112)))]; |
| tensor<fp16, [1, 77, 1280]> linear_80_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_13_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_13_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_79_cast_fp16)[name = string("linear_80_cast_fp16")]; |
| tensor<int32, [4]> var_853 = const()[name = string("op_853"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_854_cast_fp16 = reshape(shape = var_853, x = linear_78_cast_fp16)[name = string("op_854_cast_fp16")]; |
| tensor<int32, [4]> var_856 = const()[name = string("op_856"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_857_cast_fp16 = reshape(shape = var_856, x = linear_79_cast_fp16)[name = string("op_857_cast_fp16")]; |
| tensor<int32, [4]> var_859 = const()[name = string("op_859"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_860_cast_fp16 = reshape(shape = var_859, x = linear_80_cast_fp16)[name = string("op_860_cast_fp16")]; |
| tensor<int32, [4]> value_27_perm_0 = const()[name = string("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_27_cast_fp16 = mul(x = var_854_cast_fp16, y = var_11_to_fp16)[name = string("mul_27_cast_fp16")]; |
| bool matmul_13_transpose_y_0 = const()[name = string("matmul_13_transpose_y_0"), val = bool(true)]; |
| bool matmul_13_transpose_x_0 = const()[name = string("matmul_13_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_154_perm_0 = const()[name = string("transpose_154_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_155_perm_0 = const()[name = string("transpose_155_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_155 = transpose(perm = transpose_155_perm_0, x = var_857_cast_fp16)[name = string("transpose_266")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_154 = transpose(perm = transpose_154_perm_0, x = mul_27_cast_fp16)[name = string("transpose_267")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_13_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = transpose_154, y = transpose_155)[name = string("matmul_13_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_13_cast_fp16 = add(x = matmul_13_cast_fp16, y = mul_0_to_fp16)[name = string("add_13_cast_fp16")]; |
| int32 softmax_13_axis_0 = const()[name = string("softmax_13_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_13_cast_fp16 = softmax(axis = softmax_13_axis_0, x = add_13_cast_fp16)[name = string("softmax_13_cast_fp16")]; |
| bool attn_output_53_transpose_x_0 = const()[name = string("attn_output_53_transpose_x_0"), val = bool(false)]; |
| bool attn_output_53_transpose_y_0 = const()[name = string("attn_output_53_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = var_860_cast_fp16)[name = string("transpose_265")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_53_cast_fp16 = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = softmax_13_cast_fp16, y = value_27_cast_fp16)[name = string("attn_output_53_cast_fp16")]; |
| tensor<int32, [4]> var_863_perm_0 = const()[name = string("op_863_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_865 = const()[name = string("op_865"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_863_cast_fp16 = transpose(perm = var_863_perm_0, x = attn_output_53_cast_fp16)[name = string("transpose_264")]; |
| tensor<fp16, [1, 77, 1280]> var_866_cast_fp16 = reshape(shape = var_865, x = var_863_cast_fp16)[name = string("op_866_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245108736))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246337600))))[name = string("encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246347904)))]; |
| tensor<fp16, [1, 77, 1280]> linear_81_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_13_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_13_self_attn_out_proj_weight_to_fp16_palettized, x = var_866_cast_fp16)[name = string("linear_81_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_163_cast_fp16 = add(x = input_159_cast_fp16, y = linear_81_cast_fp16)[name = string("input_163_cast_fp16")]; |
| tensor<int32, [1]> input_165_axes_0 = const()[name = string("input_165_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246350528)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246353152)))]; |
| tensor<fp16, [1, 77, 1280]> input_165_cast_fp16 = layer_norm(axes = input_165_axes_0, beta = encoder_text_model_encoder_layers_13_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_13_layer_norm2_weight_to_fp16, x = input_163_cast_fp16)[name = string("input_165_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246355776))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251271040))))[name = string("encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251312064)))]; |
| tensor<fp16, [1, 77, 5120]> linear_82_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_13_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_13_mlp_fc1_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = string("linear_82_cast_fp16")]; |
| string input_169_mode_0 = const()[name = string("input_169_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_169_cast_fp16 = gelu(mode = input_169_mode_0, x = linear_82_cast_fp16)[name = string("input_169_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251322368))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256237632))))[name = string("encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256247936)))]; |
| tensor<fp16, [1, 77, 1280]> linear_83_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_13_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_13_mlp_fc2_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = string("linear_83_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_171_cast_fp16 = add(x = input_163_cast_fp16, y = linear_83_cast_fp16)[name = string("input_171_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_85_axes_0 = const()[name = string("hidden_states_85_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256250560)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256253184)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = encoder_text_model_encoder_layers_14_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_14_layer_norm1_weight_to_fp16, x = input_171_cast_fp16)[name = string("hidden_states_85_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256255808))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257484672))))[name = string("encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257494976)))]; |
| tensor<fp16, [1, 77, 1280]> linear_84_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_14_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_14_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_85_cast_fp16)[name = string("linear_84_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257497600))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258726464))))[name = string("encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258736768)))]; |
| tensor<fp16, [1, 77, 1280]> linear_85_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_14_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_14_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_85_cast_fp16)[name = string("linear_85_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258739392))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259968256))))[name = string("encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259978560)))]; |
| tensor<fp16, [1, 77, 1280]> linear_86_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_14_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_14_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_85_cast_fp16)[name = string("linear_86_cast_fp16")]; |
| tensor<int32, [4]> var_909 = const()[name = string("op_909"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_910_cast_fp16 = reshape(shape = var_909, x = linear_84_cast_fp16)[name = string("op_910_cast_fp16")]; |
| tensor<int32, [4]> var_912 = const()[name = string("op_912"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_913_cast_fp16 = reshape(shape = var_912, x = linear_85_cast_fp16)[name = string("op_913_cast_fp16")]; |
| tensor<int32, [4]> var_915 = const()[name = string("op_915"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_916_cast_fp16 = reshape(shape = var_915, x = linear_86_cast_fp16)[name = string("op_916_cast_fp16")]; |
| tensor<int32, [4]> value_29_perm_0 = const()[name = string("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_29_cast_fp16 = mul(x = var_910_cast_fp16, y = var_11_to_fp16)[name = string("mul_29_cast_fp16")]; |
| bool matmul_14_transpose_y_0 = const()[name = string("matmul_14_transpose_y_0"), val = bool(true)]; |
| bool matmul_14_transpose_x_0 = const()[name = string("matmul_14_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_156_perm_0 = const()[name = string("transpose_156_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_157_perm_0 = const()[name = string("transpose_157_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_157 = transpose(perm = transpose_157_perm_0, x = var_913_cast_fp16)[name = string("transpose_262")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_156 = transpose(perm = transpose_156_perm_0, x = mul_29_cast_fp16)[name = string("transpose_263")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_14_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = transpose_156, y = transpose_157)[name = string("matmul_14_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_14_cast_fp16 = add(x = matmul_14_cast_fp16, y = mul_0_to_fp16)[name = string("add_14_cast_fp16")]; |
| int32 softmax_14_axis_0 = const()[name = string("softmax_14_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_14_cast_fp16 = softmax(axis = softmax_14_axis_0, x = add_14_cast_fp16)[name = string("softmax_14_cast_fp16")]; |
| bool attn_output_57_transpose_x_0 = const()[name = string("attn_output_57_transpose_x_0"), val = bool(false)]; |
| bool attn_output_57_transpose_y_0 = const()[name = string("attn_output_57_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = var_916_cast_fp16)[name = string("transpose_261")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_57_cast_fp16 = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = softmax_14_cast_fp16, y = value_29_cast_fp16)[name = string("attn_output_57_cast_fp16")]; |
| tensor<int32, [4]> var_919_perm_0 = const()[name = string("op_919_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_921 = const()[name = string("op_921"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_919_cast_fp16 = transpose(perm = var_919_perm_0, x = attn_output_57_cast_fp16)[name = string("transpose_260")]; |
| tensor<fp16, [1, 77, 1280]> var_922_cast_fp16 = reshape(shape = var_921, x = var_919_cast_fp16)[name = string("op_922_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259981184))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261210048))))[name = string("encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261220352)))]; |
| tensor<fp16, [1, 77, 1280]> linear_87_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_14_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_14_self_attn_out_proj_weight_to_fp16_palettized, x = var_922_cast_fp16)[name = string("linear_87_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_175_cast_fp16 = add(x = input_171_cast_fp16, y = linear_87_cast_fp16)[name = string("input_175_cast_fp16")]; |
| tensor<int32, [1]> input_177_axes_0 = const()[name = string("input_177_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261222976)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261225600)))]; |
| tensor<fp16, [1, 77, 1280]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = encoder_text_model_encoder_layers_14_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_14_layer_norm2_weight_to_fp16, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261228224))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266143488))))[name = string("encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266184512)))]; |
| tensor<fp16, [1, 77, 5120]> linear_88_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_14_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_14_mlp_fc1_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = string("linear_88_cast_fp16")]; |
| string input_181_mode_0 = const()[name = string("input_181_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_181_cast_fp16 = gelu(mode = input_181_mode_0, x = linear_88_cast_fp16)[name = string("input_181_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266194816))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271110080))))[name = string("encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271120384)))]; |
| tensor<fp16, [1, 77, 1280]> linear_89_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_14_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_14_mlp_fc2_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = string("linear_89_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_183_cast_fp16 = add(x = input_175_cast_fp16, y = linear_89_cast_fp16)[name = string("input_183_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_91_axes_0 = const()[name = string("hidden_states_91_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271123008)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271125632)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_91_cast_fp16 = layer_norm(axes = hidden_states_91_axes_0, beta = encoder_text_model_encoder_layers_15_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_15_layer_norm1_weight_to_fp16, x = input_183_cast_fp16)[name = string("hidden_states_91_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271128256))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272357120))))[name = string("encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272367424)))]; |
| tensor<fp16, [1, 77, 1280]> linear_90_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_15_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_15_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_91_cast_fp16)[name = string("linear_90_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272370048))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273598912))))[name = string("encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273609216)))]; |
| tensor<fp16, [1, 77, 1280]> linear_91_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_15_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_15_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_91_cast_fp16)[name = string("linear_91_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273611840))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274840704))))[name = string("encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274851008)))]; |
| tensor<fp16, [1, 77, 1280]> linear_92_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_15_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_15_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_91_cast_fp16)[name = string("linear_92_cast_fp16")]; |
| tensor<int32, [4]> var_965 = const()[name = string("op_965"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_966_cast_fp16 = reshape(shape = var_965, x = linear_90_cast_fp16)[name = string("op_966_cast_fp16")]; |
| tensor<int32, [4]> var_968 = const()[name = string("op_968"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_969_cast_fp16 = reshape(shape = var_968, x = linear_91_cast_fp16)[name = string("op_969_cast_fp16")]; |
| tensor<int32, [4]> var_971 = const()[name = string("op_971"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_972_cast_fp16 = reshape(shape = var_971, x = linear_92_cast_fp16)[name = string("op_972_cast_fp16")]; |
| tensor<int32, [4]> value_31_perm_0 = const()[name = string("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_31_cast_fp16 = mul(x = var_966_cast_fp16, y = var_11_to_fp16)[name = string("mul_31_cast_fp16")]; |
| bool matmul_15_transpose_y_0 = const()[name = string("matmul_15_transpose_y_0"), val = bool(true)]; |
| bool matmul_15_transpose_x_0 = const()[name = string("matmul_15_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_158_perm_0 = const()[name = string("transpose_158_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_159_perm_0 = const()[name = string("transpose_159_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_159 = transpose(perm = transpose_159_perm_0, x = var_969_cast_fp16)[name = string("transpose_258")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_158 = transpose(perm = transpose_158_perm_0, x = mul_31_cast_fp16)[name = string("transpose_259")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_15_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = transpose_158, y = transpose_159)[name = string("matmul_15_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_15_cast_fp16 = add(x = matmul_15_cast_fp16, y = mul_0_to_fp16)[name = string("add_15_cast_fp16")]; |
| int32 softmax_15_axis_0 = const()[name = string("softmax_15_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_15_cast_fp16 = softmax(axis = softmax_15_axis_0, x = add_15_cast_fp16)[name = string("softmax_15_cast_fp16")]; |
| bool attn_output_61_transpose_x_0 = const()[name = string("attn_output_61_transpose_x_0"), val = bool(false)]; |
| bool attn_output_61_transpose_y_0 = const()[name = string("attn_output_61_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = var_972_cast_fp16)[name = string("transpose_257")]; |
| tensor<fp16, [1, 20, 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 = softmax_15_cast_fp16, y = value_31_cast_fp16)[name = string("attn_output_61_cast_fp16")]; |
| tensor<int32, [4]> var_975_perm_0 = const()[name = string("op_975_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_977 = const()[name = string("op_977"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_975_cast_fp16 = transpose(perm = var_975_perm_0, x = attn_output_61_cast_fp16)[name = string("transpose_256")]; |
| tensor<fp16, [1, 77, 1280]> var_978_cast_fp16 = reshape(shape = var_977, x = var_975_cast_fp16)[name = string("op_978_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274853632))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276082496))))[name = string("encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276092800)))]; |
| tensor<fp16, [1, 77, 1280]> linear_93_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_15_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_15_self_attn_out_proj_weight_to_fp16_palettized, x = var_978_cast_fp16)[name = string("linear_93_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_187_cast_fp16 = add(x = input_183_cast_fp16, y = linear_93_cast_fp16)[name = string("input_187_cast_fp16")]; |
| tensor<int32, [1]> input_189_axes_0 = const()[name = string("input_189_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276095424)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276098048)))]; |
| tensor<fp16, [1, 77, 1280]> input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = encoder_text_model_encoder_layers_15_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_15_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = string("input_189_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276100672))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281015936))))[name = string("encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281056960)))]; |
| tensor<fp16, [1, 77, 5120]> linear_94_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_15_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_15_mlp_fc1_weight_to_fp16_palettized, x = input_189_cast_fp16)[name = string("linear_94_cast_fp16")]; |
| string input_193_mode_0 = const()[name = string("input_193_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_193_cast_fp16 = gelu(mode = input_193_mode_0, x = linear_94_cast_fp16)[name = string("input_193_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281067264))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285982528))))[name = string("encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285992832)))]; |
| tensor<fp16, [1, 77, 1280]> linear_95_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_15_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_15_mlp_fc2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = string("linear_95_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_95_cast_fp16)[name = string("input_195_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_97_axes_0 = const()[name = string("hidden_states_97_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285995456)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285998080)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_97_cast_fp16 = layer_norm(axes = hidden_states_97_axes_0, beta = encoder_text_model_encoder_layers_16_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_16_layer_norm1_weight_to_fp16, x = input_195_cast_fp16)[name = string("hidden_states_97_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286000704))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287229568))))[name = string("encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287239872)))]; |
| tensor<fp16, [1, 77, 1280]> linear_96_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_16_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_16_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_97_cast_fp16)[name = string("linear_96_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287242496))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288471360))))[name = string("encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288481664)))]; |
| tensor<fp16, [1, 77, 1280]> linear_97_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_16_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_16_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_97_cast_fp16)[name = string("linear_97_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288484288))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289713152))))[name = string("encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289723456)))]; |
| tensor<fp16, [1, 77, 1280]> linear_98_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_16_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_16_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_97_cast_fp16)[name = string("linear_98_cast_fp16")]; |
| tensor<int32, [4]> var_1021 = const()[name = string("op_1021"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1022_cast_fp16 = reshape(shape = var_1021, x = linear_96_cast_fp16)[name = string("op_1022_cast_fp16")]; |
| tensor<int32, [4]> var_1024 = const()[name = string("op_1024"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1025_cast_fp16 = reshape(shape = var_1024, x = linear_97_cast_fp16)[name = string("op_1025_cast_fp16")]; |
| tensor<int32, [4]> var_1027 = const()[name = string("op_1027"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1028_cast_fp16 = reshape(shape = var_1027, x = linear_98_cast_fp16)[name = string("op_1028_cast_fp16")]; |
| tensor<int32, [4]> value_33_perm_0 = const()[name = string("value_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_33_cast_fp16 = mul(x = var_1022_cast_fp16, y = var_11_to_fp16)[name = string("mul_33_cast_fp16")]; |
| bool matmul_16_transpose_y_0 = const()[name = string("matmul_16_transpose_y_0"), val = bool(true)]; |
| bool matmul_16_transpose_x_0 = const()[name = string("matmul_16_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_160_perm_0 = const()[name = string("transpose_160_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_161_perm_0 = const()[name = string("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_161 = transpose(perm = transpose_161_perm_0, x = var_1025_cast_fp16)[name = string("transpose_254")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_160 = transpose(perm = transpose_160_perm_0, x = mul_33_cast_fp16)[name = string("transpose_255")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_16_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = transpose_160, y = transpose_161)[name = string("matmul_16_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_16_cast_fp16 = add(x = matmul_16_cast_fp16, y = mul_0_to_fp16)[name = string("add_16_cast_fp16")]; |
| int32 softmax_16_axis_0 = const()[name = string("softmax_16_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_16_cast_fp16 = softmax(axis = softmax_16_axis_0, x = add_16_cast_fp16)[name = string("softmax_16_cast_fp16")]; |
| bool attn_output_65_transpose_x_0 = const()[name = string("attn_output_65_transpose_x_0"), val = bool(false)]; |
| bool attn_output_65_transpose_y_0 = const()[name = string("attn_output_65_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_33_cast_fp16 = transpose(perm = value_33_perm_0, x = var_1028_cast_fp16)[name = string("transpose_253")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_65_cast_fp16 = matmul(transpose_x = attn_output_65_transpose_x_0, transpose_y = attn_output_65_transpose_y_0, x = softmax_16_cast_fp16, y = value_33_cast_fp16)[name = string("attn_output_65_cast_fp16")]; |
| tensor<int32, [4]> var_1031_perm_0 = const()[name = string("op_1031_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1033 = const()[name = string("op_1033"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1031_cast_fp16 = transpose(perm = var_1031_perm_0, x = attn_output_65_cast_fp16)[name = string("transpose_252")]; |
| tensor<fp16, [1, 77, 1280]> var_1034_cast_fp16 = reshape(shape = var_1033, x = var_1031_cast_fp16)[name = string("op_1034_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289726080))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290954944))))[name = string("encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290965248)))]; |
| tensor<fp16, [1, 77, 1280]> linear_99_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_16_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_16_self_attn_out_proj_weight_to_fp16_palettized, x = var_1034_cast_fp16)[name = string("linear_99_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_199_cast_fp16 = add(x = input_195_cast_fp16, y = linear_99_cast_fp16)[name = string("input_199_cast_fp16")]; |
| tensor<int32, [1]> input_201_axes_0 = const()[name = string("input_201_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290967872)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290970496)))]; |
| tensor<fp16, [1, 77, 1280]> input_201_cast_fp16 = layer_norm(axes = input_201_axes_0, beta = encoder_text_model_encoder_layers_16_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_16_layer_norm2_weight_to_fp16, x = input_199_cast_fp16)[name = string("input_201_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290973120))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295888384))))[name = string("encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295929408)))]; |
| tensor<fp16, [1, 77, 5120]> linear_100_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_16_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_16_mlp_fc1_weight_to_fp16_palettized, x = input_201_cast_fp16)[name = string("linear_100_cast_fp16")]; |
| string input_205_mode_0 = const()[name = string("input_205_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_205_cast_fp16 = gelu(mode = input_205_mode_0, x = linear_100_cast_fp16)[name = string("input_205_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295939712))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300854976))))[name = string("encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300865280)))]; |
| tensor<fp16, [1, 77, 1280]> linear_101_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_16_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_16_mlp_fc2_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = string("linear_101_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_207_cast_fp16 = add(x = input_199_cast_fp16, y = linear_101_cast_fp16)[name = string("input_207_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_103_axes_0 = const()[name = string("hidden_states_103_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300867904)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300870528)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_103_cast_fp16 = layer_norm(axes = hidden_states_103_axes_0, beta = encoder_text_model_encoder_layers_17_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_17_layer_norm1_weight_to_fp16, x = input_207_cast_fp16)[name = string("hidden_states_103_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300873152))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302102016))))[name = string("encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302112320)))]; |
| tensor<fp16, [1, 77, 1280]> linear_102_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_17_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_17_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_103_cast_fp16)[name = string("linear_102_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302114944))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303343808))))[name = string("encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303354112)))]; |
| tensor<fp16, [1, 77, 1280]> linear_103_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_17_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_17_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_103_cast_fp16)[name = string("linear_103_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303356736))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304585600))))[name = string("encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304595904)))]; |
| tensor<fp16, [1, 77, 1280]> linear_104_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_17_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_17_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_103_cast_fp16)[name = string("linear_104_cast_fp16")]; |
| tensor<int32, [4]> var_1077 = const()[name = string("op_1077"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1078_cast_fp16 = reshape(shape = var_1077, x = linear_102_cast_fp16)[name = string("op_1078_cast_fp16")]; |
| tensor<int32, [4]> var_1080 = const()[name = string("op_1080"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1081_cast_fp16 = reshape(shape = var_1080, x = linear_103_cast_fp16)[name = string("op_1081_cast_fp16")]; |
| tensor<int32, [4]> var_1083 = const()[name = string("op_1083"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1084_cast_fp16 = reshape(shape = var_1083, x = linear_104_cast_fp16)[name = string("op_1084_cast_fp16")]; |
| tensor<int32, [4]> value_35_perm_0 = const()[name = string("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_35_cast_fp16 = mul(x = var_1078_cast_fp16, y = var_11_to_fp16)[name = string("mul_35_cast_fp16")]; |
| bool matmul_17_transpose_y_0 = const()[name = string("matmul_17_transpose_y_0"), val = bool(true)]; |
| bool matmul_17_transpose_x_0 = const()[name = string("matmul_17_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_162_perm_0 = const()[name = string("transpose_162_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_163_perm_0 = const()[name = string("transpose_163_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_163 = transpose(perm = transpose_163_perm_0, x = var_1081_cast_fp16)[name = string("transpose_250")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_162 = transpose(perm = transpose_162_perm_0, x = mul_35_cast_fp16)[name = string("transpose_251")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_17_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = transpose_162, y = transpose_163)[name = string("matmul_17_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_17_cast_fp16 = add(x = matmul_17_cast_fp16, y = mul_0_to_fp16)[name = string("add_17_cast_fp16")]; |
| int32 softmax_17_axis_0 = const()[name = string("softmax_17_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_17_cast_fp16 = softmax(axis = softmax_17_axis_0, x = add_17_cast_fp16)[name = string("softmax_17_cast_fp16")]; |
| bool attn_output_69_transpose_x_0 = const()[name = string("attn_output_69_transpose_x_0"), val = bool(false)]; |
| bool attn_output_69_transpose_y_0 = const()[name = string("attn_output_69_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = var_1084_cast_fp16)[name = string("transpose_249")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_69_cast_fp16 = matmul(transpose_x = attn_output_69_transpose_x_0, transpose_y = attn_output_69_transpose_y_0, x = softmax_17_cast_fp16, y = value_35_cast_fp16)[name = string("attn_output_69_cast_fp16")]; |
| tensor<int32, [4]> var_1087_perm_0 = const()[name = string("op_1087_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1089 = const()[name = string("op_1089"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1087_cast_fp16 = transpose(perm = var_1087_perm_0, x = attn_output_69_cast_fp16)[name = string("transpose_248")]; |
| tensor<fp16, [1, 77, 1280]> var_1090_cast_fp16 = reshape(shape = var_1089, x = var_1087_cast_fp16)[name = string("op_1090_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304598528))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305827392))))[name = string("encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305837696)))]; |
| tensor<fp16, [1, 77, 1280]> linear_105_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_17_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_17_self_attn_out_proj_weight_to_fp16_palettized, x = var_1090_cast_fp16)[name = string("linear_105_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_211_cast_fp16 = add(x = input_207_cast_fp16, y = linear_105_cast_fp16)[name = string("input_211_cast_fp16")]; |
| tensor<int32, [1]> input_213_axes_0 = const()[name = string("input_213_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305840320)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305842944)))]; |
| tensor<fp16, [1, 77, 1280]> input_213_cast_fp16 = layer_norm(axes = input_213_axes_0, beta = encoder_text_model_encoder_layers_17_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_17_layer_norm2_weight_to_fp16, x = input_211_cast_fp16)[name = string("input_213_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305845568))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310760832))))[name = string("encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310801856)))]; |
| tensor<fp16, [1, 77, 5120]> linear_106_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_17_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_17_mlp_fc1_weight_to_fp16_palettized, x = input_213_cast_fp16)[name = string("linear_106_cast_fp16")]; |
| string input_217_mode_0 = const()[name = string("input_217_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_217_cast_fp16 = gelu(mode = input_217_mode_0, x = linear_106_cast_fp16)[name = string("input_217_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310812160))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315727424))))[name = string("encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315737728)))]; |
| tensor<fp16, [1, 77, 1280]> linear_107_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_17_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_17_mlp_fc2_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = string("linear_107_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_219_cast_fp16 = add(x = input_211_cast_fp16, y = linear_107_cast_fp16)[name = string("input_219_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_109_axes_0 = const()[name = string("hidden_states_109_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315740352)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315742976)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_109_cast_fp16 = layer_norm(axes = hidden_states_109_axes_0, beta = encoder_text_model_encoder_layers_18_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_18_layer_norm1_weight_to_fp16, x = input_219_cast_fp16)[name = string("hidden_states_109_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315745600))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316974464))))[name = string("encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316984768)))]; |
| tensor<fp16, [1, 77, 1280]> linear_108_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_18_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_18_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_109_cast_fp16)[name = string("linear_108_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316987392))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318216256))))[name = string("encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318226560)))]; |
| tensor<fp16, [1, 77, 1280]> linear_109_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_18_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_18_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_109_cast_fp16)[name = string("linear_109_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318229184))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319458048))))[name = string("encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319468352)))]; |
| tensor<fp16, [1, 77, 1280]> linear_110_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_18_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_18_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_109_cast_fp16)[name = string("linear_110_cast_fp16")]; |
| tensor<int32, [4]> var_1133 = const()[name = string("op_1133"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1134_cast_fp16 = reshape(shape = var_1133, x = linear_108_cast_fp16)[name = string("op_1134_cast_fp16")]; |
| tensor<int32, [4]> var_1136 = const()[name = string("op_1136"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1137_cast_fp16 = reshape(shape = var_1136, x = linear_109_cast_fp16)[name = string("op_1137_cast_fp16")]; |
| tensor<int32, [4]> var_1139 = const()[name = string("op_1139"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1140_cast_fp16 = reshape(shape = var_1139, x = linear_110_cast_fp16)[name = string("op_1140_cast_fp16")]; |
| tensor<int32, [4]> value_37_perm_0 = const()[name = string("value_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_37_cast_fp16 = mul(x = var_1134_cast_fp16, y = var_11_to_fp16)[name = string("mul_37_cast_fp16")]; |
| bool matmul_18_transpose_y_0 = const()[name = string("matmul_18_transpose_y_0"), val = bool(true)]; |
| bool matmul_18_transpose_x_0 = const()[name = string("matmul_18_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_164_perm_0 = const()[name = string("transpose_164_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_165_perm_0 = const()[name = string("transpose_165_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_165 = transpose(perm = transpose_165_perm_0, x = var_1137_cast_fp16)[name = string("transpose_246")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_164 = transpose(perm = transpose_164_perm_0, x = mul_37_cast_fp16)[name = string("transpose_247")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_18_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = transpose_164, y = transpose_165)[name = string("matmul_18_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_18_cast_fp16 = add(x = matmul_18_cast_fp16, y = mul_0_to_fp16)[name = string("add_18_cast_fp16")]; |
| int32 softmax_18_axis_0 = const()[name = string("softmax_18_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_18_cast_fp16 = softmax(axis = softmax_18_axis_0, x = add_18_cast_fp16)[name = string("softmax_18_cast_fp16")]; |
| bool attn_output_73_transpose_x_0 = const()[name = string("attn_output_73_transpose_x_0"), val = bool(false)]; |
| bool attn_output_73_transpose_y_0 = const()[name = string("attn_output_73_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_37_cast_fp16 = transpose(perm = value_37_perm_0, x = var_1140_cast_fp16)[name = string("transpose_245")]; |
| tensor<fp16, [1, 20, 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 = softmax_18_cast_fp16, y = value_37_cast_fp16)[name = string("attn_output_73_cast_fp16")]; |
| tensor<int32, [4]> var_1143_perm_0 = const()[name = string("op_1143_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1145 = const()[name = string("op_1145"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1143_cast_fp16 = transpose(perm = var_1143_perm_0, x = attn_output_73_cast_fp16)[name = string("transpose_244")]; |
| tensor<fp16, [1, 77, 1280]> var_1146_cast_fp16 = reshape(shape = var_1145, x = var_1143_cast_fp16)[name = string("op_1146_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319470976))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320699840))))[name = string("encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320710144)))]; |
| tensor<fp16, [1, 77, 1280]> linear_111_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_18_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_18_self_attn_out_proj_weight_to_fp16_palettized, x = var_1146_cast_fp16)[name = string("linear_111_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_223_cast_fp16 = add(x = input_219_cast_fp16, y = linear_111_cast_fp16)[name = string("input_223_cast_fp16")]; |
| tensor<int32, [1]> input_225_axes_0 = const()[name = string("input_225_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320712768)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320715392)))]; |
| tensor<fp16, [1, 77, 1280]> input_225_cast_fp16 = layer_norm(axes = input_225_axes_0, beta = encoder_text_model_encoder_layers_18_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_18_layer_norm2_weight_to_fp16, x = input_223_cast_fp16)[name = string("input_225_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320718016))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325633280))))[name = string("encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325674304)))]; |
| tensor<fp16, [1, 77, 5120]> linear_112_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_18_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_18_mlp_fc1_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = string("linear_112_cast_fp16")]; |
| string input_229_mode_0 = const()[name = string("input_229_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_229_cast_fp16 = gelu(mode = input_229_mode_0, x = linear_112_cast_fp16)[name = string("input_229_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325684608))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330599872))))[name = string("encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330610176)))]; |
| tensor<fp16, [1, 77, 1280]> linear_113_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_18_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_18_mlp_fc2_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = string("linear_113_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_231_cast_fp16 = add(x = input_223_cast_fp16, y = linear_113_cast_fp16)[name = string("input_231_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_115_axes_0 = const()[name = string("hidden_states_115_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330612800)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330615424)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_115_cast_fp16 = layer_norm(axes = hidden_states_115_axes_0, beta = encoder_text_model_encoder_layers_19_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_19_layer_norm1_weight_to_fp16, x = input_231_cast_fp16)[name = string("hidden_states_115_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330618048))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(331846912))))[name = string("encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(331857216)))]; |
| tensor<fp16, [1, 77, 1280]> linear_114_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_19_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_19_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_115_cast_fp16)[name = string("linear_114_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(331859840))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333088704))))[name = string("encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333099008)))]; |
| tensor<fp16, [1, 77, 1280]> linear_115_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_19_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_19_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_115_cast_fp16)[name = string("linear_115_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333101632))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334330496))))[name = string("encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334340800)))]; |
| tensor<fp16, [1, 77, 1280]> linear_116_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_19_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_19_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_115_cast_fp16)[name = string("linear_116_cast_fp16")]; |
| tensor<int32, [4]> var_1189 = const()[name = string("op_1189"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1190_cast_fp16 = reshape(shape = var_1189, x = linear_114_cast_fp16)[name = string("op_1190_cast_fp16")]; |
| tensor<int32, [4]> var_1192 = const()[name = string("op_1192"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1193_cast_fp16 = reshape(shape = var_1192, x = linear_115_cast_fp16)[name = string("op_1193_cast_fp16")]; |
| tensor<int32, [4]> var_1195 = const()[name = string("op_1195"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1196_cast_fp16 = reshape(shape = var_1195, x = linear_116_cast_fp16)[name = string("op_1196_cast_fp16")]; |
| tensor<int32, [4]> value_39_perm_0 = const()[name = string("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_39_cast_fp16 = mul(x = var_1190_cast_fp16, y = var_11_to_fp16)[name = string("mul_39_cast_fp16")]; |
| bool matmul_19_transpose_y_0 = const()[name = string("matmul_19_transpose_y_0"), val = bool(true)]; |
| bool matmul_19_transpose_x_0 = const()[name = string("matmul_19_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_166_perm_0 = const()[name = string("transpose_166_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_167_perm_0 = const()[name = string("transpose_167_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_167 = transpose(perm = transpose_167_perm_0, x = var_1193_cast_fp16)[name = string("transpose_242")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_166 = transpose(perm = transpose_166_perm_0, x = mul_39_cast_fp16)[name = string("transpose_243")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_19_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = transpose_166, y = transpose_167)[name = string("matmul_19_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_19_cast_fp16 = add(x = matmul_19_cast_fp16, y = mul_0_to_fp16)[name = string("add_19_cast_fp16")]; |
| int32 softmax_19_axis_0 = const()[name = string("softmax_19_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_19_cast_fp16 = softmax(axis = softmax_19_axis_0, x = add_19_cast_fp16)[name = string("softmax_19_cast_fp16")]; |
| bool attn_output_77_transpose_x_0 = const()[name = string("attn_output_77_transpose_x_0"), val = bool(false)]; |
| bool attn_output_77_transpose_y_0 = const()[name = string("attn_output_77_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = var_1196_cast_fp16)[name = string("transpose_241")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_77_cast_fp16 = matmul(transpose_x = attn_output_77_transpose_x_0, transpose_y = attn_output_77_transpose_y_0, x = softmax_19_cast_fp16, y = value_39_cast_fp16)[name = string("attn_output_77_cast_fp16")]; |
| tensor<int32, [4]> var_1199_perm_0 = const()[name = string("op_1199_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1201 = const()[name = string("op_1201"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1199_cast_fp16 = transpose(perm = var_1199_perm_0, x = attn_output_77_cast_fp16)[name = string("transpose_240")]; |
| tensor<fp16, [1, 77, 1280]> var_1202_cast_fp16 = reshape(shape = var_1201, x = var_1199_cast_fp16)[name = string("op_1202_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334343424))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335572288))))[name = string("encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335582592)))]; |
| tensor<fp16, [1, 77, 1280]> linear_117_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_19_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_19_self_attn_out_proj_weight_to_fp16_palettized, x = var_1202_cast_fp16)[name = string("linear_117_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_235_cast_fp16 = add(x = input_231_cast_fp16, y = linear_117_cast_fp16)[name = string("input_235_cast_fp16")]; |
| tensor<int32, [1]> input_237_axes_0 = const()[name = string("input_237_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335585216)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335587840)))]; |
| tensor<fp16, [1, 77, 1280]> input_237_cast_fp16 = layer_norm(axes = input_237_axes_0, beta = encoder_text_model_encoder_layers_19_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_19_layer_norm2_weight_to_fp16, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335590464))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340505728))))[name = string("encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340546752)))]; |
| tensor<fp16, [1, 77, 5120]> linear_118_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_19_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_19_mlp_fc1_weight_to_fp16_palettized, x = input_237_cast_fp16)[name = string("linear_118_cast_fp16")]; |
| string input_241_mode_0 = const()[name = string("input_241_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_241_cast_fp16 = gelu(mode = input_241_mode_0, x = linear_118_cast_fp16)[name = string("input_241_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340557056))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345472320))))[name = string("encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345482624)))]; |
| tensor<fp16, [1, 77, 1280]> linear_119_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_19_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_19_mlp_fc2_weight_to_fp16_palettized, x = input_241_cast_fp16)[name = string("linear_119_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_243_cast_fp16 = add(x = input_235_cast_fp16, y = linear_119_cast_fp16)[name = string("input_243_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_121_axes_0 = const()[name = string("hidden_states_121_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345485248)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345487872)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = encoder_text_model_encoder_layers_20_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_20_layer_norm1_weight_to_fp16, x = input_243_cast_fp16)[name = string("hidden_states_121_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345490496))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(346719360))))[name = string("encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(346729664)))]; |
| tensor<fp16, [1, 77, 1280]> linear_120_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_20_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_20_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_121_cast_fp16)[name = string("linear_120_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(346732288))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(347961152))))[name = string("encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(347971456)))]; |
| tensor<fp16, [1, 77, 1280]> linear_121_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_20_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_20_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_121_cast_fp16)[name = string("linear_121_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(347974080))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349202944))))[name = string("encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349213248)))]; |
| tensor<fp16, [1, 77, 1280]> linear_122_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_20_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_20_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_121_cast_fp16)[name = string("linear_122_cast_fp16")]; |
| tensor<int32, [4]> var_1245 = const()[name = string("op_1245"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1246_cast_fp16 = reshape(shape = var_1245, x = linear_120_cast_fp16)[name = string("op_1246_cast_fp16")]; |
| tensor<int32, [4]> var_1248 = const()[name = string("op_1248"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1249_cast_fp16 = reshape(shape = var_1248, x = linear_121_cast_fp16)[name = string("op_1249_cast_fp16")]; |
| tensor<int32, [4]> var_1251 = const()[name = string("op_1251"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1252_cast_fp16 = reshape(shape = var_1251, x = linear_122_cast_fp16)[name = string("op_1252_cast_fp16")]; |
| tensor<int32, [4]> value_41_perm_0 = const()[name = string("value_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_41_cast_fp16 = mul(x = var_1246_cast_fp16, y = var_11_to_fp16)[name = string("mul_41_cast_fp16")]; |
| bool matmul_20_transpose_y_0 = const()[name = string("matmul_20_transpose_y_0"), val = bool(true)]; |
| bool matmul_20_transpose_x_0 = const()[name = string("matmul_20_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_168_perm_0 = const()[name = string("transpose_168_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_169_perm_0 = const()[name = string("transpose_169_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_169 = transpose(perm = transpose_169_perm_0, x = var_1249_cast_fp16)[name = string("transpose_238")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_168 = transpose(perm = transpose_168_perm_0, x = mul_41_cast_fp16)[name = string("transpose_239")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_20_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = transpose_168, y = transpose_169)[name = string("matmul_20_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_20_cast_fp16 = add(x = matmul_20_cast_fp16, y = mul_0_to_fp16)[name = string("add_20_cast_fp16")]; |
| int32 softmax_20_axis_0 = const()[name = string("softmax_20_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_20_cast_fp16 = softmax(axis = softmax_20_axis_0, x = add_20_cast_fp16)[name = string("softmax_20_cast_fp16")]; |
| bool attn_output_81_transpose_x_0 = const()[name = string("attn_output_81_transpose_x_0"), val = bool(false)]; |
| bool attn_output_81_transpose_y_0 = const()[name = string("attn_output_81_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_41_cast_fp16 = transpose(perm = value_41_perm_0, x = var_1252_cast_fp16)[name = string("transpose_237")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_81_cast_fp16 = matmul(transpose_x = attn_output_81_transpose_x_0, transpose_y = attn_output_81_transpose_y_0, x = softmax_20_cast_fp16, y = value_41_cast_fp16)[name = string("attn_output_81_cast_fp16")]; |
| tensor<int32, [4]> var_1255_perm_0 = const()[name = string("op_1255_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1257 = const()[name = string("op_1257"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1255_cast_fp16 = transpose(perm = var_1255_perm_0, x = attn_output_81_cast_fp16)[name = string("transpose_236")]; |
| tensor<fp16, [1, 77, 1280]> var_1258_cast_fp16 = reshape(shape = var_1257, x = var_1255_cast_fp16)[name = string("op_1258_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349215872))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350444736))))[name = string("encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350455040)))]; |
| tensor<fp16, [1, 77, 1280]> linear_123_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_20_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_20_self_attn_out_proj_weight_to_fp16_palettized, x = var_1258_cast_fp16)[name = string("linear_123_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_247_cast_fp16 = add(x = input_243_cast_fp16, y = linear_123_cast_fp16)[name = string("input_247_cast_fp16")]; |
| tensor<int32, [1]> input_249_axes_0 = const()[name = string("input_249_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350457664)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350460288)))]; |
| tensor<fp16, [1, 77, 1280]> input_249_cast_fp16 = layer_norm(axes = input_249_axes_0, beta = encoder_text_model_encoder_layers_20_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_20_layer_norm2_weight_to_fp16, x = input_247_cast_fp16)[name = string("input_249_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350462912))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355378176))))[name = string("encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355419200)))]; |
| tensor<fp16, [1, 77, 5120]> linear_124_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_20_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_20_mlp_fc1_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = string("linear_124_cast_fp16")]; |
| string input_253_mode_0 = const()[name = string("input_253_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_253_cast_fp16 = gelu(mode = input_253_mode_0, x = linear_124_cast_fp16)[name = string("input_253_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355429504))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(360344768))))[name = string("encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(360355072)))]; |
| tensor<fp16, [1, 77, 1280]> linear_125_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_20_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_20_mlp_fc2_weight_to_fp16_palettized, x = input_253_cast_fp16)[name = string("linear_125_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_255_cast_fp16 = add(x = input_247_cast_fp16, y = linear_125_cast_fp16)[name = string("input_255_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_127_axes_0 = const()[name = string("hidden_states_127_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(360357696)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(360360320)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_127_cast_fp16 = layer_norm(axes = hidden_states_127_axes_0, beta = encoder_text_model_encoder_layers_21_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_21_layer_norm1_weight_to_fp16, x = input_255_cast_fp16)[name = string("hidden_states_127_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(360362944))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361591808))))[name = string("encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361602112)))]; |
| tensor<fp16, [1, 77, 1280]> linear_126_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_21_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_21_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_127_cast_fp16)[name = string("linear_126_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361604736))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362833600))))[name = string("encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362843904)))]; |
| tensor<fp16, [1, 77, 1280]> linear_127_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_21_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_21_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_127_cast_fp16)[name = string("linear_127_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362846528))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364075392))))[name = string("encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364085696)))]; |
| tensor<fp16, [1, 77, 1280]> linear_128_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_21_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_21_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_127_cast_fp16)[name = string("linear_128_cast_fp16")]; |
| tensor<int32, [4]> var_1301 = const()[name = string("op_1301"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1302_cast_fp16 = reshape(shape = var_1301, x = linear_126_cast_fp16)[name = string("op_1302_cast_fp16")]; |
| tensor<int32, [4]> var_1304 = const()[name = string("op_1304"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1305_cast_fp16 = reshape(shape = var_1304, x = linear_127_cast_fp16)[name = string("op_1305_cast_fp16")]; |
| tensor<int32, [4]> var_1307 = const()[name = string("op_1307"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1308_cast_fp16 = reshape(shape = var_1307, x = linear_128_cast_fp16)[name = string("op_1308_cast_fp16")]; |
| tensor<int32, [4]> value_43_perm_0 = const()[name = string("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_43_cast_fp16 = mul(x = var_1302_cast_fp16, y = var_11_to_fp16)[name = string("mul_43_cast_fp16")]; |
| bool matmul_21_transpose_y_0 = const()[name = string("matmul_21_transpose_y_0"), val = bool(true)]; |
| bool matmul_21_transpose_x_0 = const()[name = string("matmul_21_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_170_perm_0 = const()[name = string("transpose_170_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_171_perm_0 = const()[name = string("transpose_171_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_171 = transpose(perm = transpose_171_perm_0, x = var_1305_cast_fp16)[name = string("transpose_234")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_170 = transpose(perm = transpose_170_perm_0, x = mul_43_cast_fp16)[name = string("transpose_235")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_21_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = transpose_170, y = transpose_171)[name = string("matmul_21_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_21_cast_fp16 = add(x = matmul_21_cast_fp16, y = mul_0_to_fp16)[name = string("add_21_cast_fp16")]; |
| int32 softmax_21_axis_0 = const()[name = string("softmax_21_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_21_cast_fp16 = softmax(axis = softmax_21_axis_0, x = add_21_cast_fp16)[name = string("softmax_21_cast_fp16")]; |
| bool attn_output_85_transpose_x_0 = const()[name = string("attn_output_85_transpose_x_0"), val = bool(false)]; |
| bool attn_output_85_transpose_y_0 = const()[name = string("attn_output_85_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = var_1308_cast_fp16)[name = string("transpose_233")]; |
| tensor<fp16, [1, 20, 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 = softmax_21_cast_fp16, y = value_43_cast_fp16)[name = string("attn_output_85_cast_fp16")]; |
| tensor<int32, [4]> var_1311_perm_0 = const()[name = string("op_1311_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1313 = const()[name = string("op_1313"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1311_cast_fp16 = transpose(perm = var_1311_perm_0, x = attn_output_85_cast_fp16)[name = string("transpose_232")]; |
| tensor<fp16, [1, 77, 1280]> var_1314_cast_fp16 = reshape(shape = var_1313, x = var_1311_cast_fp16)[name = string("op_1314_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364088320))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365317184))))[name = string("encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365327488)))]; |
| tensor<fp16, [1, 77, 1280]> linear_129_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_21_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_21_self_attn_out_proj_weight_to_fp16_palettized, x = var_1314_cast_fp16)[name = string("linear_129_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_259_cast_fp16 = add(x = input_255_cast_fp16, y = linear_129_cast_fp16)[name = string("input_259_cast_fp16")]; |
| tensor<int32, [1]> input_261_axes_0 = const()[name = string("input_261_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365330112)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365332736)))]; |
| tensor<fp16, [1, 77, 1280]> input_261_cast_fp16 = layer_norm(axes = input_261_axes_0, beta = encoder_text_model_encoder_layers_21_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_21_layer_norm2_weight_to_fp16, x = input_259_cast_fp16)[name = string("input_261_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365335360))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370250624))))[name = string("encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370291648)))]; |
| tensor<fp16, [1, 77, 5120]> linear_130_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_21_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_21_mlp_fc1_weight_to_fp16_palettized, x = input_261_cast_fp16)[name = string("linear_130_cast_fp16")]; |
| string input_265_mode_0 = const()[name = string("input_265_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_265_cast_fp16 = gelu(mode = input_265_mode_0, x = linear_130_cast_fp16)[name = string("input_265_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370301952))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375217216))))[name = string("encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375227520)))]; |
| tensor<fp16, [1, 77, 1280]> linear_131_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_21_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_21_mlp_fc2_weight_to_fp16_palettized, x = input_265_cast_fp16)[name = string("linear_131_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_267_cast_fp16 = add(x = input_259_cast_fp16, y = linear_131_cast_fp16)[name = string("input_267_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_133_axes_0 = const()[name = string("hidden_states_133_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375230144)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375232768)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_133_cast_fp16 = layer_norm(axes = hidden_states_133_axes_0, beta = encoder_text_model_encoder_layers_22_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_22_layer_norm1_weight_to_fp16, x = input_267_cast_fp16)[name = string("hidden_states_133_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375235392))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376464256))))[name = string("encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376474560)))]; |
| tensor<fp16, [1, 77, 1280]> linear_132_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_22_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_22_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_133_cast_fp16)[name = string("linear_132_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376477184))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377706048))))[name = string("encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377716352)))]; |
| tensor<fp16, [1, 77, 1280]> linear_133_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_22_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_22_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_133_cast_fp16)[name = string("linear_133_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377718976))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378947840))))[name = string("encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378958144)))]; |
| tensor<fp16, [1, 77, 1280]> linear_134_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_22_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_22_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_133_cast_fp16)[name = string("linear_134_cast_fp16")]; |
| tensor<int32, [4]> var_1357 = const()[name = string("op_1357"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1358_cast_fp16 = reshape(shape = var_1357, x = linear_132_cast_fp16)[name = string("op_1358_cast_fp16")]; |
| tensor<int32, [4]> var_1360 = const()[name = string("op_1360"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1361_cast_fp16 = reshape(shape = var_1360, x = linear_133_cast_fp16)[name = string("op_1361_cast_fp16")]; |
| tensor<int32, [4]> var_1363 = const()[name = string("op_1363"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1364_cast_fp16 = reshape(shape = var_1363, x = linear_134_cast_fp16)[name = string("op_1364_cast_fp16")]; |
| tensor<int32, [4]> value_45_perm_0 = const()[name = string("value_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_45_cast_fp16 = mul(x = var_1358_cast_fp16, y = var_11_to_fp16)[name = string("mul_45_cast_fp16")]; |
| bool matmul_22_transpose_y_0 = const()[name = string("matmul_22_transpose_y_0"), val = bool(true)]; |
| bool matmul_22_transpose_x_0 = const()[name = string("matmul_22_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_172_perm_0 = const()[name = string("transpose_172_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_173_perm_0 = const()[name = string("transpose_173_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_173 = transpose(perm = transpose_173_perm_0, x = var_1361_cast_fp16)[name = string("transpose_230")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_172 = transpose(perm = transpose_172_perm_0, x = mul_45_cast_fp16)[name = string("transpose_231")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_22_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = transpose_172, y = transpose_173)[name = string("matmul_22_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_22_cast_fp16 = add(x = matmul_22_cast_fp16, y = mul_0_to_fp16)[name = string("add_22_cast_fp16")]; |
| int32 softmax_22_axis_0 = const()[name = string("softmax_22_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_22_cast_fp16 = softmax(axis = softmax_22_axis_0, x = add_22_cast_fp16)[name = string("softmax_22_cast_fp16")]; |
| bool attn_output_89_transpose_x_0 = const()[name = string("attn_output_89_transpose_x_0"), val = bool(false)]; |
| bool attn_output_89_transpose_y_0 = const()[name = string("attn_output_89_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_45_cast_fp16 = transpose(perm = value_45_perm_0, x = var_1364_cast_fp16)[name = string("transpose_229")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_89_cast_fp16 = matmul(transpose_x = attn_output_89_transpose_x_0, transpose_y = attn_output_89_transpose_y_0, x = softmax_22_cast_fp16, y = value_45_cast_fp16)[name = string("attn_output_89_cast_fp16")]; |
| tensor<int32, [4]> var_1367_perm_0 = const()[name = string("op_1367_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1369 = const()[name = string("op_1369"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1367_cast_fp16 = transpose(perm = var_1367_perm_0, x = attn_output_89_cast_fp16)[name = string("transpose_228")]; |
| tensor<fp16, [1, 77, 1280]> var_1370_cast_fp16 = reshape(shape = var_1369, x = var_1367_cast_fp16)[name = string("op_1370_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378960768))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380189632))))[name = string("encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380199936)))]; |
| tensor<fp16, [1, 77, 1280]> linear_135_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_22_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_22_self_attn_out_proj_weight_to_fp16_palettized, x = var_1370_cast_fp16)[name = string("linear_135_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_271_cast_fp16 = add(x = input_267_cast_fp16, y = linear_135_cast_fp16)[name = string("input_271_cast_fp16")]; |
| tensor<int32, [1]> input_273_axes_0 = const()[name = string("input_273_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380202560)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380205184)))]; |
| tensor<fp16, [1, 77, 1280]> input_273_cast_fp16 = layer_norm(axes = input_273_axes_0, beta = encoder_text_model_encoder_layers_22_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_22_layer_norm2_weight_to_fp16, x = input_271_cast_fp16)[name = string("input_273_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380207808))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385123072))))[name = string("encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385164096)))]; |
| tensor<fp16, [1, 77, 5120]> linear_136_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_22_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_22_mlp_fc1_weight_to_fp16_palettized, x = input_273_cast_fp16)[name = string("linear_136_cast_fp16")]; |
| string input_277_mode_0 = const()[name = string("input_277_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_277_cast_fp16 = gelu(mode = input_277_mode_0, x = linear_136_cast_fp16)[name = string("input_277_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385174400))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390089664))))[name = string("encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390099968)))]; |
| tensor<fp16, [1, 77, 1280]> linear_137_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_22_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_22_mlp_fc2_weight_to_fp16_palettized, x = input_277_cast_fp16)[name = string("linear_137_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_279_cast_fp16 = add(x = input_271_cast_fp16, y = linear_137_cast_fp16)[name = string("input_279_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_139_axes_0 = const()[name = string("hidden_states_139_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390102592)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390105216)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_139_cast_fp16 = layer_norm(axes = hidden_states_139_axes_0, beta = encoder_text_model_encoder_layers_23_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_23_layer_norm1_weight_to_fp16, x = input_279_cast_fp16)[name = string("hidden_states_139_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390107840))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391336704))))[name = string("encoder_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391347008)))]; |
| tensor<fp16, [1, 77, 1280]> linear_138_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_23_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_23_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_139_cast_fp16)[name = string("linear_138_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391349632))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392578496))))[name = string("encoder_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392588800)))]; |
| tensor<fp16, [1, 77, 1280]> linear_139_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_23_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_23_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_139_cast_fp16)[name = string("linear_139_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392591424))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(393820288))))[name = string("encoder_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(393830592)))]; |
| tensor<fp16, [1, 77, 1280]> linear_140_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_23_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_23_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_139_cast_fp16)[name = string("linear_140_cast_fp16")]; |
| tensor<int32, [4]> var_1413 = const()[name = string("op_1413"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1414_cast_fp16 = reshape(shape = var_1413, x = linear_138_cast_fp16)[name = string("op_1414_cast_fp16")]; |
| tensor<int32, [4]> var_1416 = const()[name = string("op_1416"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1417_cast_fp16 = reshape(shape = var_1416, x = linear_139_cast_fp16)[name = string("op_1417_cast_fp16")]; |
| tensor<int32, [4]> var_1419 = const()[name = string("op_1419"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1420_cast_fp16 = reshape(shape = var_1419, x = linear_140_cast_fp16)[name = string("op_1420_cast_fp16")]; |
| tensor<int32, [4]> value_47_perm_0 = const()[name = string("value_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_47_cast_fp16 = mul(x = var_1414_cast_fp16, y = var_11_to_fp16)[name = string("mul_47_cast_fp16")]; |
| bool matmul_23_transpose_y_0 = const()[name = string("matmul_23_transpose_y_0"), val = bool(true)]; |
| bool matmul_23_transpose_x_0 = const()[name = string("matmul_23_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_174_perm_0 = const()[name = string("transpose_174_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_175_perm_0 = const()[name = string("transpose_175_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_175 = transpose(perm = transpose_175_perm_0, x = var_1417_cast_fp16)[name = string("transpose_226")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_174 = transpose(perm = transpose_174_perm_0, x = mul_47_cast_fp16)[name = string("transpose_227")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_23_cast_fp16 = matmul(transpose_x = matmul_23_transpose_x_0, transpose_y = matmul_23_transpose_y_0, x = transpose_174, y = transpose_175)[name = string("matmul_23_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_23_cast_fp16 = add(x = matmul_23_cast_fp16, y = mul_0_to_fp16)[name = string("add_23_cast_fp16")]; |
| int32 softmax_23_axis_0 = const()[name = string("softmax_23_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_23_cast_fp16 = softmax(axis = softmax_23_axis_0, x = add_23_cast_fp16)[name = string("softmax_23_cast_fp16")]; |
| bool attn_output_93_transpose_x_0 = const()[name = string("attn_output_93_transpose_x_0"), val = bool(false)]; |
| bool attn_output_93_transpose_y_0 = const()[name = string("attn_output_93_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_47_cast_fp16 = transpose(perm = value_47_perm_0, x = var_1420_cast_fp16)[name = string("transpose_225")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_93_cast_fp16 = matmul(transpose_x = attn_output_93_transpose_x_0, transpose_y = attn_output_93_transpose_y_0, x = softmax_23_cast_fp16, y = value_47_cast_fp16)[name = string("attn_output_93_cast_fp16")]; |
| tensor<int32, [4]> var_1423_perm_0 = const()[name = string("op_1423_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1425 = const()[name = string("op_1425"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1423_cast_fp16 = transpose(perm = var_1423_perm_0, x = attn_output_93_cast_fp16)[name = string("transpose_224")]; |
| tensor<fp16, [1, 77, 1280]> var_1426_cast_fp16 = reshape(shape = var_1425, x = var_1423_cast_fp16)[name = string("op_1426_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(393833216))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395062080))))[name = string("encoder_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395072384)))]; |
| tensor<fp16, [1, 77, 1280]> linear_141_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_23_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_23_self_attn_out_proj_weight_to_fp16_palettized, x = var_1426_cast_fp16)[name = string("linear_141_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_283_cast_fp16 = add(x = input_279_cast_fp16, y = linear_141_cast_fp16)[name = string("input_283_cast_fp16")]; |
| tensor<int32, [1]> input_285_axes_0 = const()[name = string("input_285_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395075008)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395077632)))]; |
| tensor<fp16, [1, 77, 1280]> input_285_cast_fp16 = layer_norm(axes = input_285_axes_0, beta = encoder_text_model_encoder_layers_23_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_23_layer_norm2_weight_to_fp16, x = input_283_cast_fp16)[name = string("input_285_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395080256))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399995520))))[name = string("encoder_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(400036544)))]; |
| tensor<fp16, [1, 77, 5120]> linear_142_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_23_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_23_mlp_fc1_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = string("linear_142_cast_fp16")]; |
| string input_289_mode_0 = const()[name = string("input_289_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_289_cast_fp16 = gelu(mode = input_289_mode_0, x = linear_142_cast_fp16)[name = string("input_289_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(400046848))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404962112))))[name = string("encoder_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404972416)))]; |
| tensor<fp16, [1, 77, 1280]> linear_143_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_23_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_23_mlp_fc2_weight_to_fp16_palettized, x = input_289_cast_fp16)[name = string("linear_143_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_291_cast_fp16 = add(x = input_283_cast_fp16, y = linear_143_cast_fp16)[name = string("input_291_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_145_axes_0 = const()[name = string("hidden_states_145_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404975040)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404977664)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_145_cast_fp16 = layer_norm(axes = hidden_states_145_axes_0, beta = encoder_text_model_encoder_layers_24_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_24_layer_norm1_weight_to_fp16, x = input_291_cast_fp16)[name = string("hidden_states_145_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404980288))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406209152))))[name = string("encoder_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406219456)))]; |
| tensor<fp16, [1, 77, 1280]> linear_144_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_24_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_24_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_145_cast_fp16)[name = string("linear_144_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406222080))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(407450944))))[name = string("encoder_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(407461248)))]; |
| tensor<fp16, [1, 77, 1280]> linear_145_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_24_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_24_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_145_cast_fp16)[name = string("linear_145_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(407463872))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408692736))))[name = string("encoder_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408703040)))]; |
| tensor<fp16, [1, 77, 1280]> linear_146_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_24_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_24_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_145_cast_fp16)[name = string("linear_146_cast_fp16")]; |
| tensor<int32, [4]> var_1469 = const()[name = string("op_1469"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1470_cast_fp16 = reshape(shape = var_1469, x = linear_144_cast_fp16)[name = string("op_1470_cast_fp16")]; |
| tensor<int32, [4]> var_1472 = const()[name = string("op_1472"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1473_cast_fp16 = reshape(shape = var_1472, x = linear_145_cast_fp16)[name = string("op_1473_cast_fp16")]; |
| tensor<int32, [4]> var_1475 = const()[name = string("op_1475"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1476_cast_fp16 = reshape(shape = var_1475, x = linear_146_cast_fp16)[name = string("op_1476_cast_fp16")]; |
| tensor<int32, [4]> value_49_perm_0 = const()[name = string("value_49_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_49_cast_fp16 = mul(x = var_1470_cast_fp16, y = var_11_to_fp16)[name = string("mul_49_cast_fp16")]; |
| bool matmul_24_transpose_y_0 = const()[name = string("matmul_24_transpose_y_0"), val = bool(true)]; |
| bool matmul_24_transpose_x_0 = const()[name = string("matmul_24_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_176_perm_0 = const()[name = string("transpose_176_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_177_perm_0 = const()[name = string("transpose_177_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_177 = transpose(perm = transpose_177_perm_0, x = var_1473_cast_fp16)[name = string("transpose_222")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_176 = transpose(perm = transpose_176_perm_0, x = mul_49_cast_fp16)[name = string("transpose_223")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_24_cast_fp16 = matmul(transpose_x = matmul_24_transpose_x_0, transpose_y = matmul_24_transpose_y_0, x = transpose_176, y = transpose_177)[name = string("matmul_24_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_24_cast_fp16 = add(x = matmul_24_cast_fp16, y = mul_0_to_fp16)[name = string("add_24_cast_fp16")]; |
| int32 softmax_24_axis_0 = const()[name = string("softmax_24_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_24_cast_fp16 = softmax(axis = softmax_24_axis_0, x = add_24_cast_fp16)[name = string("softmax_24_cast_fp16")]; |
| bool attn_output_97_transpose_x_0 = const()[name = string("attn_output_97_transpose_x_0"), val = bool(false)]; |
| bool attn_output_97_transpose_y_0 = const()[name = string("attn_output_97_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_49_cast_fp16 = transpose(perm = value_49_perm_0, x = var_1476_cast_fp16)[name = string("transpose_221")]; |
| tensor<fp16, [1, 20, 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 = softmax_24_cast_fp16, y = value_49_cast_fp16)[name = string("attn_output_97_cast_fp16")]; |
| tensor<int32, [4]> var_1479_perm_0 = const()[name = string("op_1479_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1481 = const()[name = string("op_1481"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1479_cast_fp16 = transpose(perm = var_1479_perm_0, x = attn_output_97_cast_fp16)[name = string("transpose_220")]; |
| tensor<fp16, [1, 77, 1280]> var_1482_cast_fp16 = reshape(shape = var_1481, x = var_1479_cast_fp16)[name = string("op_1482_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408705664))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409934528))))[name = string("encoder_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409944832)))]; |
| tensor<fp16, [1, 77, 1280]> linear_147_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_24_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_24_self_attn_out_proj_weight_to_fp16_palettized, x = var_1482_cast_fp16)[name = string("linear_147_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_295_cast_fp16 = add(x = input_291_cast_fp16, y = linear_147_cast_fp16)[name = string("input_295_cast_fp16")]; |
| tensor<int32, [1]> input_297_axes_0 = const()[name = string("input_297_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409947456)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409950080)))]; |
| tensor<fp16, [1, 77, 1280]> input_297_cast_fp16 = layer_norm(axes = input_297_axes_0, beta = encoder_text_model_encoder_layers_24_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_24_layer_norm2_weight_to_fp16, x = input_295_cast_fp16)[name = string("input_297_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409952704))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414867968))))[name = string("encoder_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414908992)))]; |
| tensor<fp16, [1, 77, 5120]> linear_148_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_24_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_24_mlp_fc1_weight_to_fp16_palettized, x = input_297_cast_fp16)[name = string("linear_148_cast_fp16")]; |
| string input_301_mode_0 = const()[name = string("input_301_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_301_cast_fp16 = gelu(mode = input_301_mode_0, x = linear_148_cast_fp16)[name = string("input_301_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414919296))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419834560))))[name = string("encoder_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419844864)))]; |
| tensor<fp16, [1, 77, 1280]> linear_149_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_24_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_24_mlp_fc2_weight_to_fp16_palettized, x = input_301_cast_fp16)[name = string("linear_149_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_303_cast_fp16 = add(x = input_295_cast_fp16, y = linear_149_cast_fp16)[name = string("input_303_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_151_axes_0 = const()[name = string("hidden_states_151_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419847488)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419850112)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_151_cast_fp16 = layer_norm(axes = hidden_states_151_axes_0, beta = encoder_text_model_encoder_layers_25_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_25_layer_norm1_weight_to_fp16, x = input_303_cast_fp16)[name = string("hidden_states_151_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419852736))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421081600))))[name = string("encoder_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421091904)))]; |
| tensor<fp16, [1, 77, 1280]> linear_150_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_25_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_25_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_151_cast_fp16)[name = string("linear_150_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421094528))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422323392))))[name = string("encoder_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422333696)))]; |
| tensor<fp16, [1, 77, 1280]> linear_151_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_25_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_25_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_151_cast_fp16)[name = string("linear_151_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422336320))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423565184))))[name = string("encoder_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423575488)))]; |
| tensor<fp16, [1, 77, 1280]> linear_152_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_25_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_25_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_151_cast_fp16)[name = string("linear_152_cast_fp16")]; |
| tensor<int32, [4]> var_1525 = const()[name = string("op_1525"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1526_cast_fp16 = reshape(shape = var_1525, x = linear_150_cast_fp16)[name = string("op_1526_cast_fp16")]; |
| tensor<int32, [4]> var_1528 = const()[name = string("op_1528"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1529_cast_fp16 = reshape(shape = var_1528, x = linear_151_cast_fp16)[name = string("op_1529_cast_fp16")]; |
| tensor<int32, [4]> var_1531 = const()[name = string("op_1531"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1532_cast_fp16 = reshape(shape = var_1531, x = linear_152_cast_fp16)[name = string("op_1532_cast_fp16")]; |
| tensor<int32, [4]> value_51_perm_0 = const()[name = string("value_51_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_51_cast_fp16 = mul(x = var_1526_cast_fp16, y = var_11_to_fp16)[name = string("mul_51_cast_fp16")]; |
| bool matmul_25_transpose_y_0 = const()[name = string("matmul_25_transpose_y_0"), val = bool(true)]; |
| bool matmul_25_transpose_x_0 = const()[name = string("matmul_25_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_178_perm_0 = const()[name = string("transpose_178_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_179_perm_0 = const()[name = string("transpose_179_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_179 = transpose(perm = transpose_179_perm_0, x = var_1529_cast_fp16)[name = string("transpose_218")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_178 = transpose(perm = transpose_178_perm_0, x = mul_51_cast_fp16)[name = string("transpose_219")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_25_cast_fp16 = matmul(transpose_x = matmul_25_transpose_x_0, transpose_y = matmul_25_transpose_y_0, x = transpose_178, y = transpose_179)[name = string("matmul_25_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_25_cast_fp16 = add(x = matmul_25_cast_fp16, y = mul_0_to_fp16)[name = string("add_25_cast_fp16")]; |
| int32 softmax_25_axis_0 = const()[name = string("softmax_25_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_25_cast_fp16 = softmax(axis = softmax_25_axis_0, x = add_25_cast_fp16)[name = string("softmax_25_cast_fp16")]; |
| bool attn_output_101_transpose_x_0 = const()[name = string("attn_output_101_transpose_x_0"), val = bool(false)]; |
| bool attn_output_101_transpose_y_0 = const()[name = string("attn_output_101_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_51_cast_fp16 = transpose(perm = value_51_perm_0, x = var_1532_cast_fp16)[name = string("transpose_217")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_101_cast_fp16 = matmul(transpose_x = attn_output_101_transpose_x_0, transpose_y = attn_output_101_transpose_y_0, x = softmax_25_cast_fp16, y = value_51_cast_fp16)[name = string("attn_output_101_cast_fp16")]; |
| tensor<int32, [4]> var_1535_perm_0 = const()[name = string("op_1535_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1537 = const()[name = string("op_1537"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1535_cast_fp16 = transpose(perm = var_1535_perm_0, x = attn_output_101_cast_fp16)[name = string("transpose_216")]; |
| tensor<fp16, [1, 77, 1280]> var_1538_cast_fp16 = reshape(shape = var_1537, x = var_1535_cast_fp16)[name = string("op_1538_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423578112))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424806976))))[name = string("encoder_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424817280)))]; |
| tensor<fp16, [1, 77, 1280]> linear_153_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_25_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_25_self_attn_out_proj_weight_to_fp16_palettized, x = var_1538_cast_fp16)[name = string("linear_153_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_307_cast_fp16 = add(x = input_303_cast_fp16, y = linear_153_cast_fp16)[name = string("input_307_cast_fp16")]; |
| tensor<int32, [1]> input_309_axes_0 = const()[name = string("input_309_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424819904)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424822528)))]; |
| tensor<fp16, [1, 77, 1280]> input_309_cast_fp16 = layer_norm(axes = input_309_axes_0, beta = encoder_text_model_encoder_layers_25_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_25_layer_norm2_weight_to_fp16, x = input_307_cast_fp16)[name = string("input_309_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424825152))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429740416))))[name = string("encoder_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429781440)))]; |
| tensor<fp16, [1, 77, 5120]> linear_154_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_25_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_25_mlp_fc1_weight_to_fp16_palettized, x = input_309_cast_fp16)[name = string("linear_154_cast_fp16")]; |
| string input_313_mode_0 = const()[name = string("input_313_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_313_cast_fp16 = gelu(mode = input_313_mode_0, x = linear_154_cast_fp16)[name = string("input_313_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429791744))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434707008))))[name = string("encoder_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434717312)))]; |
| tensor<fp16, [1, 77, 1280]> linear_155_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_25_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_25_mlp_fc2_weight_to_fp16_palettized, x = input_313_cast_fp16)[name = string("linear_155_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_315_cast_fp16 = add(x = input_307_cast_fp16, y = linear_155_cast_fp16)[name = string("input_315_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_157_axes_0 = const()[name = string("hidden_states_157_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434719936)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434722560)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_157_cast_fp16 = layer_norm(axes = hidden_states_157_axes_0, beta = encoder_text_model_encoder_layers_26_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_26_layer_norm1_weight_to_fp16, x = input_315_cast_fp16)[name = string("hidden_states_157_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434725184))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(435954048))))[name = string("encoder_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(435964352)))]; |
| tensor<fp16, [1, 77, 1280]> linear_156_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_26_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_26_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_157_cast_fp16)[name = string("linear_156_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(435966976))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437195840))))[name = string("encoder_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437206144)))]; |
| tensor<fp16, [1, 77, 1280]> linear_157_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_26_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_26_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_157_cast_fp16)[name = string("linear_157_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437208768))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438437632))))[name = string("encoder_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438447936)))]; |
| tensor<fp16, [1, 77, 1280]> linear_158_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_26_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_26_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_157_cast_fp16)[name = string("linear_158_cast_fp16")]; |
| tensor<int32, [4]> var_1581 = const()[name = string("op_1581"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1582_cast_fp16 = reshape(shape = var_1581, x = linear_156_cast_fp16)[name = string("op_1582_cast_fp16")]; |
| tensor<int32, [4]> var_1584 = const()[name = string("op_1584"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1585_cast_fp16 = reshape(shape = var_1584, x = linear_157_cast_fp16)[name = string("op_1585_cast_fp16")]; |
| tensor<int32, [4]> var_1587 = const()[name = string("op_1587"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1588_cast_fp16 = reshape(shape = var_1587, x = linear_158_cast_fp16)[name = string("op_1588_cast_fp16")]; |
| tensor<int32, [4]> value_53_perm_0 = const()[name = string("value_53_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_53_cast_fp16 = mul(x = var_1582_cast_fp16, y = var_11_to_fp16)[name = string("mul_53_cast_fp16")]; |
| bool matmul_26_transpose_y_0 = const()[name = string("matmul_26_transpose_y_0"), val = bool(true)]; |
| bool matmul_26_transpose_x_0 = const()[name = string("matmul_26_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_180_perm_0 = const()[name = string("transpose_180_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_181_perm_0 = const()[name = string("transpose_181_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_181 = transpose(perm = transpose_181_perm_0, x = var_1585_cast_fp16)[name = string("transpose_214")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_180 = transpose(perm = transpose_180_perm_0, x = mul_53_cast_fp16)[name = string("transpose_215")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_26_cast_fp16 = matmul(transpose_x = matmul_26_transpose_x_0, transpose_y = matmul_26_transpose_y_0, x = transpose_180, y = transpose_181)[name = string("matmul_26_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_26_cast_fp16 = add(x = matmul_26_cast_fp16, y = mul_0_to_fp16)[name = string("add_26_cast_fp16")]; |
| int32 softmax_26_axis_0 = const()[name = string("softmax_26_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_26_cast_fp16 = softmax(axis = softmax_26_axis_0, x = add_26_cast_fp16)[name = string("softmax_26_cast_fp16")]; |
| bool attn_output_105_transpose_x_0 = const()[name = string("attn_output_105_transpose_x_0"), val = bool(false)]; |
| bool attn_output_105_transpose_y_0 = const()[name = string("attn_output_105_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_53_cast_fp16 = transpose(perm = value_53_perm_0, x = var_1588_cast_fp16)[name = string("transpose_213")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_105_cast_fp16 = matmul(transpose_x = attn_output_105_transpose_x_0, transpose_y = attn_output_105_transpose_y_0, x = softmax_26_cast_fp16, y = value_53_cast_fp16)[name = string("attn_output_105_cast_fp16")]; |
| tensor<int32, [4]> var_1591_perm_0 = const()[name = string("op_1591_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1593 = const()[name = string("op_1593"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1591_cast_fp16 = transpose(perm = var_1591_perm_0, x = attn_output_105_cast_fp16)[name = string("transpose_212")]; |
| tensor<fp16, [1, 77, 1280]> var_1594_cast_fp16 = reshape(shape = var_1593, x = var_1591_cast_fp16)[name = string("op_1594_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438450560))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(439679424))))[name = string("encoder_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(439689728)))]; |
| tensor<fp16, [1, 77, 1280]> linear_159_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_26_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_26_self_attn_out_proj_weight_to_fp16_palettized, x = var_1594_cast_fp16)[name = string("linear_159_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_319_cast_fp16 = add(x = input_315_cast_fp16, y = linear_159_cast_fp16)[name = string("input_319_cast_fp16")]; |
| tensor<int32, [1]> input_321_axes_0 = const()[name = string("input_321_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(439692352)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(439694976)))]; |
| tensor<fp16, [1, 77, 1280]> input_321_cast_fp16 = layer_norm(axes = input_321_axes_0, beta = encoder_text_model_encoder_layers_26_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_26_layer_norm2_weight_to_fp16, x = input_319_cast_fp16)[name = string("input_321_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(439697600))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444612864))))[name = string("encoder_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444653888)))]; |
| tensor<fp16, [1, 77, 5120]> linear_160_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_26_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_26_mlp_fc1_weight_to_fp16_palettized, x = input_321_cast_fp16)[name = string("linear_160_cast_fp16")]; |
| string input_325_mode_0 = const()[name = string("input_325_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_325_cast_fp16 = gelu(mode = input_325_mode_0, x = linear_160_cast_fp16)[name = string("input_325_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444664192))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449579456))))[name = string("encoder_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449589760)))]; |
| tensor<fp16, [1, 77, 1280]> linear_161_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_26_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_26_mlp_fc2_weight_to_fp16_palettized, x = input_325_cast_fp16)[name = string("linear_161_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_327_cast_fp16 = add(x = input_319_cast_fp16, y = linear_161_cast_fp16)[name = string("input_327_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_163_axes_0 = const()[name = string("hidden_states_163_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449592384)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449595008)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_163_cast_fp16 = layer_norm(axes = hidden_states_163_axes_0, beta = encoder_text_model_encoder_layers_27_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_27_layer_norm1_weight_to_fp16, x = input_327_cast_fp16)[name = string("hidden_states_163_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_27_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449597632))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450826496))))[name = string("encoder_text_model_encoder_layers_27_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450836800)))]; |
| tensor<fp16, [1, 77, 1280]> linear_162_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_27_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_27_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_163_cast_fp16)[name = string("linear_162_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_27_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450839424))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(452068288))))[name = string("encoder_text_model_encoder_layers_27_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(452078592)))]; |
| tensor<fp16, [1, 77, 1280]> linear_163_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_27_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_27_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_163_cast_fp16)[name = string("linear_163_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_27_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(452081216))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453310080))))[name = string("encoder_text_model_encoder_layers_27_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453320384)))]; |
| tensor<fp16, [1, 77, 1280]> linear_164_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_27_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_27_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_163_cast_fp16)[name = string("linear_164_cast_fp16")]; |
| tensor<int32, [4]> var_1637 = const()[name = string("op_1637"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1638_cast_fp16 = reshape(shape = var_1637, x = linear_162_cast_fp16)[name = string("op_1638_cast_fp16")]; |
| tensor<int32, [4]> var_1640 = const()[name = string("op_1640"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1641_cast_fp16 = reshape(shape = var_1640, x = linear_163_cast_fp16)[name = string("op_1641_cast_fp16")]; |
| tensor<int32, [4]> var_1643 = const()[name = string("op_1643"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1644_cast_fp16 = reshape(shape = var_1643, x = linear_164_cast_fp16)[name = string("op_1644_cast_fp16")]; |
| tensor<int32, [4]> value_55_perm_0 = const()[name = string("value_55_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_55_cast_fp16 = mul(x = var_1638_cast_fp16, y = var_11_to_fp16)[name = string("mul_55_cast_fp16")]; |
| bool matmul_27_transpose_y_0 = const()[name = string("matmul_27_transpose_y_0"), val = bool(true)]; |
| bool matmul_27_transpose_x_0 = const()[name = string("matmul_27_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_182_perm_0 = const()[name = string("transpose_182_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_183_perm_0 = const()[name = string("transpose_183_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_183 = transpose(perm = transpose_183_perm_0, x = var_1641_cast_fp16)[name = string("transpose_210")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_182 = transpose(perm = transpose_182_perm_0, x = mul_55_cast_fp16)[name = string("transpose_211")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_27_cast_fp16 = matmul(transpose_x = matmul_27_transpose_x_0, transpose_y = matmul_27_transpose_y_0, x = transpose_182, y = transpose_183)[name = string("matmul_27_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_27_cast_fp16 = add(x = matmul_27_cast_fp16, y = mul_0_to_fp16)[name = string("add_27_cast_fp16")]; |
| int32 softmax_27_axis_0 = const()[name = string("softmax_27_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_27_cast_fp16 = softmax(axis = softmax_27_axis_0, x = add_27_cast_fp16)[name = string("softmax_27_cast_fp16")]; |
| bool attn_output_109_transpose_x_0 = const()[name = string("attn_output_109_transpose_x_0"), val = bool(false)]; |
| bool attn_output_109_transpose_y_0 = const()[name = string("attn_output_109_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_55_cast_fp16 = transpose(perm = value_55_perm_0, x = var_1644_cast_fp16)[name = string("transpose_209")]; |
| tensor<fp16, [1, 20, 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 = softmax_27_cast_fp16, y = value_55_cast_fp16)[name = string("attn_output_109_cast_fp16")]; |
| tensor<int32, [4]> var_1647_perm_0 = const()[name = string("op_1647_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1649 = const()[name = string("op_1649"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1647_cast_fp16 = transpose(perm = var_1647_perm_0, x = attn_output_109_cast_fp16)[name = string("transpose_208")]; |
| tensor<fp16, [1, 77, 1280]> var_1650_cast_fp16 = reshape(shape = var_1649, x = var_1647_cast_fp16)[name = string("op_1650_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_27_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453323008))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454551872))))[name = string("encoder_text_model_encoder_layers_27_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454562176)))]; |
| tensor<fp16, [1, 77, 1280]> linear_165_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_27_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_27_self_attn_out_proj_weight_to_fp16_palettized, x = var_1650_cast_fp16)[name = string("linear_165_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_331_cast_fp16 = add(x = input_327_cast_fp16, y = linear_165_cast_fp16)[name = string("input_331_cast_fp16")]; |
| tensor<int32, [1]> input_333_axes_0 = const()[name = string("input_333_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454564800)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454567424)))]; |
| tensor<fp16, [1, 77, 1280]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = encoder_text_model_encoder_layers_27_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_27_layer_norm2_weight_to_fp16, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_27_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454570048))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(459485312))))[name = string("encoder_text_model_encoder_layers_27_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_27_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(459526336)))]; |
| tensor<fp16, [1, 77, 5120]> linear_166_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_27_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_27_mlp_fc1_weight_to_fp16_palettized, x = input_333_cast_fp16)[name = string("linear_166_cast_fp16")]; |
| string input_337_mode_0 = const()[name = string("input_337_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_337_cast_fp16 = gelu(mode = input_337_mode_0, x = linear_166_cast_fp16)[name = string("input_337_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_27_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(459536640))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(464451904))))[name = string("encoder_text_model_encoder_layers_27_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_27_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_27_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(464462208)))]; |
| tensor<fp16, [1, 77, 1280]> linear_167_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_27_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_27_mlp_fc2_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = string("linear_167_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_339_cast_fp16 = add(x = input_331_cast_fp16, y = linear_167_cast_fp16)[name = string("input_339_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_169_axes_0 = const()[name = string("hidden_states_169_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(464464832)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(464467456)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_169_cast_fp16 = layer_norm(axes = hidden_states_169_axes_0, beta = encoder_text_model_encoder_layers_28_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_28_layer_norm1_weight_to_fp16, x = input_339_cast_fp16)[name = string("hidden_states_169_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_28_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(464470080))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(465698944))))[name = string("encoder_text_model_encoder_layers_28_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(465709248)))]; |
| tensor<fp16, [1, 77, 1280]> linear_168_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_28_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_28_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_169_cast_fp16)[name = string("linear_168_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_28_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(465711872))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(466940736))))[name = string("encoder_text_model_encoder_layers_28_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(466951040)))]; |
| tensor<fp16, [1, 77, 1280]> linear_169_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_28_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_28_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_169_cast_fp16)[name = string("linear_169_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_28_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(466953664))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(468182528))))[name = string("encoder_text_model_encoder_layers_28_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(468192832)))]; |
| tensor<fp16, [1, 77, 1280]> linear_170_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_28_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_28_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_169_cast_fp16)[name = string("linear_170_cast_fp16")]; |
| tensor<int32, [4]> var_1693 = const()[name = string("op_1693"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1694_cast_fp16 = reshape(shape = var_1693, x = linear_168_cast_fp16)[name = string("op_1694_cast_fp16")]; |
| tensor<int32, [4]> var_1696 = const()[name = string("op_1696"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1697_cast_fp16 = reshape(shape = var_1696, x = linear_169_cast_fp16)[name = string("op_1697_cast_fp16")]; |
| tensor<int32, [4]> var_1699 = const()[name = string("op_1699"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1700_cast_fp16 = reshape(shape = var_1699, x = linear_170_cast_fp16)[name = string("op_1700_cast_fp16")]; |
| tensor<int32, [4]> value_57_perm_0 = const()[name = string("value_57_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_57_cast_fp16 = mul(x = var_1694_cast_fp16, y = var_11_to_fp16)[name = string("mul_57_cast_fp16")]; |
| bool matmul_28_transpose_y_0 = const()[name = string("matmul_28_transpose_y_0"), val = bool(true)]; |
| bool matmul_28_transpose_x_0 = const()[name = string("matmul_28_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_184_perm_0 = const()[name = string("transpose_184_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_185_perm_0 = const()[name = string("transpose_185_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_185 = transpose(perm = transpose_185_perm_0, x = var_1697_cast_fp16)[name = string("transpose_206")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_184 = transpose(perm = transpose_184_perm_0, x = mul_57_cast_fp16)[name = string("transpose_207")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_28_cast_fp16 = matmul(transpose_x = matmul_28_transpose_x_0, transpose_y = matmul_28_transpose_y_0, x = transpose_184, y = transpose_185)[name = string("matmul_28_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_28_cast_fp16 = add(x = matmul_28_cast_fp16, y = mul_0_to_fp16)[name = string("add_28_cast_fp16")]; |
| int32 softmax_28_axis_0 = const()[name = string("softmax_28_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_28_cast_fp16 = softmax(axis = softmax_28_axis_0, x = add_28_cast_fp16)[name = string("softmax_28_cast_fp16")]; |
| bool attn_output_113_transpose_x_0 = const()[name = string("attn_output_113_transpose_x_0"), val = bool(false)]; |
| bool attn_output_113_transpose_y_0 = const()[name = string("attn_output_113_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_57_cast_fp16 = transpose(perm = value_57_perm_0, x = var_1700_cast_fp16)[name = string("transpose_205")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_113_cast_fp16 = matmul(transpose_x = attn_output_113_transpose_x_0, transpose_y = attn_output_113_transpose_y_0, x = softmax_28_cast_fp16, y = value_57_cast_fp16)[name = string("attn_output_113_cast_fp16")]; |
| tensor<int32, [4]> var_1703_perm_0 = const()[name = string("op_1703_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1705 = const()[name = string("op_1705"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1703_cast_fp16 = transpose(perm = var_1703_perm_0, x = attn_output_113_cast_fp16)[name = string("transpose_204")]; |
| tensor<fp16, [1, 77, 1280]> var_1706_cast_fp16 = reshape(shape = var_1705, x = var_1703_cast_fp16)[name = string("op_1706_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_28_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(468195456))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469424320))))[name = string("encoder_text_model_encoder_layers_28_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469434624)))]; |
| tensor<fp16, [1, 77, 1280]> linear_171_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_28_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_28_self_attn_out_proj_weight_to_fp16_palettized, x = var_1706_cast_fp16)[name = string("linear_171_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_343_cast_fp16 = add(x = input_339_cast_fp16, y = linear_171_cast_fp16)[name = string("input_343_cast_fp16")]; |
| tensor<int32, [1]> input_345_axes_0 = const()[name = string("input_345_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469437248)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469439872)))]; |
| tensor<fp16, [1, 77, 1280]> input_345_cast_fp16 = layer_norm(axes = input_345_axes_0, beta = encoder_text_model_encoder_layers_28_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_28_layer_norm2_weight_to_fp16, x = input_343_cast_fp16)[name = string("input_345_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_28_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469442496))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(474357760))))[name = string("encoder_text_model_encoder_layers_28_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_28_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(474398784)))]; |
| tensor<fp16, [1, 77, 5120]> linear_172_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_28_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_28_mlp_fc1_weight_to_fp16_palettized, x = input_345_cast_fp16)[name = string("linear_172_cast_fp16")]; |
| string input_349_mode_0 = const()[name = string("input_349_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_349_cast_fp16 = gelu(mode = input_349_mode_0, x = linear_172_cast_fp16)[name = string("input_349_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_28_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(474409088))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(479324352))))[name = string("encoder_text_model_encoder_layers_28_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_28_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_28_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(479334656)))]; |
| tensor<fp16, [1, 77, 1280]> linear_173_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_28_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_28_mlp_fc2_weight_to_fp16_palettized, x = input_349_cast_fp16)[name = string("linear_173_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_173_cast_fp16)[name = string("input_351_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_175_axes_0 = const()[name = string("hidden_states_175_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(479337280)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(479339904)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_175_cast_fp16 = layer_norm(axes = hidden_states_175_axes_0, beta = encoder_text_model_encoder_layers_29_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_29_layer_norm1_weight_to_fp16, x = input_351_cast_fp16)[name = string("hidden_states_175_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_29_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(479342528))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480571392))))[name = string("encoder_text_model_encoder_layers_29_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480581696)))]; |
| tensor<fp16, [1, 77, 1280]> linear_174_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_29_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_29_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_175_cast_fp16)[name = string("linear_174_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_29_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480584320))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(481813184))))[name = string("encoder_text_model_encoder_layers_29_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(481823488)))]; |
| tensor<fp16, [1, 77, 1280]> linear_175_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_29_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_29_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_175_cast_fp16)[name = string("linear_175_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_29_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(481826112))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(483054976))))[name = string("encoder_text_model_encoder_layers_29_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(483065280)))]; |
| tensor<fp16, [1, 77, 1280]> linear_176_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_29_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_29_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_175_cast_fp16)[name = string("linear_176_cast_fp16")]; |
| tensor<int32, [4]> var_1749 = const()[name = string("op_1749"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1750_cast_fp16 = reshape(shape = var_1749, x = linear_174_cast_fp16)[name = string("op_1750_cast_fp16")]; |
| tensor<int32, [4]> var_1752 = const()[name = string("op_1752"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1753_cast_fp16 = reshape(shape = var_1752, x = linear_175_cast_fp16)[name = string("op_1753_cast_fp16")]; |
| tensor<int32, [4]> var_1755 = const()[name = string("op_1755"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1756_cast_fp16 = reshape(shape = var_1755, x = linear_176_cast_fp16)[name = string("op_1756_cast_fp16")]; |
| tensor<int32, [4]> value_59_perm_0 = const()[name = string("value_59_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_59_cast_fp16 = mul(x = var_1750_cast_fp16, y = var_11_to_fp16)[name = string("mul_59_cast_fp16")]; |
| bool matmul_29_transpose_y_0 = const()[name = string("matmul_29_transpose_y_0"), val = bool(true)]; |
| bool matmul_29_transpose_x_0 = const()[name = string("matmul_29_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_186_perm_0 = const()[name = string("transpose_186_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_187_perm_0 = const()[name = string("transpose_187_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_187 = transpose(perm = transpose_187_perm_0, x = var_1753_cast_fp16)[name = string("transpose_202")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_186 = transpose(perm = transpose_186_perm_0, x = mul_59_cast_fp16)[name = string("transpose_203")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_29_cast_fp16 = matmul(transpose_x = matmul_29_transpose_x_0, transpose_y = matmul_29_transpose_y_0, x = transpose_186, y = transpose_187)[name = string("matmul_29_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_29_cast_fp16 = add(x = matmul_29_cast_fp16, y = mul_0_to_fp16)[name = string("add_29_cast_fp16")]; |
| int32 softmax_29_axis_0 = const()[name = string("softmax_29_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_29_cast_fp16 = softmax(axis = softmax_29_axis_0, x = add_29_cast_fp16)[name = string("softmax_29_cast_fp16")]; |
| bool attn_output_117_transpose_x_0 = const()[name = string("attn_output_117_transpose_x_0"), val = bool(false)]; |
| bool attn_output_117_transpose_y_0 = const()[name = string("attn_output_117_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_59_cast_fp16 = transpose(perm = value_59_perm_0, x = var_1756_cast_fp16)[name = string("transpose_201")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_117_cast_fp16 = matmul(transpose_x = attn_output_117_transpose_x_0, transpose_y = attn_output_117_transpose_y_0, x = softmax_29_cast_fp16, y = value_59_cast_fp16)[name = string("attn_output_117_cast_fp16")]; |
| tensor<int32, [4]> var_1759_perm_0 = const()[name = string("op_1759_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1761 = const()[name = string("op_1761"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1759_cast_fp16 = transpose(perm = var_1759_perm_0, x = attn_output_117_cast_fp16)[name = string("transpose_200")]; |
| tensor<fp16, [1, 77, 1280]> var_1762_cast_fp16 = reshape(shape = var_1761, x = var_1759_cast_fp16)[name = string("op_1762_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_29_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(483067904))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(484296768))))[name = string("encoder_text_model_encoder_layers_29_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(484307072)))]; |
| tensor<fp16, [1, 77, 1280]> linear_177_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_29_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_29_self_attn_out_proj_weight_to_fp16_palettized, x = var_1762_cast_fp16)[name = string("linear_177_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_355_cast_fp16 = add(x = input_351_cast_fp16, y = linear_177_cast_fp16)[name = string("input_355_cast_fp16")]; |
| tensor<int32, [1]> input_357_axes_0 = const()[name = string("input_357_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(484309696)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(484312320)))]; |
| tensor<fp16, [1, 77, 1280]> input_357_cast_fp16 = layer_norm(axes = input_357_axes_0, beta = encoder_text_model_encoder_layers_29_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_29_layer_norm2_weight_to_fp16, x = input_355_cast_fp16)[name = string("input_357_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_29_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(484314944))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(489230208))))[name = string("encoder_text_model_encoder_layers_29_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_29_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(489271232)))]; |
| tensor<fp16, [1, 77, 5120]> linear_178_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_29_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_29_mlp_fc1_weight_to_fp16_palettized, x = input_357_cast_fp16)[name = string("linear_178_cast_fp16")]; |
| string input_361_mode_0 = const()[name = string("input_361_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_361_cast_fp16 = gelu(mode = input_361_mode_0, x = linear_178_cast_fp16)[name = string("input_361_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_29_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(489281536))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494196800))))[name = string("encoder_text_model_encoder_layers_29_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_29_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_29_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494207104)))]; |
| tensor<fp16, [1, 77, 1280]> linear_179_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_29_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_29_mlp_fc2_weight_to_fp16_palettized, x = input_361_cast_fp16)[name = string("linear_179_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_363_cast_fp16 = add(x = input_355_cast_fp16, y = linear_179_cast_fp16)[name = string("input_363_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_181_axes_0 = const()[name = string("hidden_states_181_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494209728)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494212352)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_181_cast_fp16 = layer_norm(axes = hidden_states_181_axes_0, beta = encoder_text_model_encoder_layers_30_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_30_layer_norm1_weight_to_fp16, x = input_363_cast_fp16)[name = string("hidden_states_181_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_30_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494214976))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(495443840))))[name = string("encoder_text_model_encoder_layers_30_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(495454144)))]; |
| tensor<fp16, [1, 77, 1280]> linear_180_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_30_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_30_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_181_cast_fp16)[name = string("linear_180_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_30_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(495456768))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496685632))))[name = string("encoder_text_model_encoder_layers_30_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496695936)))]; |
| tensor<fp16, [1, 77, 1280]> linear_181_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_30_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_30_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_181_cast_fp16)[name = string("linear_181_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_30_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496698560))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(497927424))))[name = string("encoder_text_model_encoder_layers_30_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(497937728)))]; |
| tensor<fp16, [1, 77, 1280]> linear_182_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_30_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_30_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_181_cast_fp16)[name = string("linear_182_cast_fp16")]; |
| tensor<int32, [4]> var_1805 = const()[name = string("op_1805"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1806_cast_fp16 = reshape(shape = var_1805, x = linear_180_cast_fp16)[name = string("op_1806_cast_fp16")]; |
| tensor<int32, [4]> var_1808 = const()[name = string("op_1808"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1809_cast_fp16 = reshape(shape = var_1808, x = linear_181_cast_fp16)[name = string("op_1809_cast_fp16")]; |
| tensor<int32, [4]> var_1811 = const()[name = string("op_1811"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1812_cast_fp16 = reshape(shape = var_1811, x = linear_182_cast_fp16)[name = string("op_1812_cast_fp16")]; |
| tensor<int32, [4]> value_61_perm_0 = const()[name = string("value_61_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_61_cast_fp16 = mul(x = var_1806_cast_fp16, y = var_11_to_fp16)[name = string("mul_61_cast_fp16")]; |
| bool matmul_30_transpose_y_0 = const()[name = string("matmul_30_transpose_y_0"), val = bool(true)]; |
| bool matmul_30_transpose_x_0 = const()[name = string("matmul_30_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_188_perm_0 = const()[name = string("transpose_188_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_189_perm_0 = const()[name = string("transpose_189_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_189 = transpose(perm = transpose_189_perm_0, x = var_1809_cast_fp16)[name = string("transpose_198")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_188 = transpose(perm = transpose_188_perm_0, x = mul_61_cast_fp16)[name = string("transpose_199")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_30_cast_fp16 = matmul(transpose_x = matmul_30_transpose_x_0, transpose_y = matmul_30_transpose_y_0, x = transpose_188, y = transpose_189)[name = string("matmul_30_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_30_cast_fp16 = add(x = matmul_30_cast_fp16, y = mul_0_to_fp16)[name = string("add_30_cast_fp16")]; |
| int32 softmax_30_axis_0 = const()[name = string("softmax_30_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_30_cast_fp16 = softmax(axis = softmax_30_axis_0, x = add_30_cast_fp16)[name = string("softmax_30_cast_fp16")]; |
| bool attn_output_121_transpose_x_0 = const()[name = string("attn_output_121_transpose_x_0"), val = bool(false)]; |
| bool attn_output_121_transpose_y_0 = const()[name = string("attn_output_121_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_61_cast_fp16 = transpose(perm = value_61_perm_0, x = var_1812_cast_fp16)[name = string("transpose_197")]; |
| tensor<fp16, [1, 20, 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 = softmax_30_cast_fp16, y = value_61_cast_fp16)[name = string("attn_output_121_cast_fp16")]; |
| tensor<int32, [4]> var_1815_perm_0 = const()[name = string("op_1815_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1817 = const()[name = string("op_1817"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1815_cast_fp16 = transpose(perm = var_1815_perm_0, x = attn_output_121_cast_fp16)[name = string("transpose_196")]; |
| tensor<fp16, [1, 77, 1280]> var_1818_cast_fp16 = reshape(shape = var_1817, x = var_1815_cast_fp16)[name = string("op_1818_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_30_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(497940352))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(499169216))))[name = string("encoder_text_model_encoder_layers_30_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(499179520)))]; |
| tensor<fp16, [1, 77, 1280]> linear_183_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_30_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_30_self_attn_out_proj_weight_to_fp16_palettized, x = var_1818_cast_fp16)[name = string("linear_183_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_367_cast_fp16 = add(x = input_363_cast_fp16, y = linear_183_cast_fp16)[name = string("input_367_cast_fp16")]; |
| tensor<int32, [1]> input_369_axes_0 = const()[name = string("input_369_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(499182144)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(499184768)))]; |
| tensor<fp16, [1, 77, 1280]> input_369_cast_fp16 = layer_norm(axes = input_369_axes_0, beta = encoder_text_model_encoder_layers_30_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_30_layer_norm2_weight_to_fp16, x = input_367_cast_fp16)[name = string("input_369_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_30_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(499187392))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(504102656))))[name = string("encoder_text_model_encoder_layers_30_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_30_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(504143680)))]; |
| tensor<fp16, [1, 77, 5120]> linear_184_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_30_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_30_mlp_fc1_weight_to_fp16_palettized, x = input_369_cast_fp16)[name = string("linear_184_cast_fp16")]; |
| string input_373_mode_0 = const()[name = string("input_373_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_373_cast_fp16 = gelu(mode = input_373_mode_0, x = linear_184_cast_fp16)[name = string("input_373_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_30_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(504153984))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(509069248))))[name = string("encoder_text_model_encoder_layers_30_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_30_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_30_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(509079552)))]; |
| tensor<fp16, [1, 77, 1280]> linear_185_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_30_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_30_mlp_fc2_weight_to_fp16_palettized, x = input_373_cast_fp16)[name = string("linear_185_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> hidden_embeds = add(x = input_367_cast_fp16, y = linear_185_cast_fp16)[name = string("input_375_cast_fp16")]; |
| tensor<int32, [1]> hidden_states_187_axes_0 = const()[name = string("hidden_states_187_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_layer_norm1_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(509082176)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_layer_norm1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(509084800)))]; |
| tensor<fp16, [1, 77, 1280]> hidden_states_187_cast_fp16 = layer_norm(axes = hidden_states_187_axes_0, beta = encoder_text_model_encoder_layers_31_layer_norm1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_31_layer_norm1_weight_to_fp16, x = hidden_embeds)[name = string("hidden_states_187_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_31_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(509087424))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(510316288))))[name = string("encoder_text_model_encoder_layers_31_self_attn_q_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_self_attn_q_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(510326592)))]; |
| tensor<fp16, [1, 77, 1280]> linear_186_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_31_self_attn_q_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_31_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_187_cast_fp16)[name = string("linear_186_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_31_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(510329216))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(511558080))))[name = string("encoder_text_model_encoder_layers_31_self_attn_k_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_self_attn_k_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(511568384)))]; |
| tensor<fp16, [1, 77, 1280]> linear_187_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_31_self_attn_k_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_31_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_187_cast_fp16)[name = string("linear_187_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_31_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(511571008))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(512799872))))[name = string("encoder_text_model_encoder_layers_31_self_attn_v_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_self_attn_v_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(512810176)))]; |
| tensor<fp16, [1, 77, 1280]> linear_188_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_31_self_attn_v_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_31_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_187_cast_fp16)[name = string("linear_188_cast_fp16")]; |
| tensor<int32, [4]> var_1861 = const()[name = string("op_1861"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1862_cast_fp16 = reshape(shape = var_1861, x = linear_186_cast_fp16)[name = string("op_1862_cast_fp16")]; |
| tensor<int32, [4]> var_1864 = const()[name = string("op_1864"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1865_cast_fp16 = reshape(shape = var_1864, x = linear_187_cast_fp16)[name = string("op_1865_cast_fp16")]; |
| tensor<int32, [4]> var_1867 = const()[name = string("op_1867"), val = tensor<int32, [4]>([1, 77, -1, 64])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1868_cast_fp16 = reshape(shape = var_1867, x = linear_188_cast_fp16)[name = string("op_1868_cast_fp16")]; |
| tensor<int32, [4]> value_perm_0 = const()[name = string("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> mul_63_cast_fp16 = mul(x = var_1862_cast_fp16, y = var_11_to_fp16)[name = string("mul_63_cast_fp16")]; |
| bool matmul_31_transpose_y_0 = const()[name = string("matmul_31_transpose_y_0"), val = bool(true)]; |
| bool matmul_31_transpose_x_0 = const()[name = string("matmul_31_transpose_x_0"), val = bool(false)]; |
| tensor<int32, [4]> transpose_190_perm_0 = const()[name = string("transpose_190_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_191_perm_0 = const()[name = string("transpose_191_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_191 = transpose(perm = transpose_191_perm_0, x = var_1865_cast_fp16)[name = string("transpose_194")]; |
| tensor<fp16, [1, 20, 77, 64]> transpose_190 = transpose(perm = transpose_190_perm_0, x = mul_63_cast_fp16)[name = string("transpose_195")]; |
| tensor<fp16, [1, 20, 77, 77]> matmul_31_cast_fp16 = matmul(transpose_x = matmul_31_transpose_x_0, transpose_y = matmul_31_transpose_y_0, x = transpose_190, y = transpose_191)[name = string("matmul_31_cast_fp16")]; |
| tensor<fp16, [1, 20, 77, 77]> add_31_cast_fp16 = add(x = matmul_31_cast_fp16, y = mul_0_to_fp16)[name = string("add_31_cast_fp16")]; |
| int32 softmax_31_axis_0 = const()[name = string("softmax_31_axis_0"), val = int32(-1)]; |
| tensor<fp16, [1, 20, 77, 77]> softmax_31_cast_fp16 = softmax(axis = softmax_31_axis_0, x = add_31_cast_fp16)[name = string("softmax_31_cast_fp16")]; |
| bool attn_output_125_transpose_x_0 = const()[name = string("attn_output_125_transpose_x_0"), val = bool(false)]; |
| bool attn_output_125_transpose_y_0 = const()[name = string("attn_output_125_transpose_y_0"), val = bool(false)]; |
| tensor<fp16, [1, 20, 77, 64]> value_cast_fp16 = transpose(perm = value_perm_0, x = var_1868_cast_fp16)[name = string("transpose_193")]; |
| tensor<fp16, [1, 20, 77, 64]> attn_output_125_cast_fp16 = matmul(transpose_x = attn_output_125_transpose_x_0, transpose_y = attn_output_125_transpose_y_0, x = softmax_31_cast_fp16, y = value_cast_fp16)[name = string("attn_output_125_cast_fp16")]; |
| tensor<int32, [4]> var_1871_perm_0 = const()[name = string("op_1871_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1873 = const()[name = string("op_1873"), val = tensor<int32, [3]>([1, 77, -1])]; |
| tensor<fp16, [1, 77, 20, 64]> var_1871_cast_fp16 = transpose(perm = var_1871_perm_0, x = attn_output_125_cast_fp16)[name = string("transpose_192")]; |
| tensor<fp16, [1, 77, 1280]> var_1874_cast_fp16 = reshape(shape = var_1873, x = var_1871_cast_fp16)[name = string("op_1874_cast_fp16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_model_encoder_layers_31_self_attn_out_proj_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(512812800))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(514041664))))[name = string("encoder_text_model_encoder_layers_31_self_attn_out_proj_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_self_attn_out_proj_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(514051968)))]; |
| tensor<fp16, [1, 77, 1280]> linear_189_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_31_self_attn_out_proj_bias_to_fp16, weight = encoder_text_model_encoder_layers_31_self_attn_out_proj_weight_to_fp16_palettized, x = var_1874_cast_fp16)[name = string("linear_189_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_379_cast_fp16 = add(x = hidden_embeds, y = linear_189_cast_fp16)[name = string("input_379_cast_fp16")]; |
| tensor<int32, [1]> input_381_axes_0 = const()[name = string("input_381_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_layer_norm2_weight_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(514054592)))]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_layer_norm2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(514057216)))]; |
| tensor<fp16, [1, 77, 1280]> input_381_cast_fp16 = layer_norm(axes = input_381_axes_0, beta = encoder_text_model_encoder_layers_31_layer_norm2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_encoder_layers_31_layer_norm2_weight_to_fp16, x = input_379_cast_fp16)[name = string("input_381_cast_fp16")]; |
| tensor<fp16, [5120, 1280]> encoder_text_model_encoder_layers_31_mlp_fc1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(514059840))), lut = tensor<fp16, [320, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518975104))))[name = string("encoder_text_model_encoder_layers_31_mlp_fc1_weight_to_fp16_palettized")]; |
| tensor<fp16, [5120]> encoder_text_model_encoder_layers_31_mlp_fc1_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(519016128)))]; |
| tensor<fp16, [1, 77, 5120]> linear_190_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_31_mlp_fc1_bias_to_fp16, weight = encoder_text_model_encoder_layers_31_mlp_fc1_weight_to_fp16_palettized, x = input_381_cast_fp16)[name = string("linear_190_cast_fp16")]; |
| string input_385_mode_0 = const()[name = string("input_385_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 77, 5120]> input_385_cast_fp16 = gelu(mode = input_385_mode_0, x = linear_190_cast_fp16)[name = string("input_385_cast_fp16")]; |
| tensor<fp16, [1280, 5120]> encoder_text_model_encoder_layers_31_mlp_fc2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(519026432))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(523941696))))[name = string("encoder_text_model_encoder_layers_31_mlp_fc2_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> encoder_text_model_encoder_layers_31_mlp_fc2_bias_to_fp16 = const()[name = string("encoder_text_model_encoder_layers_31_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(523952000)))]; |
| tensor<fp16, [1, 77, 1280]> linear_191_cast_fp16 = linear(bias = encoder_text_model_encoder_layers_31_mlp_fc2_bias_to_fp16, weight = encoder_text_model_encoder_layers_31_mlp_fc2_weight_to_fp16_palettized, x = input_385_cast_fp16)[name = string("linear_191_cast_fp16")]; |
| tensor<fp16, [1, 77, 1280]> input_387_cast_fp16 = add(x = input_379_cast_fp16, y = linear_191_cast_fp16)[name = string("input_387_cast_fp16")]; |
| tensor<int32, [1]> last_hidden_state_axes_0 = const()[name = string("last_hidden_state_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> encoder_text_model_final_layer_norm_weight_to_fp16 = const()[name = string("encoder_text_model_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(523954624)))]; |
| tensor<fp16, [1280]> encoder_text_model_final_layer_norm_bias_to_fp16 = const()[name = string("encoder_text_model_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(523957248)))]; |
| tensor<fp16, [1, 77, 1280]> last_hidden_state_cast_fp16 = layer_norm(axes = last_hidden_state_axes_0, beta = encoder_text_model_final_layer_norm_bias_to_fp16, epsilon = var_9_to_fp16, gamma = encoder_text_model_final_layer_norm_weight_to_fp16, x = input_387_cast_fp16)[name = string("last_hidden_state_cast_fp16")]; |
| tensor<int32, [1]> var_1902 = const()[name = string("op_1902"), val = tensor<int32, [1]>([0])]; |
| int32 var_1904_axis_0 = const()[name = string("op_1904_axis_0"), val = int32(-1)]; |
| bool var_1904_keep_dims_0 = const()[name = string("op_1904_keep_dims_0"), val = bool(false)]; |
| string var_1904_output_dtype_0 = const()[name = string("op_1904_output_dtype_0"), val = string("int32")]; |
| tensor<int32, [1]> var_1904 = reduce_argmax(axis = var_1904_axis_0, keep_dims = var_1904_keep_dims_0, output_dtype = var_1904_output_dtype_0, x = input_ids)[name = string("op_1904")]; |
| int32 stack_0_axis_0 = const()[name = string("stack_0_axis_0"), val = int32(1)]; |
| tensor<int32, [1, 2]> stack_0 = stack(axis = stack_0_axis_0, values = (var_1902, var_1904))[name = string("stack_0")]; |
| int32 input_transpose_batch_dims_0 = const()[name = string("input_transpose_batch_dims_0"), val = int32(0)]; |
| bool input_transpose_validate_indices_0 = const()[name = string("input_transpose_validate_indices_0"), val = bool(false)]; |
| string stack_0_to_uint16_dtype_0 = const()[name = string("stack_0_to_uint16_dtype_0"), val = string("uint16")]; |
| tensor<uint16, [1, 2]> stack_0_to_uint16 = cast(dtype = stack_0_to_uint16_dtype_0, x = stack_0)[name = string("cast_0")]; |
| tensor<fp16, [1, 1280]> input_transpose_cast_fp16_cast_uint16 = gather_nd(batch_dims = input_transpose_batch_dims_0, indices = stack_0_to_uint16, validate_indices = input_transpose_validate_indices_0, x = last_hidden_state_cast_fp16)[name = string("input_transpose_cast_fp16_cast_uint16")]; |
| tensor<fp16, [1280, 1280]> encoder_text_projection_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(523959872))), lut = tensor<fp16, [80, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(525188736))))[name = string("encoder_text_projection_weight_to_fp16_palettized")]; |
| tensor<fp16, [1280]> linear_192_bias_0_to_fp16 = const()[name = string("linear_192_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(525199040)))]; |
| tensor<fp16, [1, 1280]> pooled_outputs = linear(bias = linear_192_bias_0_to_fp16, weight = encoder_text_projection_weight_to_fp16_palettized, x = input_transpose_cast_fp16_cast_uint16)[name = string("linear_192_cast_fp16")]; |
| } -> (hidden_embeds, pooled_outputs); |
| } |
