Open RyusukeYamano opened 4 years ago
It is not able to embed yolov3-tiny to ultra96 due to that model require large RAM at slice operator near by yolo layer.
Verbose log when executing example/yolov3-tiny.
<class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_0 input_rams [(16, 1112), (16, 1112), (16, 1112), (16, 1112), (16, 1112), (16, 1112), (16, 1112), (16, 1112), (16, 1112), (16, 16), (16, 16), (16, 16), (16, 16), (16, 16), (16, 16), (16, 16), (16, 16), (16, 16), (32, 16), (16, 16)] output_rams [(16, 1664)] temp_rams [] <class 'nngen.operator.pool_serial.max_pool_serial'> ====================================================== MaxPool_3 input_rams [(16, 26624)] output_rams [(16, 6656)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_4 input_rams [(16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 64), (16, 64), (16, 64), (16, 64), (16, 64), (16, 64), (16, 64), (16, 64), (16, 64), (32, 32), (16, 32)] output_rams [(16, 832)] temp_rams [] <class 'nngen.operator.pool_serial.max_pool_serial'> ====================================================== MaxPool_7 input_rams [(16, 26624)] output_rams [(16, 6656)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_8 input_rams [(16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 2240), (16, 128), (16, 128), (16, 128), (16, 128), (16, 128), (16, 128), (16, 128), (16, 128), (16, 128), (32, 64), (16, 64)] output_rams [(16, 416)] temp_rams [] <class 'nngen.operator.pool_serial.max_pool_serial'> ====================================================== MaxPool_11 input_rams [(16, 26624)] output_rams [(16, 6656)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_12 input_rams [(16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 256), (16, 256), (16, 256), (16, 256), (16, 256), (16, 256), (16, 256), (16, 256), (16, 256), (32, 128), (16, 128)] output_rams [(16, 208)] temp_rams [] <class 'nngen.operator.pool_serial.max_pool_serial'> ====================================================== MaxPool_15 input_rams [(16, 26624)] output_rams [(16, 6656)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_16 input_rams [(16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 2304), (16, 512), (16, 512), (16, 512), (16, 512), (16, 512), (16, 512), (16, 512), (16, 512), (16, 512), (32, 256), (16, 256)] output_rams [(16, 104)] temp_rams [] <class 'nngen.operator.pool_serial.max_pool_serial'> ====================================================== MaxPool_19 input_rams [(16, 26624)] output_rams [(16, 6656)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_20 input_rams [(16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (32, 512), (16, 512)] output_rams [(16, 52)] temp_rams [] <class 'nngen.operator.pad.pad'> ====================================================== Pad_23 input_rams [(16, 13312)] output_rams [(16, 14336)] temp_rams [] <class 'nngen.operator.pool.max_pool'> ====================================================== MaxPool_24 input_rams [(16, 7168), (16, 7168), (16, 7168), (16, 7168)] output_rams [(16, 13312)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_25 input_rams [(16, 5120), (16, 5120), (16, 5120), (16, 5120), (16, 5120), (16, 5120), (16, 5120), (16, 5120), (16, 5120), (16, 2048), (16, 2048), (16, 2048), (16, 2048), (16, 2048), (16, 2048), (16, 2048), (16, 2048), (16, 2048), (32, 1024), (16, 1024)] output_rams [(16, 52)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_28 input_rams [(16, 26624), (16, 4096), (32, 256), (16, 256)] output_rams [(16, 52)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_55 input_rams [(16, 6656), (16, 1024), (32, 128), (16, 128)] output_rams [(16, 52)] temp_rams [] <class 'nngen.operator.upsampling2d.upsampling2d'> ====================================================== Upsample_59 input_rams [(16, 256)] output_rams [(16, 256)] temp_rams [] <class 'nngen.operator.normalize.scaled_concat'> ====================================================== Concat_60 input_rams [(16, 256)] output_rams [(16, 384)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_61 input_rams [(16, 6912), (16, 6912), (16, 6912), (16, 6912), (16, 6912), (16, 6912), (16, 6912), (16, 6912), (16, 6912), (16, 1536), (16, 1536), (16, 1536), (16, 1536), (16, 1536), (16, 1536), (16, 1536), (16, 1536), (16, 1536), (32, 256), (16, 256)] output_rams [(16, 104)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_64 input_rams [(16, 13312), (16, 1024), (32, 255), (16, 2)] output_rams [(16, 104)] temp_rams [] <class 'nngen.operator.basic._lazy_reshape'> ====================================================== Reshape_66 input_rams [(16, 255)] output_rams [(16, 255)] temp_rams [] <class 'nngen.operator.basic.transpose'> ====================================================== Transpose_67 input_rams [(16, 85)] output_rams [(16, 85)] temp_rams [] <class 'nngen.operator.basic._lazy_reshape'> ====================================================== Reshape_69 input_rams () output_rams () temp_rams () <class 'nngen.operator.slice_.slice_'> ====================================================== None act_shape [2028, 86] self.begins (0, 4) self.ends (2028, 5) self.strides (1, 1) 348816 348816 input_rams [(16, 348816)] output_rams [(16, 8112)] temp_rams [] <class 'nngen.operator.slice_.slice_'> ====================================================== None act_shape [2028, 86] self.begins (0, 5) self.ends (2028, 85) self.strides (1, 1) 348816 348816 input_rams [(16, 348816)] output_rams [(16, 324480)] temp_rams [] <class 'nngen.operator.basic.multiply'> ====================================================== Mul_82 input_rams [(16, 160), (16, 4)] output_rams [(16, 160)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_31 input_rams [(16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 2560), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (16, 1024), (32, 512), (16, 512)] output_rams [(16, 52)] temp_rams [] <class 'nngen.operator.conv2d.conv2d'> ====================================================== Conv_34 input_rams [(16, 13312), (16, 2048), (32, 255), (16, 2)] output_rams [(16, 52)] temp_rams [] <class 'nngen.operator.basic._lazy_reshape'> ====================================================== Reshape_36 input_rams [(16, 255)] output_rams [(16, 255)] temp_rams [] <class 'nngen.operator.basic.transpose'> ====================================================== Transpose_37 input_rams [(16, 85)] output_rams [(16, 85)] temp_rams [] <class 'nngen.operator.basic._lazy_reshape'> ====================================================== Reshape_39 input_rams () output_rams () temp_rams () <class 'nngen.operator.slice_.slice_'> ====================================================== None act_shape [507, 86] self.begins (0, 4) self.ends (507, 5) self.strides (1, 1) 87204 87204 input_rams [(16, 87204)] output_rams [(16, 2028)] temp_rams [] <class 'nngen.operator.slice_.slice_'> ====================================================== None act_shape [507, 86] self.begins (0, 5) self.ends (507, 85) self.strides (1, 1) 87204 87204 input_rams [(16, 87204)] output_rams [(16, 81120)] temp_rams [] <class 'nngen.operator.basic.multiply'> ====================================================== Mul_52 input_rams [(16, 160), (16, 4)] output_rams [(16, 160)] temp_rams [] <class 'nngen.operator.normalize.scaled_concat'> ====================================================== Concat_85 input_rams [(16, 80)] output_rams [(16, 80)] temp_rams [] act_shape [2028, 86] self.begins (0, 5) self.ends (2028, 85) self.strides (1, 1) 348816 348816 act_shape [507, 86] self.begins (0, 4) self.ends (507, 5) self.strides (1, 1) 87204 87204 act_shape [507, 86] self.begins (0, 5) self.ends (507, 85) self.strides (1, 1) 87204 87204 NNgen: Neural Network Accelerator Generator (version 1.2.0) [IP-XACT] Output: yolov3tiny [Configuration] (AXI Master Interface) Data width : 32 Address width: 32 (AXI Slave Interface) Data width : 32 Address width: 32 [Schedule Table] (Stage 0) (Stage 1) <conv2d Conv_0 dtype:int16 shape:(1, 416, 416, 16) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(16,) scale:(16,) cshamt_out:13 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:2 stationary:filter default_addr:19502656 g_index:0 l_index:1 word_alignment:2 aligned_shape:(1, 416, 416, 16) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:434.483748> | <placeholder act dtype:int16 shape:(1, 416, 416, 3) default_addr:405632 g_index:2 word_alignment:2 aligned_shape:(1, 416, 416, 4) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:128.000000> | <variable module_list.0.Conv2d.weight dtype:int16 shape:(16, 3, 3, 3) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(16, 3, 3, 4) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:13903.479945> | <variable onnx_Conv_0_conv.bias dtype:int32 shape:(16,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(16,) scale_factor:1779645.432922> | <variable onnx_Conv_0_conv.scale dtype:int16 shape:(16,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(16,) scale_factor:2.000000> (Stage 2) <max_pool_serial MaxPool_3 dtype:int16 shape:(1, 208, 208, 16) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:25040448 g_index:0 l_index:2 word_alignment:2 aligned_shape:(1, 208, 208, 16) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:434.483748> | <conv2d Conv_0 dtype:int16 shape:(1, 416, 416, 16) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(16,) scale:(16,) cshamt_out:13 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:2 stationary:filter default_addr:19502656 g_index:0 l_index:1 word_alignment:2 aligned_shape:(1, 416, 416, 16) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:434.483748> (Stage 3) <conv2d Conv_4 dtype:int16 shape:(1, 208, 208, 32) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(32,) scale:(32,) cshamt_out:19 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter default_addr:26424896 g_index:0 l_index:3 word_alignment:2 aligned_shape:(1, 208, 208, 32) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:845.394497> | <max_pool_serial MaxPool_3 dtype:int16 shape:(1, 208, 208, 16) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:25040448 g_index:0 l_index:2 word_alignment:2 aligned_shape:(1, 208, 208, 16) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:434.483748> | <variable module_list.2.Conv2d.weight dtype:int16 shape:(32, 3, 3, 16) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(32, 3, 3, 16) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:15939.541460> | <variable onnx_Conv_4_conv.bias dtype:int32 shape:(32,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(32,) scale_factor:6925471.719068> | <variable onnx_Conv_4_conv.scale dtype:int16 shape:(32,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(32,) scale_factor:64.000000> (Stage 4) <max_pool_serial MaxPool_7 dtype:int16 shape:(1, 104, 104, 32) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:29193792 g_index:0 l_index:4 word_alignment:2 aligned_shape:(1, 104, 104, 32) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:845.394497> | <conv2d Conv_4 dtype:int16 shape:(1, 208, 208, 32) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(32,) scale:(32,) cshamt_out:19 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter default_addr:26424896 g_index:0 l_index:3 word_alignment:2 aligned_shape:(1, 208, 208, 32) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:845.394497> (Stage 5) <conv2d Conv_8 dtype:int16 shape:(1, 104, 104, 64) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(64,) scale:(64,) cshamt_out:19 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:8 stationary:filter default_addr:29886016 g_index:0 l_index:5 word_alignment:2 aligned_shape:(1, 104, 104, 64) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1575.078006> | <max_pool_serial MaxPool_7 dtype:int16 shape:(1, 104, 104, 32) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:29193792 g_index:0 l_index:4 word_alignment:2 aligned_shape:(1, 104, 104, 32) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:845.394497> | <variable module_list.4.Conv2d.weight dtype:int16 shape:(64, 3, 3, 32) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(64, 3, 3, 32) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:15262.743102> | <variable onnx_Conv_8_conv.bias dtype:int32 shape:(64,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(64,) scale_factor:12903039.026557> | <variable onnx_Conv_8_conv.scale dtype:int16 shape:(64,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(64,) scale_factor:64.000000> (Stage 6) <max_pool_serial MaxPool_11 dtype:int16 shape:(1, 52, 52, 64) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:31270464 g_index:0 l_index:6 word_alignment:2 aligned_shape:(1, 52, 52, 64) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1575.078006> | <conv2d Conv_8 dtype:int16 shape:(1, 104, 104, 64) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(64,) scale:(64,) cshamt_out:19 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:8 stationary:filter default_addr:29886016 g_index:0 l_index:5 word_alignment:2 aligned_shape:(1, 104, 104, 64) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1575.078006> (Stage 7) <conv2d Conv_12 dtype:int16 shape:(1, 52, 52, 128) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(128,) scale:(128,) cshamt_out:21 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:16 stationary:filter default_addr:31616576 g_index:0 l_index:7 word_alignment:2 aligned_shape:(1, 52, 52, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1093.161615> | <max_pool_serial MaxPool_11 dtype:int16 shape:(1, 52, 52, 64) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:31270464 g_index:0 l_index:6 word_alignment:2 aligned_shape:(1, 52, 52, 64) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1575.078006> | <variable module_list.6.Conv2d.weight dtype:int16 shape:(128, 3, 3, 64) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(128, 3, 3, 64) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:22742.187791> | <variable onnx_Conv_12_conv.bias dtype:int32 shape:(128,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(128,) scale_factor:35820719.801639> | <variable onnx_Conv_12_conv.scale dtype:int16 shape:(128,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(128,) scale_factor:64.000000> (Stage 8) <max_pool_serial MaxPool_15 dtype:int16 shape:(1, 26, 26, 128) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:32308800 g_index:0 l_index:8 word_alignment:2 aligned_shape:(1, 26, 26, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1093.161615> | <conv2d Conv_12 dtype:int16 shape:(1, 52, 52, 128) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(128,) scale:(128,) cshamt_out:21 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:16 stationary:filter default_addr:31616576 g_index:0 l_index:7 word_alignment:2 aligned_shape:(1, 52, 52, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1093.161615> (Stage 9) <conv2d Conv_16 dtype:int16 shape:(1, 26, 26, 256) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(256,) scale:(256,) cshamt_out:19 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:8 stationary:filter default_addr:32481856 g_index:0 l_index:9 word_alignment:2 aligned_shape:(1, 26, 26, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:2745.399286> | <max_pool_serial MaxPool_15 dtype:int16 shape:(1, 26, 26, 128) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:32308800 g_index:0 l_index:8 word_alignment:2 aligned_shape:(1, 26, 26, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1093.161615> | <variable module_list.8.Conv2d.weight dtype:int16 shape:(256, 3, 3, 128) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(256, 3, 3, 128) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:20573.637640> | <variable onnx_Conv_16_conv.bias dtype:int32 shape:(256,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(256,) scale_factor:22490310.949332> | <variable onnx_Conv_16_conv.scale dtype:int16 shape:(256,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(256,) scale_factor:64.000000> (Stage 10) <max_pool_serial MaxPool_19 dtype:int16 shape:(1, 13, 13, 256) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:32827968 g_index:0 l_index:10 word_alignment:2 aligned_shape:(1, 13, 13, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:2745.399286> | <conv2d Conv_16 dtype:int16 shape:(1, 26, 26, 256) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(256,) scale:(256,) cshamt_out:19 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:8 stationary:filter default_addr:32481856 g_index:0 l_index:9 word_alignment:2 aligned_shape:(1, 26, 26, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:2745.399286> (Stage 11) <conv2d Conv_20 dtype:int16 shape:(1, 13, 13, 512) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(512,) scale:(512,) cshamt_out:20 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter keep_input default_addr:32914496 g_index:0 l_index:11 word_alignment:2 aligned_shape:(1, 13, 13, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7199.522558> | <max_pool_serial MaxPool_19 dtype:int16 shape:(1, 13, 13, 256) ksize:(1, 2, 2, 1) strides:(1, 2, 2, 1) padding:(0, 0, 0, 0) no_reuse default_addr:32827968 g_index:0 l_index:10 word_alignment:2 aligned_shape:(1, 13, 13, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:2745.399286> | <variable module_list.10.Conv2d.weight dtype:int16 shape:(512, 3, 3, 256) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(512, 3, 3, 256) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:42965.326831> | <variable onnx_Conv_20_conv.bias dtype:int32 shape:(512,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(512,) scale_factor:117956977.595654> | <variable onnx_Conv_20_conv.scale dtype:int16 shape:(512,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(512,) scale_factor:64.000000> (Stage 12) <pad Pad_23 dtype:int16 shape:(1, 14, 14, 512) padding:(0, 1, 0, 1) default_addr:33087552 g_index:0 l_index:12 word_alignment:2 aligned_shape:(1, 14, 14, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7199.522558> | <conv2d Conv_20 dtype:int16 shape:(1, 13, 13, 512) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(512,) scale:(512,) cshamt_out:20 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter keep_input default_addr:32914496 g_index:0 l_index:11 word_alignment:2 aligned_shape:(1, 13, 13, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7199.522558> (Stage 13) <max_pool MaxPool_24 dtype:int16 shape:(1, 13, 13, 512) ksize:(1, 2, 2, 1) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) default_addr:33288256 g_index:0 l_index:13 word_alignment:2 aligned_shape:(1, 13, 13, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7199.522558> | <pad Pad_23 dtype:int16 shape:(1, 14, 14, 512) padding:(0, 1, 0, 1) default_addr:33087552 g_index:0 l_index:12 word_alignment:2 aligned_shape:(1, 14, 14, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7199.522558> (Stage 14) <conv2d Conv_25 dtype:int16 shape:(1, 13, 13, 1024) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(1024,) scale:(1024,) cshamt_out:22 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:2 stationary:filter default_addr:33461312 g_index:0 l_index:14 word_alignment:2 aligned_shape:(1, 13, 13, 1024) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1523.915455> | <max_pool MaxPool_24 dtype:int16 shape:(1, 13, 13, 512) ksize:(1, 2, 2, 1) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) default_addr:33288256 g_index:0 l_index:13 word_alignment:2 aligned_shape:(1, 13, 13, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7199.522558> | <variable module_list.12.Conv2d.weight dtype:int16 shape:(1024, 3, 3, 512) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(1024, 3, 3, 512) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:55487.747957> | <variable onnx_Conv_25_conv.bias dtype:int32 shape:(1024,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(1024,) scale_factor:399485293.128401> | <variable onnx_Conv_25_conv.scale dtype:int16 shape:(1024,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(1024,) scale_factor:16.000000> (Stage 15) <conv2d Conv_28 dtype:int16 shape:(1, 13, 13, 256) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(256,) scale:(256,) cshamt_out:20 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter keep_input default_addr:33807424 g_index:0 l_index:15 word_alignment:2 aligned_shape:(1, 13, 13, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7965.936386> | <conv2d Conv_25 dtype:int16 shape:(1, 13, 13, 1024) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(1024,) scale:(1024,) cshamt_out:22 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:2 stationary:filter default_addr:33461312 g_index:0 l_index:14 word_alignment:2 aligned_shape:(1, 13, 13, 1024) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1523.915455> | <variable module_list.13.Conv2d.weight dtype:int16 shape:(256, 1, 1, 1024) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(256, 1, 1, 1024) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:42821.897134> | <variable onnx_Conv_28_conv.bias dtype:int32 shape:(256,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(256,) scale_factor:65256950.870254> | <variable onnx_Conv_28_conv.scale dtype:int16 shape:(256,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(256,) scale_factor:128.000000> (Stage 16) <conv2d Conv_55 dtype:int16 shape:(1, 13, 13, 128) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(128,) scale:(128,) cshamt_out:20 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:16 stationary:filter keep_input default_addr:34411200 g_index:0 l_index:22 word_alignment:2 aligned_shape:(1, 13, 13, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:6841.001736> | <conv2d Conv_28 dtype:int16 shape:(1, 13, 13, 256) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(256,) scale:(256,) cshamt_out:20 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter keep_input default_addr:33807424 g_index:0 l_index:15 word_alignment:2 aligned_shape:(1, 13, 13, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7965.936386> | <variable module_list.18.Conv2d.weight dtype:int16 shape:(128, 1, 1, 256) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(128, 1, 1, 256) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:56281.128556> | <variable onnx_Conv_55_conv.bias dtype:int32 shape:(128,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(128,) scale_factor:448331889.781508> | <variable onnx_Conv_55_conv.scale dtype:int16 shape:(128,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(128,) scale_factor:16.000000> (Stage 17) <upsampling2d Upsample_59 dtype:int16 shape:(1, 26, 26, 128) default_addr:34454464 g_index:0 l_index:23 word_alignment:2 aligned_shape:(1, 26, 26, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:6841.001736> | <conv2d Conv_55 dtype:int16 shape:(1, 13, 13, 128) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(128,) scale:(128,) cshamt_out:20 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:16 stationary:filter keep_input default_addr:34411200 g_index:0 l_index:22 word_alignment:2 aligned_shape:(1, 13, 13, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:6841.001736> (Stage 18) <scaled_concat Concat_60 dtype:int16 shape:(1, 26, 26, 384) buffered scales:(128, 319) shamt:8 default_addr:34627520 g_index:0 l_index:24 word_alignment:2 aligned_shape:(1, 26, 26, 384) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:3420.500868> | <upsampling2d Upsample_59 dtype:int16 shape:(1, 26, 26, 128) default_addr:34454464 g_index:0 l_index:23 word_alignment:2 aligned_shape:(1, 26, 26, 128) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:6841.001736> | <conv2d Conv_16 dtype:int16 shape:(1, 26, 26, 256) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(256,) scale:(256,) cshamt_out:19 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:8 stationary:filter default_addr:32481856 g_index:0 l_index:9 word_alignment:2 aligned_shape:(1, 26, 26, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:2745.399286> (Stage 19) <conv2d Conv_61 dtype:int16 shape:(1, 26, 26, 256) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(256,) scale:(256,) cshamt_out:22 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:2 stationary:filter default_addr:35146688 g_index:0 l_index:25 word_alignment:2 aligned_shape:(1, 26, 26, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:4418.436110> | <scaled_concat Concat_60 dtype:int16 shape:(1, 26, 26, 384) buffered scales:(128, 319) shamt:8 default_addr:34627520 g_index:0 l_index:24 word_alignment:2 aligned_shape:(1, 26, 26, 384) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:3420.500868> | <variable module_list.21.Conv2d.weight dtype:int16 shape:(256, 3, 3, 384) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(256, 3, 3, 384) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:84656.206817> | <variable onnx_Conv_61_conv.bias dtype:int32 shape:(256,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(256,) scale_factor:289566628.905870> | <variable onnx_Conv_61_conv.scale dtype:int16 shape:(256,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(256,) scale_factor:64.000000> (Stage 20) <conv2d Conv_64 dtype:int16 shape:(1, 26, 26, 255) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(255,) scale:(1,) cshamt_out:17 sum_dtype:int64 concur_och:16 stationary:filter keep_input default_addr:35492800 g_index:0 l_index:26 word_alignment:2 aligned_shape:(1, 26, 26, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1457.059043> | <conv2d Conv_61 dtype:int16 shape:(1, 26, 26, 256) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(256,) scale:(256,) cshamt_out:22 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:2 stationary:filter default_addr:35146688 g_index:0 l_index:25 word_alignment:2 aligned_shape:(1, 26, 26, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:4418.436110> | <variable module_list.22.Conv2d.weight dtype:int16 shape:(255, 1, 1, 256) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(255, 1, 1, 256) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:43223.357344> | <variable module_list.22.Conv2d.bias dtype:int32 shape:(255,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(255,) scale_factor:190979642.883883> | <variable onnx_Conv_64_conv.scale dtype:int16 shape:(1,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(2,) scale_factor:1.000000> (Stage 21) <_lazy_reshape Reshape_66 dtype:int16 shape:(1, 26, 26, 3, 85) default_addr:35838912 g_index:0 l_index:27 word_alignment:2 aligned_shape:(1, 26, 26, 3, 86) layout:('N', 'H', 'W', 'X0', 'X1') onnx_layout:('N', 'X0', 'X1', 'H', 'W') scale_factor:1457.059043> | <conv2d Conv_64 dtype:int16 shape:(1, 26, 26, 255) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(255,) scale:(1,) cshamt_out:17 sum_dtype:int64 concur_och:16 stationary:filter keep_input default_addr:35492800 g_index:0 l_index:26 word_alignment:2 aligned_shape:(1, 26, 26, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1457.059043> (Stage 22) <transpose Transpose_67 dtype:int16 shape:(1, 3, 26, 26, 85) perm:(0, 3, 1, 2, 4) onnx_perm:(0, 1, 3, 4, 2) default_addr:36187776 g_index:0 l_index:28 word_alignment:2 aligned_shape:(1, 3, 26, 26, 86) layout:('N', 'X0', 'H', 'W', 'X1') onnx_layout:('N', 'X0', 'H', 'W', 'X1') scale_factor:1457.059043> | <_lazy_reshape Reshape_66 dtype:int16 shape:(1, 26, 26, 3, 85) default_addr:35838912 g_index:0 l_index:27 word_alignment:2 aligned_shape:(1, 26, 26, 3, 86) layout:('N', 'H', 'W', 'X0', 'X1') onnx_layout:('N', 'X0', 'X1', 'H', 'W') scale_factor:1457.059043> (Stage 23) <_lazy_reshape Reshape_69 dtype:int16 shape:(2028, 85) alias_of:Transpose_67 default_addr:36187776 g_index:0 l_index:28 word_alignment:2 aligned_shape:(2028, 86) onnx_layout:('X0', 'X1') scale_factor:1457.059043> | <transpose Transpose_67 dtype:int16 shape:(1, 3, 26, 26, 85) perm:(0, 3, 1, 2, 4) onnx_perm:(0, 1, 3, 4, 2) default_addr:36187776 g_index:0 l_index:28 word_alignment:2 aligned_shape:(1, 3, 26, 26, 86) layout:('N', 'X0', 'H', 'W', 'X1') onnx_layout:('N', 'X0', 'H', 'W', 'X1') scale_factor:1457.059043> (Stage 24) <slice_ None dtype:int16 shape:(2028, 1) begins:(0, 4) ends:(2028, 5) strides:(1, 1) default_addr:36861120 g_index:0 l_index:30 word_alignment:2 aligned_shape:(2028, 2) onnx_layout:('X0', 'X1') scale_factor:1457.059043> | <_lazy_reshape Reshape_69 dtype:int16 shape:(2028, 85) alias_of:Transpose_67 default_addr:36187776 g_index:0 l_index:28 word_alignment:2 aligned_shape:(2028, 86) onnx_layout:('X0', 'X1') scale_factor:1457.059043> (Stage 25) <slice_ None dtype:int16 shape:(2028, 80) begins:(0, 5) ends:(2028, 85) strides:(1, 1) default_addr:36536640 g_index:0 l_index:29 word_alignment:2 aligned_shape:(2028, 80) onnx_layout:('X0', 'X1') scale_factor:1457.059043> | <_lazy_reshape Reshape_69 dtype:int16 shape:(2028, 85) alias_of:Transpose_67 default_addr:36187776 g_index:0 l_index:28 word_alignment:2 aligned_shape:(2028, 86) onnx_layout:('X0', 'X1') scale_factor:1457.059043> (Stage 26) <multiply Mul_82 dtype:int16 shape:(2028, 80) default_addr:36950400 g_index:0 l_index:32 word_alignment:2 aligned_shape:(2028, 80) onnx_layout:('X0', 'X1') scale_factor:31130.000000> | <sigmoid Sigmoid_79 dtype:int16 shape:(2028, 80) chained default_addr:0 word_alignment:2 aligned_shape:(2028, 80) onnx_layout:('X0', 'X1') scale_factor:31130.000000> | | <slice_ None dtype:int16 shape:(2028, 80) begins:(0, 5) ends:(2028, 85) strides:(1, 1) default_addr:36536640 g_index:0 l_index:29 word_alignment:2 aligned_shape:(2028, 80) onnx_layout:('X0', 'X1') scale_factor:1457.059043> | <sigmoid Sigmoid_81 dtype:int16 shape:(2028, 1) chained default_addr:0 word_alignment:2 aligned_shape:(2028, 2) onnx_layout:('X0', 'X1') scale_factor:31130.000000> | | <slice_ None dtype:int16 shape:(2028, 1) begins:(0, 4) ends:(2028, 5) strides:(1, 1) default_addr:36861120 g_index:0 l_index:30 word_alignment:2 aligned_shape:(2028, 2) onnx_layout:('X0', 'X1') scale_factor:1457.059043> (Stage 27) <conv2d Conv_31 dtype:int16 shape:(1, 13, 13, 512) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(512,) scale:(512,) cshamt_out:21 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter keep_input default_addr:33893952 g_index:0 l_index:16 word_alignment:2 aligned_shape:(1, 13, 13, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:4674.111764> | <conv2d Conv_28 dtype:int16 shape:(1, 13, 13, 256) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(256,) scale:(256,) cshamt_out:20 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter keep_input default_addr:33807424 g_index:0 l_index:15 word_alignment:2 aligned_shape:(1, 13, 13, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:7965.936386> | <variable module_list.14.Conv2d.weight dtype:int16 shape:(512, 3, 3, 256) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(512, 3, 3, 256) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:76908.118207> | <variable onnx_Conv_31_conv.bias dtype:int32 shape:(512,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(512,) scale_factor:612645177.166219> | <variable onnx_Conv_31_conv.scale dtype:int16 shape:(512,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(512,) scale_factor:16.000000> (Stage 28) <conv2d Conv_34 dtype:int16 shape:(1, 13, 13, 255) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(255,) scale:(1,) cshamt_out:17 sum_dtype:int64 concur_och:8 stationary:filter keep_input default_addr:34067008 g_index:0 l_index:17 word_alignment:2 aligned_shape:(1, 13, 13, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1117.906799> | <conv2d Conv_31 dtype:int16 shape:(1, 13, 13, 512) strides:(1, 1, 1, 1) padding:(1, 1, 1, 1) bias:(512,) scale:(512,) cshamt_out:21 act_func:leaky_relu_214748368_31 sum_dtype:int64 concur_och:4 stationary:filter keep_input default_addr:33893952 g_index:0 l_index:16 word_alignment:2 aligned_shape:(1, 13, 13, 512) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:4674.111764> | <variable module_list.15.Conv2d.weight dtype:int16 shape:(255, 1, 1, 512) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(255, 1, 1, 512) layout:('O', 'H', 'W', 'I') onnx_layout:('O', 'I', 'H', 'W') scale_factor:31348.475905> | <variable module_list.15.Conv2d.bias dtype:int32 shape:(255,) default_addr:1790080 g_index:3 word_alignment:1 aligned_shape:(255,) scale_factor:146526280.017658> | <variable onnx_Conv_34_conv.scale dtype:int16 shape:(1,) default_addr:1790080 g_index:3 word_alignment:2 aligned_shape:(2,) scale_factor:1.000000> (Stage 29) <_lazy_reshape Reshape_36 dtype:int16 shape:(1, 13, 13, 3, 85) default_addr:34153536 g_index:0 l_index:18 word_alignment:2 aligned_shape:(1, 13, 13, 3, 86) layout:('N', 'H', 'W', 'X0', 'X1') onnx_layout:('N', 'X0', 'X1', 'H', 'W') scale_factor:1117.906799> | <conv2d Conv_34 dtype:int16 shape:(1, 13, 13, 255) strides:(1, 1, 1, 1) padding:(0, 0, 0, 0) bias:(255,) scale:(1,) cshamt_out:17 sum_dtype:int64 concur_och:8 stationary:filter keep_input default_addr:34067008 g_index:0 l_index:17 word_alignment:2 aligned_shape:(1, 13, 13, 256) layout:('N', 'H', 'W', 'C') onnx_layout:('N', 'C', 'H', 'W') scale_factor:1117.906799> (Stage 30) <transpose Transpose_37 dtype:int16 shape:(1, 3, 13, 13, 85) perm:(0, 3, 1, 2, 4) onnx_perm:(0, 1, 3, 4, 2) default_addr:34240768 g_index:0 l_index:19 word_alignment:2 aligned_shape:(1, 3, 13, 13, 86) layout:('N', 'X0', 'H', 'W', 'X1') onnx_layout:('N', 'X0', 'H', 'W', 'X1') scale_factor:1117.906799> | <_lazy_reshape Reshape_36 dtype:int16 shape:(1, 13, 13, 3, 85) default_addr:34153536 g_index:0 l_index:18 word_alignment:2 aligned_shape:(1, 13, 13, 3, 86) layout:('N', 'H', 'W', 'X0', 'X1') onnx_layout:('N', 'X0', 'X1', 'H', 'W') scale_factor:1117.906799> (Stage 31) <_lazy_reshape Reshape_39 dtype:int16 shape:(507, 85) alias_of:Transpose_37 default_addr:34240768 g_index:0 l_index:19 word_alignment:2 aligned_shape:(507, 86) onnx_layout:('X0', 'X1') scale_factor:1117.906799> | <transpose Transpose_37 dtype:int16 shape:(1, 3, 13, 13, 85) perm:(0, 3, 1, 2, 4) onnx_perm:(0, 1, 3, 4, 2) default_addr:34240768 g_index:0 l_index:19 word_alignment:2 aligned_shape:(1, 3, 13, 13, 86) layout:('N', 'X0', 'H', 'W', 'X1') onnx_layout:('N', 'X0', 'H', 'W', 'X1') scale_factor:1117.906799> (Stage 32) <slice_ None dtype:int16 shape:(507, 1) begins:(0, 4) ends:(507, 5) strides:(1, 1) default_addr:34409152 g_index:0 l_index:21 word_alignment:2 aligned_shape:(507, 2) onnx_layout:('X0', 'X1') scale_factor:1117.906799> | <_lazy_reshape Reshape_39 dtype:int16 shape:(507, 85) alias_of:Transpose_37 default_addr:34240768 g_index:0 l_index:19 word_alignment:2 aligned_shape:(507, 86) onnx_layout:('X0', 'X1') scale_factor:1117.906799> (Stage 33) <slice_ None dtype:int16 shape:(507, 80) begins:(0, 5) ends:(507, 85) strides:(1, 1) default_addr:34328000 g_index:0 l_index:20 word_alignment:2 aligned_shape:(507, 80) onnx_layout:('X0', 'X1') scale_factor:1117.906799> | <_lazy_reshape Reshape_39 dtype:int16 shape:(507, 85) alias_of:Transpose_37 default_addr:34240768 g_index:0 l_index:19 word_alignment:2 aligned_shape:(507, 86) onnx_layout:('X0', 'X1') scale_factor:1117.906799> (Stage 34) <multiply Mul_52 dtype:int16 shape:(507, 80) default_addr:36869248 g_index:0 l_index:31 word_alignment:2 aligned_shape:(507, 80) onnx_layout:('X0', 'X1') scale_factor:31130.000000> | <sigmoid Sigmoid_49 dtype:int16 shape:(507, 80) chained default_addr:0 word_alignment:2 aligned_shape:(507, 80) onnx_layout:('X0', 'X1') scale_factor:31130.000000> | | <slice_ None dtype:int16 shape:(507, 80) begins:(0, 5) ends:(507, 85) strides:(1, 1) default_addr:34328000 g_index:0 l_index:20 word_alignment:2 aligned_shape:(507, 80) onnx_layout:('X0', 'X1') scale_factor:1117.906799> | <sigmoid Sigmoid_51 dtype:int16 shape:(507, 1) chained default_addr:0 word_alignment:2 aligned_shape:(507, 2) onnx_layout:('X0', 'X1') scale_factor:31130.000000> | | <slice_ None dtype:int16 shape:(507, 1) begins:(0, 4) ends:(507, 5) strides:(1, 1) default_addr:34409152 g_index:0 l_index:21 word_alignment:2 aligned_shape:(507, 2) onnx_layout:('X0', 'X1') scale_factor:1117.906799> (Stage 35) <scaled_concat Concat_85 dtype:int16 shape:(2535, 80) buffered scales:(1, 1) shamt:7 default_addr:0 g_index:1 word_alignment:2 aligned_shape:(2535, 80) onnx_layout:('X0', 'X1') scale_factor:243.203125> | <multiply Mul_52 dtype:int16 shape:(507, 80) default_addr:36869248 g_index:0 l_index:31 word_alignment:2 aligned_shape:(507, 80) onnx_layout:('X0', 'X1') scale_factor:31130.000000> | <multiply Mul_82 dtype:int16 shape:(2028, 80) default_addr:36950400 g_index:0 l_index:32 word_alignment:2 aligned_shape:(2028, 80) onnx_layout:('X0', 'X1') scale_factor:31130.000000> [RAM (spec: num)] 32-bit 1024-entry 2-port 1-bank RAM: 1 16-bit 524288-entry 2-port 2-bank RAM: 2 # <= Huge RAM is requested. 16-bit 8192-entry 2-port 2-bank RAM: 9 16-bit 2048-entry 2-port 2-bank RAM: 10 [Substream (spec: num)] ('_max', (16, 0, True, 4)): 1 ('acc_rshift_round_frac', (64, 0, True, 64, 0, True)): 1 ('add_tree', (64, 0, True, 1)): 1 ('add_tree', (64, 0, True, 9)): 1 ('mul_rshift_round_clip', (64, 0, True, 16, 0, True, 80, 0, True, 16, 0, True, False)): 1 ('mul_rshift_round_madd', (16, 0, True, 16, 0, True, 32, 0, True)): 9 ('reduce_max', (16, 0, True)): 1 [Stream (spec: num)] (((<class 'nngen.operator.conv2d.conv2d'>, <dtype int16>, <dtype int16>, <dtype int32>, <dtype int16>), <dtype int16>, 1), 3, 3, None, <dtype int64>, 1, 1, 1, 1, 9, 9): 1 (((<class 'nngen.operator.pool_serial.max_pool_serial'>, <dtype int16>), <dtype int16>, 1), 2, 2, True, 1): 1 (((<class 'nngen.operator.pad.pad'>, <dtype int16>), <dtype int16>, 1), 1, 1, 1): 1 (((<class 'nngen.operator.pool.max_pool'>, <dtype int16>), <dtype int16>, 1), 2, 2, 1): 1 (((<class 'nngen.operator.conv2d.conv2d'>, <dtype int16>, <dtype int16>, <dtype int32>, <dtype int16>), <dtype int16>, 1), 1, 1, None, <dtype int64>, 1, 1, 1, 1, 1, 1): 1 ((<class 'nngen.operator.upsampling2d.upsampling2d'>, <dtype int16>), <dtype int16>, 1): 1 ((<class 'nngen.operator.normalize.scaled_concat'>, <dtype int16>, <dtype int16>), <dtype int16>, 1): 1 (((<class 'nngen.operator.basic._lazy_reshape'>, <dtype int16>), <dtype int16>, 1), False): 1 ((<class 'nngen.operator.basic.transpose'>, <dtype int16>), <dtype int16>, 1): 1 (((<class 'nngen.operator.basic._lazy_reshape'>, <dtype int16>), <dtype int16>, 1), True): 1 (((<class 'nngen.operator.slice_.slice_'>, <dtype int16>), <dtype int16>, 1), 2, 1): 1 ((<class 'nngen.operator.basic.multiply'>, (((<class 'nngen.operator.sigmoid.sigmoid'>, <dtype int16>), <dtype int16>, 1), 8, 6.0, 0.95), (((<class 'nngen.operator.sigmoid.sigmoid'>, <dtype int16>), <dtype int16>, 1), 8, 6.0, 0.95)), <dtype int16>, 1): 1 [State IDs in main_fsm] (3, 4, 'act', 'None') (12, 14, 'Conv_0', 'control_conv2d_73') (19, 21, 'MaxPool_3', 'control_max_pool_serial_75') (29, 31, 'Conv_4', 'control_conv2d_73') (36, 38, 'MaxPool_7', 'control_max_pool_serial_75') (46, 48, 'Conv_8', 'control_conv2d_73') (53, 55, 'MaxPool_11', 'control_max_pool_serial_75') (63, 65, 'Conv_12', 'control_conv2d_73') (70, 72, 'MaxPool_15', 'control_max_pool_serial_75') (80, 82, 'Conv_16', 'control_conv2d_73') (87, 89, 'MaxPool_19', 'control_max_pool_serial_75') (97, 99, 'Conv_20', 'control_conv2d_73') (103, 105, 'Pad_23', 'control_pad_100') (109, 111, 'MaxPool_24', 'control_max_pool_101') (119, 121, 'Conv_25', 'control_conv2d_73') (129, 131, 'Conv_28', 'control_conv2d_108') (139, 141, 'Conv_55', 'control_conv2d_108') (145, 147, 'Upsample_59', 'control_upsampling2d_128') (153, 155, 'Concat_60', 'control_scaled_concat_129') (163, 165, 'Conv_61', 'control_conv2d_73') (173, 175, 'Conv_64', 'control_conv2d_108') (180, 182, 'Reshape_66', 'control__lazy_reshape_136') (187, 189, 'Transpose_67', 'control_transpose_137') (190, 191, 'Reshape_69', 'None') (196, 198, None, 'control_slice__141') (203, 205, None, 'control_slice__141') (211, 213, 'Mul_82', 'control_multiply_143') (221, 223, 'Conv_31', 'control_conv2d_73') (231, 233, 'Conv_34', 'control_conv2d_108') (238, 240, 'Reshape_36', 'control__lazy_reshape_136') (245, 247, 'Transpose_37', 'control_transpose_137') (248, 249, 'Reshape_39', 'None') (254, 256, None, 'control_slice__141') (261, 263, None, 'control_slice__141') (269, 271, 'Mul_52', 'control_multiply_143') (277, 279, 'Concat_85', 'control_scaled_concat_129') [Control (name (# states: num))] main_fsm (# states: 285) control_conv2d_73 (# states: 56) control_max_pool_serial_75 (# states: 26) control_pad_100 (# states: 20) control_max_pool_101 (# states: 28) control_conv2d_108 (# states: 40) control_upsampling2d_128 (# states: 31) control_scaled_concat_129 (# states: 37) control__lazy_reshape_136 (# states: 22) control_transpose_137 (# states: 22) control_slice__141 (# states: 19) control_multiply_143 (# states: 44) [Register Map] 0 (R ): header0 (default: 0x00000000) 4 (R ): header1 (default: 0x00000000) 8 (R ): header2 (default: 0x00000000) 12 (R ): header3 (default: 0x00000000) 16 ( W): Start (set '1' to run) 20 (R ): Busy (returns '1' when running) 24 ( W): Reset (set '1' to initialize internal logic) 28 (R ): Opcode from extern objects to SW (returns '0' when idle) 32 ( W): Resume extern objects (set '1' to resume) 36 (R ): Interrupt Status Register 40 ( W): Interrupt Enable Register 44 ( W): Interrupt Acknowledge Register 48 (R ): State Counter 52 ( W): Count Target 56 ( W): Count Divider 60 (X): reserved .. 120 (X): .. reserved 124 (R ): Address space amount 128 (RW): Global address offset (default: 0) 132 (RW): Address of temporal storages (size: 17356KB) 136 (RW): Address of output (scaled_concat) 'Concat_85' (size: 397KB, dtype: int16, shape: (2535, 80), alignment: 2 words (4 bytes)), aligned shape: (2535, 80) 140 (RW): Address of placeholder 'act' (size: 1352KB, dtype: int16, shape: (1, 416, 416, 3), alignment: 2 words (4 bytes)), aligned shape: (1, 416, 416, 4) 144 (RW): Address of variables 'module_list.0.Conv2d.weight', 'onnx_Conv_0_conv.bias', 'onnx_Conv_0_conv.scale', 'module_list.2.Conv2d.weight', 'onnx_Conv_4_conv.bias', 'onnx_Conv_4_conv.scale', 'module_list.4.Conv2d.weight', 'onnx_Conv_8_conv.bias', 'onnx_Conv_8_conv.scale', 'module_list.6.Conv2d.weight', 'onnx_Conv_12_conv.bias', 'onnx_Conv_12_conv.scale', 'module_list.8.Conv2d.weight', 'onnx_Conv_16_conv.bias', 'onnx_Conv_16_conv.scale', 'module_list.10.Conv2d.weight', 'onnx_Conv_20_conv.bias', 'onnx_Conv_20_conv.scale', 'module_list.12.Conv2d.weight', 'onnx_Conv_25_conv.bias', 'onnx_Conv_25_conv.scale', 'module_list.13.Conv2d.weight', 'onnx_Conv_28_conv.bias', 'onnx_Conv_28_conv.scale', 'module_list.14.Conv2d.weight', 'onnx_Conv_31_conv.bias', 'onnx_Conv_31_conv.scale', 'module_list.15.Conv2d.weight', 'module_list.15.Conv2d.bias', 'onnx_Conv_34_conv.scale', 'module_list.18.Conv2d.weight', 'onnx_Conv_55_conv.bias', 'onnx_Conv_55_conv.scale', 'module_list.21.Conv2d.weight', 'onnx_Conv_61_conv.bias', 'onnx_Conv_61_conv.scale', 'module_list.22.Conv2d.weight', 'module_list.22.Conv2d.bias', 'onnx_Conv_64_conv.scale' (size: 17298KB) [Default Memory Map (start - end)] (entire range: [0 - 37274879], size: 36402KB) [ 0 - 405631]: output (scaled_concat) 'Concat_85' (size: 397KB, dtype: int16, shape: (2535, 80), alignment: 2 words (4 bytes)), aligned shape: (2535, 80) [ 405632 - 1790079]: placeholder 'act' (size: 1352KB, dtype: int16, shape: (1, 416, 416, 3), alignment: 2 words (4 bytes)), aligned shape: (1, 416, 416, 4) [ 1790080 - 1791231]: variable 'module_list.0.Conv2d.weight' (size: 2KB, dtype: int16, shape: (16, 3, 3, 3), alignment: 2 words (4 bytes)), aligned shape: (16, 3, 3, 4) [ 1791232 - 1791295]: variable 'onnx_Conv_0_conv.bias' (size: 64B, dtype: int32, shape: (16,), alignment: 1 words (4 bytes)), aligned shape: (16,) [ 1791296 - 1791359]: variable 'onnx_Conv_0_conv.scale' (size: 64B, dtype: int16, shape: (16,), alignment: 2 words (4 bytes)), aligned shape: (16,) [ 1791360 - 1800575]: variable 'module_list.2.Conv2d.weight' (size: 9KB, dtype: int16, shape: (32, 3, 3, 16), alignment: 2 words (4 bytes)), aligned shape: (32, 3, 3, 16) [ 1800576 - 1800703]: variable 'onnx_Conv_4_conv.bias' (size: 128B, dtype: int32, shape: (32,), alignment: 1 words (4 bytes)), aligned shape: (32,) [ 1800704 - 1800767]: variable 'onnx_Conv_4_conv.scale' (size: 64B, dtype: int16, shape: (32,), alignment: 2 words (4 bytes)), aligned shape: (32,) [ 1800768 - 1837631]: variable 'module_list.4.Conv2d.weight' (size: 36KB, dtype: int16, shape: (64, 3, 3, 32), alignment: 2 words (4 bytes)), aligned shape: (64, 3, 3, 32) [ 1837632 - 1837887]: variable 'onnx_Conv_8_conv.bias' (size: 256B, dtype: int32, shape: (64,), alignment: 1 words (4 bytes)), aligned shape: (64,) [ 1837888 - 1838015]: variable 'onnx_Conv_8_conv.scale' (size: 128B, dtype: int16, shape: (64,), alignment: 2 words (4 bytes)), aligned shape: (64,) [ 1838016 - 1985471]: variable 'module_list.6.Conv2d.weight' (size: 144KB, dtype: int16, shape: (128, 3, 3, 64), alignment: 2 words (4 bytes)), aligned shape: (128, 3, 3, 64) [ 1985472 - 1985983]: variable 'onnx_Conv_12_conv.bias' (size: 512B, dtype: int32, shape: (128,), alignment: 1 words (4 bytes)), aligned shape: (128,) [ 1985984 - 1986239]: variable 'onnx_Conv_12_conv.scale' (size: 256B, dtype: int16, shape: (128,), alignment: 2 words (4 bytes)), aligned shape: (128,) [ 1986240 - 2576063]: variable 'module_list.8.Conv2d.weight' (size: 576KB, dtype: int16, shape: (256, 3, 3, 128), alignment: 2 words (4 bytes)), aligned shape: (256, 3, 3, 128) [ 2576064 - 2577087]: variable 'onnx_Conv_16_conv.bias' (size: 1KB, dtype: int32, shape: (256,), alignment: 1 words (4 bytes)), aligned shape: (256,) [ 2577088 - 2577599]: variable 'onnx_Conv_16_conv.scale' (size: 512B, dtype: int16, shape: (256,), alignment: 2 words (4 bytes)), aligned shape: (256,) [ 2577600 - 4936895]: variable 'module_list.10.Conv2d.weight' (size: 2304KB, dtype: int16, shape: (512, 3, 3, 256), alignment: 2 words (4 bytes)), aligned shape: (512, 3, 3, 256) [ 4936896 - 4938943]: variable 'onnx_Conv_20_conv.bias' (size: 2KB, dtype: int32, shape: (512,), alignment: 1 words (4 bytes)), aligned shape: (512,) [ 4938944 - 4939967]: variable 'onnx_Conv_20_conv.scale' (size: 1KB, dtype: int16, shape: (512,), alignment: 2 words (4 bytes)), aligned shape: (512,) [ 4939968 - 14377151]: variable 'module_list.12.Conv2d.weight' (size: 9216KB, dtype: int16, shape: (1024, 3, 3, 512), alignment: 2 words (4 bytes)), aligned shape: (1024, 3, 3, 512) [14377152 - 14381247]: variable 'onnx_Conv_25_conv.bias' (size: 4KB, dtype: int32, shape: (1024,), alignment: 1 words (4 bytes)), aligned shape: (1024,) [14381248 - 14383295]: variable 'onnx_Conv_25_conv.scale' (size: 2KB, dtype: int16, shape: (1024,), alignment: 2 words (4 bytes)), aligned shape: (1024,) [14383296 - 14907583]: variable 'module_list.13.Conv2d.weight' (size: 512KB, dtype: int16, shape: (256, 1, 1, 1024), alignment: 2 words (4 bytes)), aligned shape: (256, 1, 1, 1024) [14907584 - 14908607]: variable 'onnx_Conv_28_conv.bias' (size: 1KB, dtype: int32, shape: (256,), alignment: 1 words (4 bytes)), aligned shape: (256,) [14908608 - 14909119]: variable 'onnx_Conv_28_conv.scale' (size: 512B, dtype: int16, shape: (256,), alignment: 2 words (4 bytes)), aligned shape: (256,) [14909120 - 17268415]: variable 'module_list.14.Conv2d.weight' (size: 2304KB, dtype: int16, shape: (512, 3, 3, 256), alignment: 2 words (4 bytes)), aligned shape: (512, 3, 3, 256) [17268416 - 17270463]: variable 'onnx_Conv_31_conv.bias' (size: 2KB, dtype: int32, shape: (512,), alignment: 1 words (4 bytes)), aligned shape: (512,) [17270464 - 17271487]: variable 'onnx_Conv_31_conv.scale' (size: 1KB, dtype: int16, shape: (512,), alignment: 2 words (4 bytes)), aligned shape: (512,) [17271488 - 17532607]: variable 'module_list.15.Conv2d.weight' (size: 255KB, dtype: int16, shape: (255, 1, 1, 512), alignment: 2 words (4 bytes)), aligned shape: (255, 1, 1, 512) [17532608 - 17533631]: variable 'module_list.15.Conv2d.bias' (size: 1KB, dtype: int32, shape: (255,), alignment: 1 words (4 bytes)), aligned shape: (255,) [17533632 - 17533695]: variable 'onnx_Conv_34_conv.scale' (size: 64B, dtype: int16, shape: (1,), alignment: 2 words (4 bytes)), aligned shape: (2,) [17533696 - 17599231]: variable 'module_list.18.Conv2d.weight' (size: 64KB, dtype: int16, shape: (128, 1, 1, 256), alignment: 2 words (4 bytes)), aligned shape: (128, 1, 1, 256) [17599232 - 17599743]: variable 'onnx_Conv_55_conv.bias' (size: 512B, dtype: int32, shape: (128,), alignment: 1 words (4 bytes)), aligned shape: (128,) [17599744 - 17599999]: variable 'onnx_Conv_55_conv.scale' (size: 256B, dtype: int16, shape: (128,), alignment: 2 words (4 bytes)), aligned shape: (128,) [17600000 - 19369471]: variable 'module_list.21.Conv2d.weight' (size: 1728KB, dtype: int16, shape: (256, 3, 3, 384), alignment: 2 words (4 bytes)), aligned shape: (256, 3, 3, 384) [19369472 - 19370495]: variable 'onnx_Conv_61_conv.bias' (size: 1KB, dtype: int32, shape: (256,), alignment: 1 words (4 bytes)), aligned shape: (256,) [19370496 - 19371007]: variable 'onnx_Conv_61_conv.scale' (size: 512B, dtype: int16, shape: (256,), alignment: 2 words (4 bytes)), aligned shape: (256,) [19371008 - 19501567]: variable 'module_list.22.Conv2d.weight' (size: 128KB, dtype: int16, shape: (255, 1, 1, 256), alignment: 2 words (4 bytes)), aligned shape: (255, 1, 1, 256) [19501568 - 19502591]: variable 'module_list.22.Conv2d.bias' (size: 1KB, dtype: int32, shape: (255,), alignment: 1 words (4 bytes)), aligned shape: (255,) [19502592 - 19502655]: variable 'onnx_Conv_64_conv.scale' (size: 64B, dtype: int16, shape: (1,), alignment: 2 words (4 bytes)), aligned shape: (2,) [19502656 - 37274879]: temporal storages (size: 17356KB)
It is not able to embed yolov3-tiny to ultra96 due to that model require large RAM at slice operator near by yolo layer.
Verbose log when executing example/yolov3-tiny.