【免费】gkt上板训练编译结果资源-CSDN文库

共26个文件

png：14个

tar：3个

onnx：3个

需积分: 0 85 浏览量 2024-04-15 08:45:05 上传评论收藏 152.59MB ZIP 举报

"gkt上板训练编译结果" 指的是使用gkt工具进行的在硬件板上运行的训练程序的编译过程所得到的最终输出。这通常涉及到嵌入式系统、软件开发和机器学习等多个IT领域的知识点。GKT可能是一个专为特定平台或设备设计的开发框架，用于在硬件平台上执行训练任务，如神经网络模型的训练。【gkt】标签暗示了这是一个专门针对gkt工具或系统的讨论。gkt可能是某种开源库、开发环境或者是针对特定硬件的加速器库。它可能包含了优化的算法，以便在资源受限的设备上高效运行计算密集型任务，例如在嵌入式设备或物联网(IoT)装置上进行机器学习。中的“上板训练”意味着将模型部署到实际的硬件板上进行训练，而不是在模拟环境中。这种做法可以更好地反映出实际运行时的性能，因为硬件限制和实时数据处理是无法在仿真中完全模拟的。这通常涉及到交叉编译、设备驱动程序的适配以及对硬件性能特性的理解。在IT领域，这个过程涵盖以下几个关键知识点： 1. **交叉编译**：由于开发环境与目标硬件平台可能不同，需要使用交叉编译工具链来编译代码，使其能够在目标硬件上运行。 2. **嵌入式系统**：gkt可能被用于嵌入式系统，这些系统通常有严格的资源限制（内存、处理器速度等），因此需要优化代码以提高效率。 3. **硬件加速**：gkt可能利用硬件加速功能，如GPU或专用AI芯片，来提升训练速度和效率。 4. **机器学习框架**：gkt可能与TensorFlow、PyTorch等常见的机器学习框架兼容，允许开发者在其上构建和训练模型。 5. **设备驱动程序**：为了充分利用硬件，需要编写或适配设备驱动程序，以确保软件能够正确地与硬件交互。 6. **实时数据处理**：在硬件板上进行训练可能涉及到实时数据采集和处理，这需要对实时操作系统和数据流管理有深入理解。 7. **调试与性能分析**：在硬件板上进行编译和训练，需要具备远程调试技巧以及性能分析能力，以优化代码并解决可能出现的问题。 8. **软件版本控制**：在整个开发过程中，版本控制工具如Git对于跟踪代码变更、协同开发至关重要。 9. **文档与日志**：详细记录编译过程和结果，便于问题排查和后期维护。 "gkt上板训练编译结果"是一个涉及嵌入式开发、机器学习、硬件优化等多个领域的复杂工程，需要开发者具备跨学科的知识和技能。而压缩包中的"gkt"文件可能包含了相关的源代码、编译脚本、配置文件或者训练日志，用于进一步分析和研究编译过程和训练效果。

资源推荐

资源详情

资源评论

收起资源包目录

gkt.zip （26个子文件）

gkt

上板fps.png 119KB

float测试.png 98KB

calib测试.png 100KB

上板latency.png 122KB

onnx

calibration-checkpoint-best.pth.tar 17.71MB

float和int算子差异

2024-04-09 09-20-19屏幕截图.png 9KB

2024-04-09 09-24-00屏幕截图.png 5KB

resize浮点模型中.png 6KB

2024-04-09 09-24-17屏幕截图.png 4KB

2024-04-09 09-24-45屏幕截图.png 47KB

float.onnx 19.74MB

float.onnx.png 8.54MB

float-checkpoint-best.pth.tar 47.84MB

int_infer-checkpoint-last.pth.tar 17.09MB

qat.onnx 21.11MB

int_infer.onnx 16.09MB

qat.onnx.png 5.49MB

int_infer.onnx.png 8.48MB

int测试.png 101KB

nusc_pred.png 726KB

compile_gkt

bev_gkt_mixvargenet_multitask_nuscenes_model.json 71KB

model.svg 383KB

model.pt 18.13MB

model.hbir 5MB

model-26d1052d.hbm 8.13MB

bev_gkt_mixvargenet_multitask_nuscenes_model.html 2.19MB

<!DOCTYPE html> <html> <head><title>bev_gkt_mixvargenet_multitask_nuscenes_model</title></head> <body> <div id="App"></div> </body> <script type="text/babel"> class App extends React.Component { constructor(props) { super(props) this.state = { jsonObject: {"BPU OPs per frame (effective)":34492078080,"BPU OPs per run (effective)":206952468480,"BPU PE number":1,"BPU core number":1,"BPU march":"BAYES","DDR bytes per frame":17934784,"DDR bytes per run":107608704,"DDR bytes per second":4693316802,"DDR megabytes per frame":17.104,"DDR megabytes per run":102.624,"DDR megabytes per second":4475.9,"FPS":261.69,"HBDK version":"3.48.6","compiling options":"--march bayes -m compile_gkt/model.hbir -f hbir --O3 -o compile_gkt/model.hbm --jobs 4 -n bev_gkt_mixvargenet_multitask_nuscenes_model -i pyramid,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr,ddr --input-name arg0[img],arg0[points0]_torch_native,arg0[points1]_torch_native,arg0[points2]_torch_native,arg0[points3]_torch_native,arg0[points4]_torch_native,arg0[points5]_torch_native,arg0[points6]_torch_native,arg0[points7]_torch_native,arg0[points8]_torch_native --output-layout NCHW --progressbar --debug","frame per run":6,"frame per second":261.69,"input features":[["input name","input size"],["arg0[img]","6x512x960x3"],["arg0[points0]_torch_native","6x64x64x2"],["arg0[points1]_torch_native","6x64x64x2"],["arg0[points2]_torch_native","6x64x64x2"],["arg0[points3]_torch_native","6x64x64x2"],["arg0[points4]_torch_native","6x64x64x2"],["arg0[points5]_torch_native","6x64x64x2"],["arg0[points6]_torch_native","6x64x64x2"],["arg0[points7]_torch_native","6x64x64x2"],["arg0[points8]_torch_native","6x64x64x2"]],"interval computing unit utilization":[0.413,0.999,0.999,0.957,0.998,0.957,0.999,0.998,0.938,1.0,1.0,1.0,1.0,0.968,0.948,0.926,0.788,0.962,0.957,0.876,0.791,0.634,0.575,0.708,0.684,0.613,0.754,0.584,0.794,0.493,0.511,0.493,0.644,0.569,0.67,0.986,0.888,0.999,0.824,0.984,0.98,0.977,0.982,0.975,0.983,0.811],"interval computing units utilization":[0.413,0.999,0.999,0.957,0.998,0.957,0.999,0.998,0.938,1.0,1.0,1.0,1.0,0.968,0.948,0.926,0.788,0.962,0.957,0.876,0.791,0.634,0.575,0.708,0.684,0.613,0.754,0.584,0.794,0.493,0.511,0.493,0.644,0.569,0.67,0.986,0.888,0.999,0.824,0.984,0.98,0.977,0.982,0.975,0.983,0.811],"interval loading bandwidth (megabytes/s)":[2349,1781,1117,1174,1253,1150,1185,1317,1325,1027,747,747,791,1367,2306,2582,2742,3719,4520,5070,6030,5252,4678,5423,5414,5540,5391,5441,5370,5295,5208,4747,4430,5027,5605,2938,960,903,1168,1155,409,373,414,414,457,378],"interval number":46,"interval storing bandwidth (megabytes/s)":[3172,1914,615,613,611,621,631,625,864,1141,1168,1168,1189,946,873,2429,4702,3822,2065,2314,2695,1443,2218,3489,2544,2324,2495,2429,2613,3337,3722,3908,2847,914,0,688,1042,867,927,1059,1162,1202,1329,1285,1226,1242],"interval time (ms)":0.5,"latency (ms)":22.93,"latency (ms) by segments":[22.928],"latency (us)":22928.1,"layer details":[["layer","ops","original output shape","aligned output shape","computing cost (no DDR)","load/store cost","active period of time"],["_backbone_mod1_0_hz_conv2d","1,274,019,840","[6,256,480,32]","[6,256,480,32]","205 us (0.8% of model)","685 us (2.9% of model)","317 ~ 3913 us (3596)"],["_backbone_mod2_0_head_layer_conv_0_0_hz_conv2d","27,179,089,920","[6,256,480,64]","[6,256,480,64]","1376 us (6.0% of model)","3 us (0% of model)","327 ~ 3944 us (3617)"],["_backbone_mod2_0_head_layer_short_add_hz_conv2d","3,019,898,880","[6,256,480,32]","[6,256,480,32]","329 us (1.4% of model)","1 us (0% of model)","333 ~ 3948 us (3615)"],["_backbone_mod3_0_head_layer_conv_0_0_hz_conv2d","13,589,544,960","[6,128,240,128]","[6,128,240,128]","849 us (3.7% of model)","5 us (0% of model)","338 ~ 3966 us (3628)"],["_backbone_mod3_0_head_layer_downsample_0_hz_conv2d","377,487,360","[6,128,240,32]","[6,128,240,32]","189 us (0.8% of model)","1 us (0% of model)","346 ~ 3964 us (3618)"],["_backbone_mod3_0_head_layer_short_add_hz_conv2d","1,509,949,440","[6,128,240,32]","[6,128,240,32]","82 us (0.3% of model)","1 us (0% of model)","347 ~ 3968 us (3621)"],["_backbone_mod3_0_stack_layers_0_conv_0_0_hz_conv2d","13,589,544,960","[6,128,240,128]","[6,128,240,128]","691 us (3.0% of model)","5 us (0% of model)","349 ~ 3999 us (3650)"],["_backbone_mod3_0_stack_layers_0_short_add_hz_conv2d","1,509,949,440","[6,128,240,32]","[6,128,240,32]","82 us (0.3% of model)","289 us (1.2% of model)","357 ~ 4004 us (3647)"],["_backbone_mod3_0_stack_layers_1_conv_0_0_hz_conv2d","13,589,544,960","[6,128,240,128]","[6,128,240,128]","691 us (3.0% of model)","292 us (1.2% of model)","3878 ~ 6663 us (2785)"],["_backbone_mod3_0_stack_layers_1_short_add_hz_conv2d","1,509,949,440","[6,128,240,32]","[6,128,240,32]","123 us (0.5% of model)","289 us (1.2% of model)","3948 ~ 6696 us (2748)"],["_backbone_mod4_0_head_layer_conv_0_0_hz_conv2d","3,397,386,240","[6,64,120,128]","[6,64,128,128]","228 us (0.9% of model)","5 us (0% of model)","3884 ~ 6698 us (2814)"],["_backbone_mod4_0_head_layer_downsample_0_hz_conv2d","188,743,680","[6,64,120,64]","[6,64,128,64]","54 us (0.2% of model)","1 us (0% of model)","3955 ~ 6698 us (2743)"],["_backbone_mod4_0_head_layer_short_add_hz_conv2d","754,974,720","[6,64,120,64]","[6,64,128,64]","44 us (0.1% of model)","2 us (0% of model)","3981 ~ 6701 us (2720)"],["_backbone_mod4_0_stack_layers_0_conv_0_0_hz_conv2d","13,589,544,960","[6,64,120,256]","[6,64,128,256]","734 us (3.2% of model)","17 us (0% of model)","3964 ~ 6768 us (2804)"],["_backbone_mod4_0_stack_layers_0_short_add_hz_conv2d","1,509,949,440","[6,64,120,64]","[6,64,128,64]","85 us (0.3% of model)","3 us (0% of model)","3984 ~ 6775 us (2791)"],["_backbone_mod4_0_stack_layers_1_conv_0_0_hz_conv2d","13,589,544,960","[6,64,120,256]","[6,64,128,256]","734 us (3.2% of model)","17 us (0% of model)","3987 ~ 6840 us (2853)"],["_backbone_mod4_0_stack_layers_1_short_add_hz_conv2d","1,509,949,440","[6,64,120,64]","[6,64,128,64]","106 us (0.4% of model)","307 us (1.3% of model)","4028 ~ 7122 us (3094)"],["_backbone_mod5_0_head_layer_conv_0_0_hz_conv2d","424,673,280","[6,32,60,128]","[6,32,64,128]","33 us (0.1% of model)","3 us (0% of model)","6663 ~ 7132 us (469)"],["_backbone_mod5_0_head_layer_downsample_0_hz_conv2d","141,557,760","[6,32,60,96]","[6,32,64,96]","31 us (0.1% of model)","2 us (0% of model)","6685 ~ 7129 us (444)"],["_backbone_mod5_0_head_layer_short_add_hz_conv2d","283,115,520","[6,32,60,96]","[6,32,64,96]","17 us (0% of model)","2 us (0% of model)","6686 ~ 7134 us (448)"],["_backbone_mod5_0_stack_layers_0_conv_0_0_hz_conv2d","637,009,920","[6,32,60,192]","[6,32,64,192]","42 us (0.1% of model)","4 us (0% of model)","6699 ~ 7144 us (445)"],["_backbone_mod5_0_stack_layers_0_short_add_hz_conv2d","424,673,280","[6,32,60,96]","[6,32,64,96]","24 us (0.1% of model)","3 us (0% of model)","6703 ~ 7148 us (445)"],["_backbone_mod5_0_stack_layers_1_conv_0_0_hz_conv2d","637,009,920","[6,32,60,192]","[6,32,64,192]","42 us (0.1% of model)","4 us (0% of model)","6706 ~ 7158 us (452)"],["_backbone_mod5_0_stack_layers_1_short_add_hz_conv2d","424,673,280","[6,32,60,96]","[6,32,64,96]","24 us (0.1% of model)","3 us (0% of model)","6776 ~ 7162 us (386)"],["_backbone_mod5_0_stack_layers_2_conv_0_0_hz_conv2d","637,009,920","[6,32,60,192]","[6,32,64,192]","42 us (0.1% of model)","4 us (0% of model)","6779 ~ 7172 us (393)"],["_backbone_mod5_0_stack_layers_2_short_add_hz_conv2d","424,673,280","[6,32,60,96]","[6,32,64,96]","24 us (0.1% of model)","3 us (0% of model)","6844 ~ 7174 us (330)"],["_backbone_mod5_0_stack_layers_3_conv_0_0_hz_conv2d","637,009,920","[6,32,60,192]","[6,32,64,192]","36 us (0.1% of model)","4 us (0% of model)","7122 ~ 7337 us (215)"],["_backbone_mod5_0_stack_layers_3_short_add_hz_conv2d","424,673,280","[6,32,60,96]","[6,32,64,96]","23 us (0.1% of model)","3 us (0% of model)","7136 ~ 7345 us (209)"],["_backbone_mod5_0_stack_layers_4_conv_0_0_hz_conv2d","637,009,920","[6,32,60,192]","[6,32,64,192]","36

评论收藏

内容反馈