PaddleSlim icon indicating copy to clipboard operation
PaddleSlim copied to clipboard

PaddleSlim is an open-source library for deep model compression and architecture search.

Results 256 PaddleSlim issues
Sort by recently updated
recently updated
newest added

大佬们,我用自动压缩工具压缩了ppyoloe的,但是我用C++代码进行推断的时间没有什么提升,请问会是什么问题? 这是我压缩前后的模型: 链接:https://pan.baidu.com/s/1QYiPK_zu6j--rYp0ncq06g 提取码:9dcw

使用方法参考[官方示例文档](https://github.com/PaddlePaddle/PaddleSlim/tree/develop/example/auto_compression/detection), 通过指令导出模型: ```python python tools/export_model.py \ -c configs/ppyoloe/ppyoloe_plus_crn_s_80e_coco.yml \ -o weights=~/ss/code/PaddleYOLO/output/ppyoloe_plus_crn_s_80e_coco_shrimp/best_model.pdparams \ trt=True exclude_nms=True ``` 通过指令训练模型: ``` CUDA_VISIBLE_DEVICES=0,1 python -m paddle.distributed.launch --log_dir=log --gpus 0,1 run.py \ --config_path=./configs/ppyoloe_x_qat_dis.yaml --save_dir='./output/' ``` 模型训练时的部分log如下:...

报错如图,按照流程使用官方提供的eyolov7-tiny.onnx已经实现压缩。 ![QQ截图20230412182823](https://user-images.githubusercontent.com/91938255/231431187-f87242df-ade4-41c4-9079-122cb0fc80ba.png) 模型训练的是自己的数据集,格式是从voc转的,如图 ![image](https://user-images.githubusercontent.com/91938255/231430185-ca45dae6-8ce6-4aa4-bcd0-d5f6174e4878.png) 求解答,万分感谢!!!!!

用yolov5 -l的检测模型训练了自己的数据集,然后转onnx用run.py自动化压缩。最后得到的onnx模型转engine之后。推理没有任何的输出信息。但是程序没有报告bug。

您好,请问PaddleSlim离线量化以及其他压缩方式所需要的数据集格式均必须为COCO格式的吗?如果用VOC格式,则会报错: ![image](https://user-images.githubusercontent.com/71055342/222896873-591ec53a-89fb-448e-9cfc-f8e95f287728.png) ,但是我自己的VOC格式数据集,转成COCO格式,精度下降很多(40-50百分点),是否还有其他解决办法?

使用pruner = L1NormFilterPruner(trainer.model, input_spec)时,对正常卷积的模型可以运行,但模型里面有深度可分离卷积的就会报错误,2023-06-10 16:31:29,724-WARNING: Leaves ['eager_tmp_5', 'batch_norm_75.tmp_0', 'batch_norm_75.tmp_1', 'batch_norm_76.tmp_0', 'batch_norm_76.tmp_1', 'batch_norm_77.tmp_0', 'batch_norm_77.tmp_1', 'batch_norm_78.tmp_0', 'batch_norm_78.tmp_1', 'batch_norm_79.tmp_0', 'batch_norm_79.tmp_1', 'batch_norm_80.tmp_0', 'batch_norm_80.tmp_1', 'batch_norm_81.tmp_0', 'batch_norm_81.tmp_1', 'batch_norm_82.tmp_0', 'batch_norm_82.tmp_1', 'batch_norm_83.tmp_0', 'batch_norm_83.tmp_1', 'batch_norm_84.tmp_0', 'batch_norm_84.tmp_1', 'batch_norm_85.tmp_0', 'batch_norm_85.tmp_1', 'batch_norm_86.tmp_0',...

您好,请问CNN+Transformer部署的思路过程是什么?有参考学习的代码吗?

https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/quant/analysis.py#L416 paddle gpu:2.4.2 post112 paddleslim2.4.1 分析精度损失:评估函数不可使用,利用fp_int_cosine_similarity 内存占用一直上升,直接到最后就是ResourceExhaustedError: Fail to alloc memory of 524288000 size, error code is 12. Sampling stage, Run batch:| | 0/1W0531 14:29:24.876821 1627 sampler.cpp:189] bvar is busy...