Daniel

Results 101 comments of Daniel

Looking forward to your good news

You could just use the adjusted darknet.

https://blog.csdn.net/linghu8812/article/details/109270320

in the folder of its source code, you can find an original yolo-fastest.weights , then get it with below command line: ./darknet partial yolo-fastest.cfg yolo-fastest.weights yolo-fastest.conv.109 109

直接运行torchscript模型比较好。建议支持torchscript和onnx模型。 MegEninge支持的太少了,试了一下,从onnx转换到megEngine模型就是挑战。

库主没有计划支持吗? 想法很好,但是通用性欠佳。

是的,我测试过通过onnx 转到mengine类型,好多转换不了。 建议能不能直接读取pnnx格式进行编译?类似ncnn这样的,通用性会好很多,可以直接打通pytorch到megcc之间的通道。

Epoch 2, acc: 0.91542254, loss: 0.18063137 model file: checkpoint/open_pt_toutiao_opt_prompt_1_prefix_cn_lr_0.0001_bsize_32_seed_42_a100/model_epoch_2_acc_0.91542254_loss_0.180631369.pt