nncase
nncase copied to clipboard
kendryte
Open deep learning compiler stack for Kendryte AI accelerators ✨
- [x] layernorm ir - [x] layernorm importer(onnx) - [x] layernorm kernels - [x] layernorm pytest
Unfortunately, #310 broke ONNX import for models with graphs which don't have inputs and outputs of the nodes mentioned in `value_info` field of the graph, which is [optional](https://github.com/onnx/onnx/blob/f81558d4b24e910cbd05990f09e4ae78b7452560/onnx/onnx.proto#L463) and not...
**Describe the bug** 分别使用nncase1.6和nncase0.2beta4版本编译tflite，在k210上都能跑通， 但是nncase1.6的kmodel耗时约为400ms，而nncase0.2beta4的kmodel大约150ms。 **To Reproduce** 为什么新版本的模型的运行时间会这么长呢？ **Origin model and code** 原始tflite模型为[MobileFaceNet.tflite](https://github.com/wei8171023/LightFaceNet/blob/master/output/ckpt_best/mobilefacenet_best_ckpt/MobileFaceNet.tflite) **Environment (please complete the following information):** - OS: Ubuntu20.04] - nncase 1.6/0.2beta4 - DL Framework TensorFlow1.14 **Additional...
**Describe the bug** 使自设voc数据集训练得到的pth转出onnx，且经过测试没有问题，ap也比较高。按照教程进行编译k210所使用的定点模型，测试时时出现报错 我的voc数据集是20类，我已经将文件中类别相关的地方都做了修改，但是量化后是没有结果的，也不能标出检测框 可以详细问一下量化文件是什么原理 **Origin model and code** If applicable, add orgin model and code to help explain your problem. ****Environment (please complete the following information):** - OS:...
**Describe the bug** After i run the script to convert caffemodel to model ,I am getting the following error **ERROR** python convert.py 1. Import graph... 2. Optimize target independent... **Segmentation...
Hello, I get a "core dump: illegal instruction" in the k210_runtime_module::visit(const kpu_conv2d_options &op) function on the kendrye. It is raised by an alignment check : CHECK_WITH_ERR(layer.kernel_calc_type_cfg.data.active_addr % 256 == 0,...
how to fix the issue "Fatal: Not supported tflite opcode: DEQUANTIZE",please
