chenbohua3

Could you please upload the Pytorch version of SK-ResNeXt and hence we can reproduce the paper performance of it. Thank you:)

I searched the snpe docs but got nothing:(

Hello Jie, Could you please share the hyper-parameters for GENet on CIFAR10/100, I can not reproduce the result based on hyper-parameters mentioned in [here](https://arxiv.org/pdf/1603.05027.pdf). Currently, the result(original preact-Resnet110 on CIFAR10/100)...

Any ideas about the structure of unsupervised SUM-FCN

1

After reading chapter 3.3 in FCSN several times, I can not figure out what exactly structure of the unsupervised part. Is that mean: 1. select Y frames: choose the top...

Also apply linter to the codes

The overall process can be divided into the following steps: - [x] make each subgraph executable, this allows us to collect the inputs of each subgraph when inference with the...

Hi, thank you for your codes, it dose help me a lot:) But there is still some error in the code like `transforms.RandomSizedCrop(size0)` in `train_imagenet.py ` and it seems that...

[Feature] 如何贡献GPTQ相关的代码以及模型

6

### Is your feature request related to a problem? Please describe. Hi，我目前已经用GPTQ对ChatGLM-6B做了8/4bit的量化，量化后的模型依旧能够流畅对话。想问一下如果如果贡献这部分内容的话，该怎么操作。我想了一些可能的形式哈： 1. 直接贡献GPTQ calibrated 的模型参数，并适配ChatGLM的推理流程。 2. 我对GPTQ做了一定的工具化，可以把这部分内容直接直接贡献到.quantize接口里来 上面两种方法背后都还有一些待讨论的点，例如GPTQ理论上应该是比min/max的算法要好，但是他需要一定的calibration set，我手边没有相关数据集来采样calibration set，也没办法定量说明GPTQ calibrated的模型精度好于min/max。以及GPTQ的校准过程会有更高的内存/显存的要求。 ### Solutions 见上：） ### Additional context _No response_

Numerical overflow happens on GEMMLowpOutputStage

3

I am currently unable to judge whether my `GEMMLowpOutputStageInfo` configuration is wrong or there are some bugs in the codes. The `GEMMLowpOutputStageInfo` I configured is listed below: ``` arm_compute::GEMMLowpOutputStageInfo osInfo;...