Zening Lin
Zening Lin
我也认为不妥。之前做过一个ESP32的项目，LDO脑抽没看手册选择了250mA的，导致一开射频（WiFi、BLE等）就有几率导致芯片重启，即使成功运行产生的射频信号也极弱，无法满足项目的需求，还是选一个1A的LDO比较好。
> 之前用的是RT9080-33GJ5，600mA，相比原来的效果有改善，至少不会因为电平被拉低重启了
change the variable _URL in huggingface/modules/datasets_modules/datasets/xfun/xxxxxxxxxx/xfun.py to "https://github.com/doc-analysis/XFUND/releases/download/v1.0/"
看样子是某个库版本对不上 检查一下当前的环境配置，按照README里面的要求重新安装一下conda环境
Hello, Sorry for my late reply. The OCR I used is an private engine of IntSig, which is for in-house development only. The public version of the API is [TextIn](https://www.textin.com/experience/text_recognize_3d1)...
Hi, Something seems weird in your returned data. The `ocr_corpus` refers to the token ids converted by the tokenizer, and the `seg_indices` corresponds to the segment index that a token...
The returns of your wrapper seems correct this time. Now we should try to figure out which part of the pipeline failed, and I need more information. Try to add...
Hello, It seems that you set the classifier to "full" mode, which has the same architecture described in the paper. I also found that this setting gives poor results, as...
Hello~ A pre-train weight is available at [google drive](https://drive.google.com/file/d/1VI5g9wbtc0uihhCmM3g5_g20UFVOGhw6/view?usp=sharing), try to load it using the `simp` classifier mode and see whether it gives results or not.
Hello~ Sorry for my delayed response. I found that the `weights` arg in your config file is empty, which will not load any checkpoint and will use the randomly initialized...