Yunfeng Wang

Hi @vchoutas, Could you please merge this pull request so that we can use smplx to dump torchscript

Same question here. If we can train SPIN and VIBE together, we can adopt both parts of them and scale to different network sizes.

Same concern here

Hi Lea, Thanks for your clean answer. Actually I didn't figure it out and stopped the experiments after several trials 🤣 . This answer definitely will help me too! Thanks.

Combination with Implicit human reconstruction: [ICON: Implicit Clothed humans Obtained from Normals](https://icon.is.tue.mpg.de)

Hi @Avatair , Thanks for trying talkGPT4All and reporting this issue and solution. I will check it and fix it as soon as possible.

你好，感谢使用和反馈，能否发一下错误报告的详细内容，我分析一下。 pyttsx3的依赖问题我看看

@sologuy 我看了log，查了下是Python的内存越界问题，比较底层，由于这个仓库使用的是GPT4All编译好的二进制文件，我目前也没比较好的方法来定位问题。你看使用GPT4All的Python Binding是否正常 https://github.com/nomic-ai/gpt4all/blob/main/gpt4all-bindings/python/README.md

这个问题还存在，核心原因是mmlu_pro eval时的正则表达式有问题，提取了第一个符合格式`answer: x` 的字母X作为回答。 mmlu_pro的[eval_cfg](https://github.com/open-compass/opencompass/blob/main/opencompass/configs/datasets/mmlu_pro/mmlu_pro_0shot_cot_gen_08c1de.py#L52): ```python mmlu_pro_eval_cfg = dict( evaluator=dict(type=AccEvaluator), pred_postprocessor=dict( type=match_answer_pattern, answer_pattern=r'(?i)ANSWER\s*:\s*([A-P])') ) ``` 对应的[match_answer_pattern](https://github.com/open-compass/opencompass/blob/main/opencompass/utils/text_postprocessors.py#L241)： ```python def match_answer_pattern(response_text: str, answer_pattern: str): match = re.search(answer_pattern, response_text) extracted_answer = match.group(1) if...

Same issue here. 160 is too short for applications like iOS.