luhairong11

> Hello, pre-trained weights is depend on your backbone. > for darknet53 backbone, use darknet53.conv.74. > for csresnext50 backbone, use csresnext50.conv.80. > for csdarknet53 backbone, use csdarknet53.conv.104. > etc. where...

> @HuangYG123 I got the results here when using file pretrained model CurricularFace_Backbone.pth. These are far more different from the pubic results. Please check your file CurricularFace_Backbone.pth > Evaluation: LFW...

> @luhairong11 I downloaded the (preprocessed) LFW data including the pairs file from [here](https://github.com/IrvingMeng/MagFace) and achieved at least on LFW = 99.783 using the provided CurricularFace model. Still a bit...

有输出的格式示例吗： def parse_output_json(self, data, prompt): usage = data.get("usage", None) assert len(data["choices"]) == 1, f"Too many choices {len(data['choices'])}" choice = data["choices"][0] if self.parsed_options.chat: if self.parsed_options.stream: text = choice["delta"].get("content", "") else: text...

 已经跑通了

我实现了，但是每次运行的结果不一样，有遇到过的吗

> > 我实现了，但是每次运行的结果不一样，有遇到过的吗 > > 我也实现了 **但结果是一样的**，用的opencompass里的openai的api改的，方法为： 修改openai接口为 `os.environ.get('OPENAI_BASE_URL', 'http://localhost:11434/v1/'),` 然后我举一个使用例子： > > ``` > from opencompass.models import OpenAI > > api_meta_template = dict(round=[ > dict(role='HUMAN', api_role='HUMAN'), > dict(role='BOT', api_role='BOT',...

> > 我实现了，但是每次运行的结果不一样，有遇到过的吗 > > 我也实现了 **但结果是一样的**，用的opencompass里的openai的api改的，方法为： 修改openai接口为 `os.environ.get('OPENAI_BASE_URL', 'http://localhost:11434/v1/'),` 然后我举一个使用例子： > > ``` > from opencompass.models import OpenAI > > api_meta_template = dict(round=[ > dict(role='HUMAN', api_role='HUMAN'), > dict(role='BOT', api_role='BOT',...