momo

I have the same question, i can not understand here.

> Hi, > > 1. I notice that the labels created in InfoNCE loss is always a zero-vector:(https://github.com/salesforce/PCL/blob/964da1fb7c0546e8ce55627fa3c0debde4b7e456/pcl/builder.py#L163 > ) > I think this is wrong since otherwise the loss...

> 运行 InternVideo2_stage2_1B 的 demo.ipynb 时，intern_model, tokenizer = setup_internvideo2(config) 会报如下警告： load_state_dict: _IncompatibleKeys(missing_keys=[], unexpected_keys=['temp', 'itm_head.weight', 'itm_head.bias']) > > 最终能够得到如下运行结果： text: A man in a gray sweater plays fetch with his dog...

> 想要在自有中文检索数据集上应用internvideo2，看了下最接近的是vatex_cn中使用internvideo2_clip模型，但是在加载过程中遇到些疑问。 > > 1. 共计需要加载哪些ckpt？ > 目前看是这几个组件 > chinese_alpaca_lora_7b > InternVideo2-stage2_1b-224p-f4.pt > 1B_clip.pth > internvl_c_13b_224px.pth > 配置文件如下 > tokenizer_path="chinese_alpaca_lora_7b", > vision_ckpt_path="OpenGVLab__InternVideo2-Stage2_1B-224p-f4/InternVideo2-stage2_1b-224p-f4.pt", > load_vision_ckpt_from_internvideo2_stage2=True, > text_ckpt_path="internvl_c_13b_224px.pth" > extra_ckpt_path="OpenGVLab__InternVideo2-CLIP-1B-224p-f8/1B_clip.pth" > >...

Hello, have you resolved this issue? I'm encountering the same problem now.

> > @nsreeprem The 0-shot performance I measured on MSRVTT by using the s2-1B model is 51.8 (0.1% lower) for T2V R@1 and 49.3 (1.6% lower) for V2T R@1. These...

> @nsreeprem The 0-shot performance I measured on MSRVTT by using the s2-1B model is 51.8 (0.1% lower) for T2V R@1 and 49.3 (1.6% lower) for V2T R@1. These results...