Bowen Yuan

icon location Ph.D student, focusing on image and video understanding, i.e., visual question answering, video question answering, etc.

Results 9 comments of Bowen Yuan

Some files for VQA v2 may be missing..

Sorry for disturbing you, I find it in LXMERT. : )

Llava-video slowfast mode

I'm concerned about that, too. I find the following statements in the paper: "_We consider the same video representation configurations for the training and inference stages. On 128 NVIDIA H100...

Llava-video slowfast mode

Also, I'm curious why not use it on the 7B model.

Llava-video slowfast mode

> > Maybe the slowfast mode also used in the 72B model's training stage, instead of only for 72B model's inference stage? > > Based on the config.json of 72B...

Do not support for finetuning on only one dataset..

besides, why i have some unexpected key when loading the model like this: `unexpected_keys=['llama_model.model.layers.0.self_attn.rotary_emb.inv_freq', 'llama_model.model.layers.1.self_attn.rotary_emb.inv_freq', 'llama_model.model.layers.2.self_attn.rotary_emb.inv_freq', 'llama_model.model.layers.3.self_attn.rotary_emb.inv_freq',.....................`

Questions about VideoChat2_HD

Hi, I found `shared_utils_ds.py` has a bug in line 58. `optimizer_params = create_optimizer(config.optimizer, model, return_group=True)` the `optimizer.py` may need to be updated.

llava-video使用llms-eval测试出错

Hi, I'm trying it but found an error when using function `read_video_pyav()`. 最新版本的lmms-eval中的 `read_video_pyav()` 似乎仍然没有更新 具体错误： `read_video_pyav() got an unexpected keyword argument 'force_sample'`

llava-video使用llms-eval测试出错

我把`video_decode_backend: str = "pyav"` 默认值暂时换成了`video_decode_backend: str = "decord"`，可以跑通

llava-video使用llms-eval测试出错

我在用于训练的scripts里看到下面的设置： ``` --add_time_instruction True \ --force_sample True \ ``` 这两个值是不是默认也需要改为True