import pretrained model error
Hi, I am trying to run the video inference, but got an error when import the pretrained InternVideo2-Chat-8B, can you help me to fix this issue ? thanks !
Sorry, I moved the definition of the logger up one line. Please try to pull this code again. If possible, it is recommended that you install flash_attn.
@ynhe
Thank you so much! It's solved now, but I just encountered another issue related to Bert Tokenizer, and I tried to set the local_files_only=False but still not working, below is the the error message, can you help me to fix it ? thanks !
@daisyyayueyue Thank you for your feedback. This issue might be caused by the need to pull the Bert model from the local. I just revised the loading pipeline of BERT. If possible, you can try again.
awesome it worked ! I was able to get the full demo tested and got a chat response yesterday, but looks like there were some updates since then and it's giving me empty response when set {'do_sample':False} and throw
an error when set {'do_sample':True} , see the full error message below:
May I know if there were any warnings when you loaded the model and, if possible, I would like to know your PEFT version :). We can run with the PEFT version 0.5.0.
Here is the warning I got when load the model: "WARNING:transformers_modules.OpenGVLab.InternVideo2-Chat-8B.83a2c5753362b7b8b4042c8028c24be40ee54f48.modeling_internvideo2_vit:DropoutAddRMSNorm of flash_attn is not installed!!!", and I have flash_attn 2.6.3 installed and peft 0.12.0 installed. thank you so much for looking into this
Here is the warning I got when load the model: "WARNING:transformers_modules.OpenGVLab.InternVideo2-Chat-8B.83a2c5753362b7b8b4042c8028c24be40ee54f48.modeling_internvideo2_vit:DropoutAddRMSNorm of flash_attn is not installed!!!", and I have flash_attn 2.6.3 installed and peft 0.12.0 installed. thank you so much for looking into this
@daisyyayueyue
You need to downgrade peft to 0.5.0. For DropoutAddRMSNorm, this is not mandatory to install. You can refer to https://github.com/Dao-AILab/flash-attention/tree/main/csrc/layer_norm
Here is the warning I got when load the model: "WARNING:transformers_modules.OpenGVLab.InternVideo2-Chat-8B.83a2c5753362b7b8b4042c8028c24be40ee54f48.modeling_internvideo2_vit:DropoutAddRMSNorm of flash_attn is not installed!!!", and I have flash_attn 2.6.3 installed and peft 0.12.0 installed. thank you so much for looking into this
@daisyyayueyue You need to downgrade peft to
0.5.0. ForDropoutAddRMSNorm, this is not mandatory to install. You can refer to https://github.com/Dao-AILab/flash-attention/tree/main/csrc/layer_norm
it works now thank you! does this stage only takes visual info (i.e. audio info are not captured) ? will there be a new version have both visual and audio as inputs?
Certainly. We will support audio inputs along with other essential functionalities. We are planning to release an initial version that accommodates both audio and visual inputs in the near future.