ValueError: When load_4bit=True in from longva.model.builder.load_pretrained_model

[hungvtm]

Description
When using the load_pretrained_model function from longva.model.builder with load_4bit=True, the following error is encountered: ValueError: You can't pass `load_in_4bit` or `load_in_8bit` as a kwarg when passing `quantization_config` argument at the same time.

This issue occurs because the quantization_config argument is being set alongside the load_in_4bit argument, which leads to a conflict.

Proposed Solution
To resolve this issue, please remove the following line from the function:

# kwargs["load_in_4bit"] = True

This change will ensure that the conflicting argument is not set, preventing the ValueError.
This adjustment ensures that the function will work correctly when load_4bit=True.
Thank you for your attention to this issue.