The Official Implementation of STA

1. Download the Dataset

Download datasets from HuggingFace:

• MOSEI
• SIMS-V2
• MELD
• CHERMA

Place them under the same folder, and set root_dataset_dir in parse_args of run.py to the dataset path.

2. Download the Backbone LLM

Download THUDM/chatglm3-6b and set pretrain_LM in parse_args of run.py to the LLM path.

If download is too slow, try Modelscope or HF-mirrors.

3. Acknowledgment

Our code is structurally referenced to MSE-Adapter and Self-MM. Thanks for their open-source spirit!