[Inference] Fix multibatch inference

<!-- Demo: https://github.com/PaddlePaddle/PaddleNLP/pull/26 --> #### Before submitting - [x] Lint code. If there are lint issues, please format the code first. ```shell # Install and register `pre-commit` in the project folder pip install pre-commit && pre-commit install # Process previous code files separately pre-commit run --file XXXX.py ``` - [ ] Add test cases into `tests` folder. If there are codecov issues, please add tests cases first. ### PR types <!-- One of [ New features | Bug fixes | Function optimization | Performance optimization | Breaking changes | Others ] --> Bug fixes ### PR changes <!-- One of [ Models | APIs | Docs | Others ] --> Others ### Description <!-- Describe what this PR does --> Fix multibatch infra for Qwen2