| # If it's full parameter training, use `--model xxx` instead of `--adapters xxx`. | |
| # If you are using the validation set for inference, add the parameter `--load_data_args true`. | |
| CUDA_VISIBLE_DEVICES=0 \ | |
| swift infer \ | |
| --adapters output/vx-xxx/checkpoint-xxx \ | |
| --stream true \ | |
| --temperature 0 \ | |
| --max_new_tokens 2048 | |