Skip to content

Latest commit

 

History

History
15 lines (11 loc) · 986 Bytes

README.md

File metadata and controls

15 lines (11 loc) · 986 Bytes
Raw

Inference on MMedBench Testset

We provide the code for inference on MMedBench Testset. Note that currently MMedLM is a multilingual medical foundation model that hasn't undergone Supervised Fine-Tuning. You may need to first finetune the model with code in the finetune folder.

After finetuning, you need to change the following parameters in inference.sh before running it.

  • MODEL_NAME_OR_PATH : The folder containing the fintuned checkpoints of the model.
  • IS_LORA: If the model is fine-tuned with LoRA. If True, we will use AutoPeftModel to load the weight.
  • OUTPUT_DIR: The directory to save the response of the model.
  • IS_WITH_RATIONALE: If the model need to output the rationale of it's choice. If True, the prompt which instructs the model to output both selected option and rationale is applied. If False, model will output only the selected option, which is easier to calculate Accuracy.

Then you can start inference by

sbatch inference.sh