Checking the Configurations
Model Configuration
- Check whether the Ascend platform supports model inference.
The Ascend platform, based on the MindIE inference framework, is compatible with various open-source large models. It is necessary to ensure that the model, device, and data types are all properly matched and meet the requirements. For details about MindIE, refer to What Is MindIE?. For the supported LLMs, see the "Supported Models > Large Language Models ".
- Check whether the model's loaded weights come from the same file. If the model weights are inconsistent, it will directly lead to variations in inference computation, and the inference accuracy will not align.
- Check the model configuration file to ensure that the parameter settings are consistent with the benchmark model configuration.
Common parameter configurations include pad_token_id, eos_token_id, max_sequence, and others. These configuration files are usually stored under the model weight path, and aligning the config files can resolve some accuracy issues.
Environment Configuration
If the model delivers normal accuracy in one environment but exhibits abnormal accuracy in other environments, or if accuracy issues arise after changing the software version, you need to check the environment configuration. Verify whether the versions of all components are consistent and compatible with the hardware environment. For version compatibility issues, For details, see section " Version Mapping" in MindStudio Release Notes..