E0202: input shape mismatch
What happened
A hardware backend received an input tensor with got features per timestep, but the model’s first layer expects expected features.
Why
The input was encoded or resized for a different model. Each compiled artifact has a fixed input dimension (stored in the first layer’s in_features). If the encoder was configured for a different model size, the input will not match.
How to fix
- Use
thrindex.encoders with the correct n_channels matching the model’s first layer.
- Check the artifact’s input dimension:
thrindex doctor --check model.thx prints Dense N → M for the first layer; N is the required input feature count.
- Confirm you are loading the correct artifact for your encoder configuration.
Example
