GEMS Webserver for Raman spectroscopy

Accelerate your spectral analysis with the GEMS foundation model. Please choose your desired workflow below:

Fine-tune (Left Panel): Upload your raw spectral data and labels. The system will automatically execute a complete AutoML pipeline before fine-tuning the pretrained GEMS model and generating comprehensive evaluation reports.
Predict (Right Panel): Upload a previously fine-tuned .pth model weight file alongside your unknown spectral data for instant classification and confident Top-5 logit predictions.

Fine-tune

Upload the raw training data and a pretrained model to generate a new classifier checkpoint and test results.

True Label Mapping File (optional, .json/.txt)

No file selected

Spectral (.npy)

No file selected

Labels (.npy)

No file selected

Wavenumbers (.npy)

No file selected

Pretrained Model (.pth) - Optional

No file selected

💡 Leave empty for the built-in GEMS Model. Upload to resume your checkpoint.

Epochs

Batch Size

Learning Rate

Weight Decay

Patience

Label Smoothing

Predict

Upload the fine-tuned classifier checkpoint exported after training, such as final_model.pth. Then upload the spectrum file you want to classify. If the spectrum file already includes wavenumbers, you can leave the optional wavelength file empty.

True Label Mapping File (optional, .json/.txt)

No file selected

Saved Model (.pth)

No file selected

Spectral (.npy/.txt/.csv)

No file selected

Wavelengths / Wavenumbers (optional, .npy/.txt/.csv)

No file selected

Manual wavelength range fallback (optional)

Low value

High value

Live Training Status

The results panel will appear here after you start a job.

Prediction Results

The results will appear here after you run prediction.