---
license: mit
tags:
- audio
- pytorch
- torchscript
- guitar-amp-simulation
- real-time
inference:
  framework: pytorch
  task: audio-to-audio
  inputs:
    - name: input
      type: float[]
      description: "Input waveform or features (e.g. [batch, channels, samples])"
  outputs:
    - name: output
      type: float[]
      description: "Output waveform or processed features"
---

## Usage
This is a model I trained to mimic a JCM 800 AMP. It doesn't sound very good, but as a first pass, I'm glad I have it.

![](infer.PNG)

Download [GuneAmp.exe](GuneAmp.exe) and try running your own conversion.

Read my notes [GuneAmpNotes](GuneAmpNotes.pdf)

## Using the TorchScript Model from Hugging Face

If you wish to use the TorchScript version of the model directly, you can download it from Hugging Face and load it using the following Python code.

First, ensure you have the necessary libraries installed:
```bash
pip install torch huggingface_hub
```

Then, use the following Python code to load and use the model:

```python
import torch
from huggingface_hub import hf_hub_download

model_id = 'sgune/gune-amp'
model_filename = 'metal_amp_v2_ts.pt'

model_path = hf_hub_download(repo_id=model_id, filename=model_filename)

#LOAD the model on GPU or CPU
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f"Loading model on device: {device}")

model = torch.jit.load(model_path, map_location=device)
model.eval()

print("Model loaded successfully!")

input_size = 1024
dummy_input = torch.randn(1, input_size, dtype=torch.float32).to(device)

print(f"Running inference with dummy input of shape: {dummy_input.shape}")

with torch.no_grad(): # Disable gradient calculations for inference
    output = model(dummy_input)

print("Inference complete!")
print("Example output shape:", output.shape)
print("Example output values:", output)
```


## COMING SOON
`infer.py`, `model.py`, `train.py` and `config.py` deepdives.