After successfully training a few conditioned models on google colab, I took a little time to make a clean notebook based on the aidadsp original. Feel free to try it and report bugs or make suggestions.
You may need to buy comput units in order to complete the training process. Set the environnement to “High RAM” to prevent a crash during the file parsing step (2).
I made a less complex version of the trainer that will allow free google accounts to train simple conditioned models with 1 param. It needs only four reamped files and limited epochs. The result is pretty decent.
Simple conditioned model trainer
Here is a model quickly made for testing purpose. Mesa DC-5 ch1l
Hello. Do you think the conditioned trainer is better than the standard one?
In terms of sound rendering, they are identical. The main advantage of this version is that it allows modeling the behavior of a gain potentiometer or a tone potentiometer, for example. Otherwise, the training parameters are identical.
Thank you very much for the information.
I had some mixed results with high gain amp and pedals.
If you are using the official trainer, once in google collab, edit the config file of your choice (8, 12, 16 or 20) and replace “loss_functions” with “loss_fcns”. You can try “ESRPre” instead of ESR. For “pre_filter” you can try A-weighting instead of A-weighting-lp (low pass filter). Or more radically you can chose “none”.
Wow I didn’t know that, Thanks.
Sorry for the noob question. Where is the config file on the Google Colab trainer? I couldn’t locate it.
When you load the notebook on google collab you have to run the first steps in order for the environnement to download the project files. Once it is done use the file browser on the left pane to navigate into the Configs folder and edit the config files.