I'm hoping to learn more by getting a discussion going about how some of you are capturing models / training your NAM files for maximum quality with respect to either outboard gear and / or software plugin FX (Chains).

Personally, I do not own a guitar, nor an amp/cab, so this use case is of little interest to me, but certainly feel free to discuss your methods related to that here as certainly there will be others who would benefit.

Currently, I'm doing most of my captures using the most recent publicly available input.wav trainer by Steve Atkinson and am merely training the models by using the method provided by neuralampmodeler.com.

Everything I do is in 48K, so I record, train, and work in 48k.

I do my best to reamp my training files at 0dBVU, but sometimes capture softer or harder. Sometimes, either/or turns out more pleasing, or achieves a much better ESR score. I tend to run 1000 epochs while training.

Apart from the virtually indistinguishable and incredible tone/sound, the reason why this is of special interest to me is because of the recent GPU Audio SDK that has come out, allowing an insane amount of instances of Neural Amp Modeler that is powered by the GPU. This has dramatically changed my workflow and approach.

This shouldn't be a thread about GPU Audio, though. There's already a dedicated thread about that here: GPU Audio SDK Released | 100+ Instances of Neural Amp Modeler on your GPU!

Are any of you using specialized files to "reamp" and train? Since I'm interested in using .nam files based on FX gear, such as preamps, eq's, compressors, etc, on virtually all types of audio, drums, bass, synths, mix bus, etc., Is there a better way to capture the action of these devices apart from the standard reamping file (input.wav)

Certainly this tech will be evolving for some time, so we can refer to this space for the most up to date techniques.

For now, let's hear your current techniques!