Los-D
Shredder
- Messages
- 1,525
Hi @ArteraDSP, is it true you are woking on a possible hardware pedal version?
NAM: Neural Amp Modeler
- Thread starter James Freeman
- Start date
Dom Mcsweeney
Roadie
- Messages
- 147
New NAM Trainer
>>> https://www.tonezone3000.com/
I helped with testing and can vouch that it's a great alternative for folks who weren't that pleased using Google Colab, or hadn't yet jumped into the local NAM training solution. There are a few extra features built in that are really handy too.
Have a look and give it a try =)
>>> https://www.tonezone3000.com/
I helped with testing and can vouch that it's a great alternative for folks who weren't that pleased using Google Colab, or hadn't yet jumped into the local NAM training solution. There are a few extra features built in that are really handy too.
Have a look and give it a try =)
2dor
Roadie
- Messages
- 850
It's great!
I tried it and it performed roughly about the same as training on my local RTX 4070.
It allows tweaking the learning rate and learning rate decay which is great if you want to push for longer epochs / training times to squeeze out some lower ESRs.
I still don't know if there are any strings attached to it because it seems just too good to be true lol
Dom Mcsweeney
Roadie
- Messages
- 147
Awesome man! Yeah it's speedy, and lots of very useful options.
Re: 'free' -
I don't know too much on that, but I believe they have a vc backed studio working on ML/audio projects, which affords them access to lots of GPU resources. They are offering free resources to the NAM community - no strings attached there.
Re: 'free' -
I don't know too much on that, but I believe they have a vc backed studio working on ML/audio projects, which affords them access to lots of GPU resources. They are offering free resources to the NAM community - no strings attached there.
2dor
Roadie
- Messages
- 850
I did read that piece on the FB group but yeah - don't know what to make of it.
Anyway, it's great as long as it's available.
2dor
Roadie
- Messages
- 850
It's offline now due to high demand.
I don't know how anybody can afford to keep something like this for free; at least in the long term. Maybe including adds on the page like Tonehunt does now would help with the cost but I still think it may not be enough to break even. One thing I've learned over the years, there's no such thing as a free lunch.
![[Nathan]](/data/avatars/m/0/613.jpg?1726885163){kind=avatar}
[Nathan]
Roadie
- Messages
- 419
Hah that didn’t take long at all to jack up usage and cause a cutoff.
Dom Mcsweeney
Roadie
- Messages
- 147
Dom Mcsweeney
Roadie
- Messages
- 147
Regardless of how long they make it last, and I really do hope it lasts, it's an awesome example of what can be done and it's only just the beginning there. It's really important people do take these chances and show that it's possible to take the code and make cool products with it.
That it's not just the 'the open source NAM plugin' that can be used to make new products, but also the open source trainer is good to be made into a product too. And also if people do take that initiative it allows Steve to focus on improving the code, rather than spending a lot of time on supporting users. That part, we all do our best with, but Steve does a huge amount in his limited spare time and I think he'd love to be tinkering and innovating a lot more.
That it's not just the 'the open source NAM plugin' that can be used to make new products, but also the open source trainer is good to be made into a product too. And also if people do take that initiative it allows Steve to focus on improving the code, rather than spending a lot of time on supporting users. That part, we all do our best with, but Steve does a huge amount in his limited spare time and I think he'd love to be tinkering and innovating a lot more.
MirrorProfiles
Rock Star
- Messages
- 3,332
Was anything said about what happens with the models that are generated? Like who owns them, where they’re stored, what can be done with them etc?
I love the idea but with so much that seems too good to be true, I wonder what the catch is.
Seems like an amazing tool regardless. I hope it survives and copes well with the demand.
I love the idea but with so much that seems too good to be true, I wonder what the catch is.
Seems like an amazing tool regardless. I hope it survives and copes well with the demand.
2dor
Roadie
- Messages
- 850
Yeah - no word on it & that's exactly what I was wondering about. Anyway, that aside, takes 32 minutes for a 1000 epoch training run which is very fast - faster than what I can pull off with my local RTX 4070.
If it could do batch training, then it's going to be very hard to pass on it.
paisleywookiee
Rock Star
- Messages
- 5,443
Is there not a local way to train models?
2dor
Roadie
- Messages
- 850
There is but it's not that easy to get started and you need a good nvidia GPU to get decent training times.
paisleywookiee
Rock Star
- Messages
- 5,443
Ugh. No thanks.
Dom Mcsweeney
Roadie
- Messages
- 147
A great response from Woody (dev of ToneZone) to @2dor in the NAM Fb group. Thought I'd just add it here:
(Thanks 2dor for your interaction in the fb group as always)
2dor asked:
1. Will there be an option to do a "batch" training run? Say even if limited to batches of 10, 20 reamps for instance?
2. What happens to these trained models and the reamps? Is the data considered your IP, the owner's etc? Do you reserve the right to reshare / resell these models further?
3. You've probably seen there's quite a big demand for this platform (and for good reason - implementation is great & very fast training). How do you foresee maintenance / upkeep is going to look like?
Answers:
1. Yes batching is definitely on the roadmap, I knew that was going to be a # one request. it'd be cool to hear feedback from you/others on the ideal workflow for both batching and organizing training jobs as packs.
2. They are hosted on s3 right now. Only you and people you share the link with should have access - we don't have like a public search page or anything (but please share on tonehunt!!!). we aren't doing anything with the files other than hosting them for you. this reminds me to put up some kind of data privacy and terms pages for more transparency - will start working on that!
3. I'm cofounder of a vc backed studio which means resources to offer up this compute. if the tool is well liked and actually starts to grow at an unsustainable rate we'll probably need a "business model" like a pro tier or something to maintain it (don't quote me on that literally just making it up haha). for now i am focused on releasing this beta and gathering feedback from the community so I can keep iterating and hopefully makes something great that makes creating and contributing to the community as easy and accessible as possible.
(Thanks 2dor for your interaction in the fb group as always)
2dor asked:
1. Will there be an option to do a "batch" training run? Say even if limited to batches of 10, 20 reamps for instance?
2. What happens to these trained models and the reamps? Is the data considered your IP, the owner's etc? Do you reserve the right to reshare / resell these models further?
3. You've probably seen there's quite a big demand for this platform (and for good reason - implementation is great & very fast training). How do you foresee maintenance / upkeep is going to look like?
Answers:
1. Yes batching is definitely on the roadmap, I knew that was going to be a # one request. it'd be cool to hear feedback from you/others on the ideal workflow for both batching and organizing training jobs as packs.
2. They are hosted on s3 right now. Only you and people you share the link with should have access - we don't have like a public search page or anything (but please share on tonehunt!!!). we aren't doing anything with the files other than hosting them for you. this reminds me to put up some kind of data privacy and terms pages for more transparency - will start working on that!
3. I'm cofounder of a vc backed studio which means resources to offer up this compute. if the tool is well liked and actually starts to grow at an unsustainable rate we'll probably need a "business model" like a pro tier or something to maintain it (don't quote me on that literally just making it up haha). for now i am focused on releasing this beta and gathering feedback from the community so I can keep iterating and hopefully makes something great that makes creating and contributing to the community as easy and accessible as possible.
Los-D
Shredder
- Messages
- 1,525
Los-D
Shredder
- Messages
- 1,525
@ArteraDSP , ran into an issue when loading @2112 ’s LT TV Mix 2 IR on iOS (AUM crashed but I could hear reflections from the IR before… like along reverb/delay). Not sure if a local problem or something else.
IR can be found here:
forum.fractalaudio.com
IR can be found here:
Marshall TV Greenback IR's - Free Cab Pack
I took some test shots of my Marshall TV cab using 3 mics (57, SM7 & 121) and a Dynamount (mic robot) and made these 3 mixes that I'm pretty happy with. I can't remember the recipes but Mix 2 is my go to for anything from clean to mean. UPDATE - TV Mix 2 seems pretty popular so I've uploaded it...
2dor
Roadie
- Messages
- 850
NAM just went another level above:
www.neuralampmodeler.com
TONEZONE3000: Training Made Simple
NAM is the fastest, most accurate--and now (I'd argue), easiest--way to make a neural model of your gear.TONEZONE3000 is a third-party website that puts the standard snapshot training process in a streamlined, easy-to-use website. Like the GUI and Colab trainers I've developed, users can reamp a...
www.neuralampmodeler.com