•⎳•
•⎳• @Tijmencrpt ·
#SN56 #Gradients #bittensor $TAO
lenno lenno @Lennobeer ·
The subnet i am talking about is Gradients (Sn56) I am going to explain why this subnet is massively underrated $TAO 🧵👇
403
Georg Martius
Georg Martius @GMartius ·
You are working on a simulation or generally code with control flow logic: you can probably make that yield informative gradients with our new softjax or softtorch libraries. #Gradients #JAX
Anselm Paulus Anselm Paulus @AnselmPaulus ·
Autodiff libraries compute gradients, but many operations (e.g. sorting, comparisons, logic, indexing) yield uninformative gradients. We release SoftJAX & SoftTorch, which bundle 40+ drop-in replacements that make these functions smoothly differentiable. github.com/a-paulus/softj…
1
490