Popular repositories Loading

1. dulub dulub Public

   Forked from SakaguchiAngosan/dulub

   TypeScript

2. dullub dullub Public

   TypeScript 1

3. KANX-tst KANX-tst Public

   Forked from DavidMacha/KANX-test

   Production-grade Kolmogorov-Arnold Networks TensorFlow + PyTorch + ONNX — one library, four surfaces. || 265× lower MSE than an MLP with 5× fewer parameters. One library. Two backends. Real ONNX ex…

   Python

4. KANX KANX Public

   Forked from Mattral/KANX

   One library, four surfaces. Production-grade Kolmogorov-Arnold Networks || TensorFlow + PyTorch + ONNX. || 265× lower MSE than an MLP with 5× fewer parameters. One library. Two backends. Real ONNX …

   Python

5. A-Composed-Mixture-of-Experts-Engine A-Composed-Mixture-of-Experts-Engine Public

   Forked from Mattral/Gesture_Arm

   hyperscale MoE training runtime. Designed to keep a 10,000-GPU pre-training job alive end-to-end: sparse Top-K routing in custom Triton, true 4-D parallelism (TP × PP × FSDP2 × EP) on PyTorch 2.12+…

   Python

6. Composed-Mixture-of-Experts-Engine Composed-Mixture-of-Experts-Engine Public

   Forked from SakaguchiAngosan/Composed-Mixture-of-Experts-Engine

   Production-grade sparse MoE training runtime. Designed to keep large-scale pre-training jobs alive end-to-end: sparse Top-K routing in custom Triton, DP+EP distributed training with TP support in c…

   Python