KellerJordan / Muon
Muon optimizer for neural networks: >30% extra sample efficiency, <3% wallclock overhead
☆69Updated this week
Related projects ⓘ
Alternatives and complementary repositories for Muon
- ☆122Updated this week
- Scaling is a distributed training library and installable dependency designed to scale up neural networks, with a dedicated module for tr…☆46Updated 2 weeks ago
- The simplest, fastest repository for training/finetuning medium-sized GPTs.☆84Updated last week
- code for training & evaluating Contextual Document Embedding models☆93Updated this week
- Minimal (400 LOC) implementation Maximum (multi-node, FSDP) GPT training☆112Updated 6 months ago
- A MAD laboratory to improve AI architecture designs 🧪☆95Updated 6 months ago
- ☆72Updated 4 months ago
- ☆49Updated 7 months ago
- Experiments for efforts to train a new and improved t5☆76Updated 6 months ago
- Latent Diffusion Language Models☆67Updated last year
- ☆53Updated 9 months ago
- Scalable neural net training via automatic normalization in the modular norm.☆119Updated 2 months ago
- One Initialization to Rule them All: Fine-tuning via Explained Variance Adaptation☆29Updated 3 weeks ago
- ☆76Updated 6 months ago
- Collection of autoregressive model implementation☆66Updated last week
- Experiment of using Tangent to autodiff triton☆72Updated 9 months ago
- Understand and test language model architectures on synthetic tasks.☆161Updated 6 months ago
- ☆50Updated 5 months ago
- Tree Attention: Topology-aware Decoding for Long-Context Attention on GPU clusters☆104Updated last month
- ☆50Updated last week
- ☆43Updated 2 months ago
- ☆61Updated 2 months ago
- Explorations into the proposal from the paper "Grokfast, Accelerated Grokking by Amplifying Slow Gradients"☆84Updated 2 months ago
- Efficient optimizers☆58Updated this week
- Code to reproduce "Transformers Can Do Arithmetic with the Right Embeddings", McLeish et al (NeurIPS 2024)☆177Updated 5 months ago
- ☆55Updated 11 months ago
- ☆27Updated 4 months ago
- ☆18Updated last month
- Pytorch implementation of the PEER block from the paper, Mixture of A Million Experts, by Xu Owen He at Deepmind☆111Updated 2 months ago
- ☆46Updated last month