LucasPrietoAl / grokking-at-the-edge-of-numerical-stabilityLinks
☆102Updated 2 months ago
Alternatives and similar repositories for grokking-at-the-edge-of-numerical-stability
Users that are interested in grokking-at-the-edge-of-numerical-stability are comparing it to the libraries listed below
Sorting:
- Token Omission Via Attention☆128Updated 11 months ago
- ☆82Updated last year
- Landing repository for the paper "Softpick: No Attention Sink, No Massive Activations with Rectified Softmax"☆84Updated 3 weeks ago
- ☆53Updated last year
- Explorations into the proposal from the paper "Grokfast, Accelerated Grokking by Amplifying Slow Gradients"☆102Updated 9 months ago
- EvaByte: Efficient Byte-level Language Models at Scale☆109Updated 5 months ago
- DeMo: Decoupled Momentum Optimization☆193Updated 10 months ago
- The simplest, fastest repository for training/finetuning medium-sized GPTs.☆164Updated 3 months ago
- Collection of autoregressive model implementation☆86Updated 5 months ago
- A byte-level decoder architecture that matches the performance of tokenized Transformers.☆66Updated last year
- Normalized Transformer (nGPT)☆191Updated 10 months ago
- Tree Attention: Topology-aware Decoding for Long-Context Attention on GPU clusters☆130Updated 10 months ago
- an open source reproduction of NVIDIA's nGPT (Normalized Transformer with Representation Learning on the Hypersphere)☆105Updated 6 months ago
- H-Net Dynamic Hierarchical Architecture☆80Updated 3 weeks ago
- A MAD laboratory to improve AI architecture designs 🧪☆129Updated 9 months ago
- Understand and test language model architectures on synthetic tasks.☆229Updated last week
- Attention Kernels for Symmetric Power Transformers☆120Updated last week
- PyTorch implementation of models from the Zamba2 series.☆185Updated 8 months ago
- Code to reproduce "Transformers Can Do Arithmetic with the Right Embeddings", McLeish et al (NeurIPS 2024)☆193Updated last year
- RWKV-7: Surpassing GPT☆95Updated 10 months ago
- Pytorch implementation of the PEER block from the paper, Mixture of A Million Experts, by Xu Owen He at Deepmind