lucidrains / grokfast-pytorch
Explorations into the proposal from the paper "Grokfast, Accelerated Grokking by Amplifying Slow Gradients"
☆85Updated 2 months ago
Related projects ⓘ
Alternatives and complementary repositories for grokfast-pytorch
- ☆76Updated 7 months ago
- Normalized Transformer (nGPT)☆66Updated this week
- ☆128Updated this week
- Explorations into the recently proposed Taylor Series Linear Attention☆90Updated 3 months ago
- ☆53Updated 10 months ago
- Collection of autoregressive model implementation☆67Updated this week
- Muon optimizer for neural networks: >30% extra sample efficiency, <3% wallclock overhead☆109Updated last week
- Some personal experiments around routing tokens to different autoregressive attention, akin to mixture-of-experts☆108Updated last month
- Implementation of GateLoop Transformer in Pytorch and Jax☆86Updated 5 months ago
- Implementation of the proposed Spline-Based Transformer from Disney Research☆76Updated last week
- Just some miscellaneous utility functions / decorators / modules related to Pytorch and Accelerate to help speed up implementation of new…☆119Updated 3 months ago
- σ-GPT: A New Approach to Autoregressive Models☆59Updated 3 months ago
- Pytorch implementation of the PEER block from the paper, Mixture of A Million Experts, by Xu Owen He at Deepmind☆112Updated 2 months ago
- Efficient optimizers☆79Updated this week
- Latent Diffusion Language Models☆67Updated last year
- The simplest, fastest repository for training/finetuning medium-sized GPTs.☆84Updated last week
- Implementation of a Light Recurrent Unit in Pytorch☆46Updated last month
- ☆49Updated 8 months ago
- The official repository for HyperZ⋅Z⋅W Operator Connects Slow-Fast Networks for Full Context Interaction.☆31Updated 2 months ago
- Mobile Viewer for W&B, built on top of Flutter.☆30Updated 8 months ago
- Implementation of Infini-Transformer in Pytorch☆104Updated last month
- ☆39Updated 10 months ago
- ☆73Updated 4 months ago
- Unofficial but Efficient Implementation of "Mamba: Linear-Time Sequence Modeling with Selective State Spaces" in JAX☆79Updated 9 months ago
- Minimal (400 LOC) implementation Maximum (multi-node, FSDP) GPT training☆113Updated 7 months ago
- ☆57Updated 11 months ago
- Transformer with Mu-Parameterization, implemented in Jax/Flax. Supports FSDP on TPU pods.☆29Updated 2 weeks ago
- Implementation of the Llama architecture with RLHF + Q-learning☆157Updated 10 months ago
- Exploring an idea where one forgets about efficiency and carries out attention across each edge of the nodes (tokens)☆43Updated last month
- A MAD laboratory to improve AI architecture designs 🧪☆95Updated 6 months ago