thinking-machines-lab / manifoldsLinks
Supporting code for the blog post on modular manifolds.
☆39Updated this week
Alternatives and similar repositories for manifolds
Users that are interested in manifolds are comparing it to the libraries listed below
Sorting:
- supporting pytorch FSDP for optimizers☆84Updated 9 months ago
- ☆120Updated 3 months ago
- ☆89Updated last year
- WIP☆93Updated last year
- Accelerated First Order Parallel Associative Scan☆188Updated last year
- 📄Small Batch Size Training for Language Models☆62Updated last week
- ☆67Updated 10 months ago
- ☆58Updated 11 months ago
- ☆30Updated 10 months ago
- ☆215Updated 10 months ago
- The simplest, fastest repository for training/finetuning medium-sized GPTs.☆164Updated 3 months ago
- Simple implementation of muP, based on Spectral Condition for Feature Learning. The implementation is SGD only, dont use it for Adam☆85Updated last year
- Minimal (400 LOC) implementation Maximum (multi-node, FSDP) GPT training☆132Updated last year
- research impl of Native Sparse Attention (2502.11089)☆61Updated 7 months ago
- ☆40Updated 3 weeks ago
- DeMo: Decoupled Momentum Optimization☆193Updated 10 months ago
- Focused on fast experimentation and simplicity☆75Updated 9 months ago
- Efficient optimizers☆265Updated last week
- JAX bindings for Flash Attention v2☆92Updated 3 weeks ago
- Pytorch implementation of the PEER block from the paper, Mixture of A Million Experts, by Xu Owen He at Deepmind☆128Updated last year
- Normalized Transformer (nGPT)☆191Updated 10 months ago
- A library for unit scaling in PyTorch☆130Updated 2 months ago
- Minimal but scalable implementation of large language models in JAX☆35Updated last month
- Code for the paper "Function-Space Learning Rates"☆23Updated 4 months ago
- FlashRNN - Fast RNN Kernels with I/O Awareness☆98Updated 3 months ago
- Explorations into the recently proposed Taylor Series Linear Attention☆100Updated last year
- Tiny re-implementation of MDM in style of LLaDA and nano-gpt speedrun☆56Updated 6 months ago
- CIFAR-10 speedruns: 94% in 2.6 seconds and 96% in 27 seconds☆301Updated 2 months ago
- 🧱 Modula software package☆277Updated last month
- H-Net Dynamic Hierarchical Architecture☆80Updated 3 weeks ago