ClashLuke / SOAPLinks
☆21Updated 8 months ago
Alternatives and similar repositories for SOAP
Users that are interested in SOAP are comparing it to the libraries listed below
Sorting:
- ☆34Updated 10 months ago
- Fast, Modern, and Low Precision PyTorch Optimizers☆103Updated last week
- ☆53Updated last year
- ☆83Updated last year
- supporting pytorch FSDP for optimizers☆84Updated 7 months ago
- research impl of Native Sparse Attention (2502.11089)☆60Updated 5 months ago
- ☆19Updated 2 months ago
- Minimal (400 LOC) implementation Maximum (multi-node, FSDP) GPT training☆130Updated last year
- Triton Implementation of HyperAttention Algorithm☆48Updated last year
- ☆81Updated last year
- H-Net Dynamic Hierarchical Architecture☆65Updated last week
- Explorations into the proposal from the paper "Grokfast, Accelerated Grokking by Amplifying Slow Gradients"☆101Updated 7 months ago
- ☆53Updated 10 months ago
- Implementation of Gradient Agreement Filtering, from Chaubard et al. of Stanford, but for single machine microbatches, in Pytorch☆25Updated 6 months ago
- Utilities for PyTorch distributed☆24Updated 5 months ago
- ☆49Updated last year
- Experiments for efforts to train a new and improved t5☆76Updated last year
- ☆20Updated 2 years ago
- Collection of autoregressive model implementation☆86Updated 3 months ago
- ☆100Updated 2 weeks ago
- Tiny re-implementation of MDM in style of LLaDA and nano-gpt speedrun☆55Updated 4 months ago
- Griffin MQA + Hawk Linear RNN Hybrid☆88Updated last year
- Demonstration that finetuning RoPE model on larger sequences than the pre-trained model adapts the model context limit☆63Updated 2 years ago
- Train a SmolLM-style llm on fineweb-edu in JAX/Flax with an assortment of optimizers.☆19Updated last week
- LayerNorm(SmallInit(Embedding)) in a Transformer to improve convergence☆59Updated 3 years ago
- Implementation of GateLoop Transformer in Pytorch and Jax☆89Updated last year
- Latent Diffusion Language Models☆68Updated last year
- GoldFinch and other hybrid transformer components☆46Updated last year
- An implementation of the Llama architecture, to instruct and delight☆21Updated 2 months ago
- Mixture of A Million Experts☆46Updated last year