repoGitHubTrust 82 · PrimaryPublished 1mo agoLive · yesterday
lucidrains/torch-einops-utils
Some utility functions to help myself (and perhaps others) go faster with ML/AI work
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 62%Beyond Adam: SOAP and Muon for Faster, Label-Efficient Training of Machine Learning Interatomic Potentials →
- PossiblePossibly related (embedding) · 51%DeepSeek open-sources inference optimizations with 60–85% faster generation [pdf] →
- PossiblePossibly related (embedding) · 48%H64LM: A 249M-parameter Mixture-of-Experts Transformer built from scratch in PyTorch [P] →
- PossiblePossibly related (embedding) · 47%Optimize model training on Amazon SageMaker AI with NVIDIA Blackwell →
- PossiblePossibly related (embedding) · 52%TorchJD: Training with multiple losses in PyTorch [P] →
Implements
Covers
Covers (incoming)
newsMimo & deepseek are really amazing at optimizing ai. Read the the official blog page i linked, it will give amazing insight on how they pulled off this kind of low pricing with 2x - 3x profit margins.newsTorchJD: Training with multiple losses in PyTorch [P]newstried predicting which MoE experts get used next token to speed up cpu/gpu offload, got some real numbers, is this actually implementable or am i wasting my time (30tg/s -> 150-200tg/s)
Related across the graph
newsTorchJD: Training with multiple losses in PyTorch [P]newstried predicting which MoE experts get used next token to speed up cpu/gpu offload, got some real numbers, is this actually implementable or am i wasting my time (30tg/s -> 150-200tg/s)paperBeyond Adam: SOAP and Muon for Faster, Label-Efficient Training of Machine Learning Interatomic PotentialsnewsMimo & deepseek are really amazing at optimizing ai. Read the the official blog page i linked, it will give amazing insight on how they pulled off this kind of low pricing with 2x - 3x profit margins.newsOptimize model training on Amazon SageMaker AI with NVIDIA BlackwellnewsH64LM: A 249M-parameter Mixture-of-Experts Transformer built from scratch in PyTorch [P]newsDeepSeek open-sources inference optimizations with 60–85% faster generation [pdf]