evanatyourservice / kron_torchLinks
An implementation of PSGD Kron second-order optimizer for PyTorch
☆94Updated last week
Alternatives and similar repositories for kron_torch
Users that are interested in kron_torch are comparing it to the libraries listed below
Sorting:
- Efficient optimizers☆252Updated last week
- 🧱 Modula software package☆216Updated last week
- supporting pytorch FSDP for optimizers☆84Updated 7 months ago
- ☆150Updated 11 months ago
- The AdEMAMix Optimizer: Better, Faster, Older.☆184Updated 10 months ago
- ☆206Updated 8 months ago
- Getting crystal-like representations with harmonic loss☆192Updated 4 months ago
- WIP☆94Updated 11 months ago
- ☆65Updated 8 months ago
- DeMo: Decoupled Momentum Optimization☆190Updated 8 months ago
- Scalable and Performant Data Loading☆291Updated this week
- ☆100Updated 2 weeks ago
- Dion optimizer algorithm☆193Updated this week
- ☆83Updated last year
- ☆81Updated last year
- Explorations into the proposal from the paper "Grokfast, Accelerated Grokking by Amplifying Slow Gradients"☆101Updated 7 months ago
- Focused on fast experimentation and simplicity☆76Updated 7 months ago
- ☆115Updated last month
- NanoGPT-speedrunning for the poor T4 enjoyers☆68Updated 3 months ago
- $100K or 100 Days: Trade-offs when Pre-Training with Academic Resources☆143Updated 2 months ago
- Minimal (400 LOC) implementation Maximum (multi-node, FSDP) GPT training☆130Updated last year
- Fast, Modern, and Low Precision PyTorch Optimizers☆103Updated last week
- Official implementation of the paper: "ZClip: Adaptive Spike Mitigation for LLM Pre-Training".☆131Updated last month
- ☆27Updated last year
- Train a SmolLM-style llm on fineweb-edu in JAX/Flax with an assortment of optimizers.☆19Updated last week
- an open source reproduction of NVIDIA's nGPT (Normalized Transformer with Representation Learning on the Hypersphere)☆103Updated 5 months ago
- ☆304Updated last year
- Implementations of attention with the softpick function, naive and FlashAttention-2☆81Updated 3 months ago
- H-Net Dynamic Hierarchical Architecture☆65Updated 2 weeks ago
- CIFAR-10 speedruns: 94% in 2.6 seconds and 96% in 27 seconds☆274Updated 2 weeks ago