Skip to content

Leaderboard

PyTorch

Explore public practice and performance rankings. Each board shows up to 25 entries.

How rankings work
  • The benchmark and solved-count boards include published problems in the selected mode from the main catalogue. Guided curriculum exercises are excluded, as are solves made after viewing any part of the reference solution.
  • A streak counts consecutive days when you solve an exercise, across all modes. Guided exercises and solves made with reference help count too. Days reset at midnight UTC. Your most recent solve must be today or yesterday to keep a streak active.
  • Boards refresh about once a minute, so a solve you just made can take a minute to appear.
  • Only public profiles appear. You can change your profile visibility in Settings.

Best benchmark ratios

lower ratio is faster

Each ratio compares a saved solution's runtime with the stored reference time for that exercise. A ratio of 0.50× means half the reference time; 1.00× means the same time. The lowest ratios come first. These are relative comparisons across different exercises, not a ranking by raw execution time.

One learner can appear for several exercises. Results depend on the code, hardware and software used; the rankings do not adjust for differences between machines.

  1. 1 banutha 0.71× Self-Attention Layer
  2. 2 Anupa Adikary 0.82× Masked Fill
  3. 3 Anupa Adikary 1.00× Tensor Indexing and Slicing
  4. 4 Anupa Adikary 1.98× Apply Along an Axis
  5. 5 banutha 2.10× Implement Scaled Dot-Product Attention
  6. 6 Anupa Adikary 2.42× Matrix Multiplication
  7. 7 Anupa Adikary 3.92× Concatenate Tensors
  8. 8 Anupa Adikary 4.18× SwiGLU Activation

Most solved

PyTorch
  1. 1 Anupa Adikary 6 problems solved
  2. 2 banutha 2 problems solved
  3. 3 testadmin 1 problem solved

Longest current streaks

all modes

No current public streaks. Solve a problem today to start one.