Implement Softmax from Scratch in PyTorch
Easyclassical-mlauto-graded
Build numerically stable softmax using the log-sum-exp trick, handling overflow and underflow in raw tensor math.
Solve it
Check your answer
The grader verifies properties of your implementation, so a correct solution written differently from ours still passes.
pip install torchleet
from torchleet import check
check("softmax", softmax)