Implement Temperature Sampling in PyTorch

Easyllm-decodingauto-graded

Divide logits by a temperature scalar before softmax to sharpen or flatten the token probability distribution.

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("temperature-sampling", temperature_sample)

Company tags

How these tags are sourced