Watch logits become a probability distribution through softmax, reshape with temperature, and collapse to a single sampled word.
Được dịch máy từ bản gốc tiếng Anh. Mở bản gốc tiếng Anh
For the continuation of "Once upon a ___", the model produces a score (a logit) for every word in its vocabulary. Here we line up only the leading candidates as pillars of light. Each pillar's height is its logit. You can see "time" standing out above the rest.
Look at the pillar heights and find which word's logit stands out.
Sharpness (entropy) 19%. 0% is a single point, 100% is a uniform distribution.
