Unverified50% confidenceFactExact time
In the example step, top-k=3 raises the probability of 'moon' from 78.04% to 92.1%; top-p=0.9 keeps 8 tokens, but in another step where 'of' has a probability of 97.9%, it keeps only 1 token.
1
Sources
50%
Confidence
Long-term
Relevance
11/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn Qwen2.5-0.5B-Instruct, after the input prompt 'Write one sentence about the moon.', following 'The', the model scores 151,936 vocabulary tokens, among which 'moon' has a probability of 78.04%, 'full' 4.46%, 'vast' 2.24%, 'Moon' 1.51%, and 'lunar' 1.50%.67% similarVerifiedTop-P (nucleus sampling) samples from the smallest set of tokens whose cumulative probability exceeds the threshold P, while Top-K selects only from the K candidates with the highest probabilities.66% similarUnverifiedIn a moon landing question, gpt-5.4-mini gave P(someone lands on the moon) = 0.74 and P(no one lands on the moon) = 0.74, summing to 1.48.63% similarUnverifiedThe authors propose a top-k approximation method: consider only the k tokens with the highest probabilities and ensure the accuracy of the approximation by adjusting the mass of the tail probabilities.62% similarUnverifiedStepFun's Step 3.5 can reach an inference speed of six to seven hundred tokens per second, while by Step 5 the speed drops to under one hundred per second, yet capability has not improved noticeably.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1140783API
curl https://kongchang.com/api/v1/knowledge/claims/1140783MCP
get_claim(id=1140783)