Unverified50% confidenceFactExact time
The format instruction on the last line of the prompt is the key variable determining the output quality of small models: on math tasks, after unifying the last-line format, differences in question text style caused at most a 5-point fluctuation, while changing the format line could produce a 45-point gap.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn tests covering context lengths of 512–4096, increasing the draft length from 4 tokens to 16 tokens resulted in only a 0.2%–1.3% increase in overall inference latency.64% similarUnverifiedGemma 4 26B terminated with a runtime error before producing a score, because the scoring code assumed at least one sentence would be returned in the correct structured format.63% similarUnverifiednomic-embed-text嵌入模型有137M参数，all-MiniLM-L6-v2有22M参数，BGE-small为24M-109M参数59% similarUnverifiedThe author used the parameter --sref 2231018166 4162055066 at the end of the prompt.59% similarUnverifiedThe degree-3 and degree-4 polynomial coefficients in the paper are obtained by minimizing the maximum error (minimax approximation) in the BF16/FP16 target format.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1018307API
curl https://kongchang.com/api/v1/knowledge/claims/1018307MCP
get_claim(id=1018307)