Unverified50% confidenceBenchmarkExact time
In tests on 2,456 held-out prompts, increasing the sampling steps from 20 to 50 improved FID only from 27.0 to 26.6 (a gain of 0.4), while removing the shift caused FID to rise from 27.0 to 27.3 and FD-DINOv2 to rise from 218 to 228.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn guided mode, the average end-to-end accuracy increased from 28.6% to 33.7%, and the ambiguity detection F1 score increased from 45.3% to 64.9%65% similarUnverifiedIn the queue admission control task, the RL policy improved from a baseline of −16.2 to 25.6.63% similarUnverifiedOn the AIME-2024 mathematical reasoning benchmark, pass@1 improved from 0.39 to 0.49.60% similarUnverifiedINT4量化将原本由65536个不同值（FP16）表示的权重压缩到仅16个离散级别60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902324API
curl https://kongchang.com/api/v1/knowledge/claims/902324MCP
get_claim(id=902324)