Unverified50% confidenceFactExact time
Under a matched sampling budget, some heavily pruned models can match or even exceed the original model on pass@k, even though their pass@1 is lower.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe best-performing model achieved 38.7% on pass@3, but only 5.9% on pass^3.60% similarUnverifiedK折交叉验证的K通常取5或10，这在偏差-方差权衡上被实证为较优选择59% similarUnverifiedOn strong models such as Claude Opus 4.5, DeReAct did not improve Pass@1 but made trajectories more evidence-complete and more compliant with constraints.58% similarUnverifiedUnder four-view test-time augmentation, ensembling the flat model with the label-correlation model achieved the best configuration on six of eight metrics: exact match 0.694, micro-F1 0.678, macro-F1 0.456, and macro-AUC 0.972.57% similarUnverifiedUnder matched batch-size-1 decoding latency, LRCC simultaneously improves both perplexity and downstream accuracy on Llama-3.2-1B.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1129102API
curl https://kongchang.com/api/v1/knowledge/claims/1129102MCP
get_claim(id=1129102)