Unverified50% confidenceFactExact time
When individually tuning dedicated learning rates for each control group and repeating experiments with three random seeds, the advantage of repairing only the encoder disappears, reaching parity with the other approaches.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
Unverified即便固定所有随机种子，在不同硬件或并行度下，训练结果仍可能出现被大模型放大为显著能力差异的细微偏差75% similarUnverifiedThe encoder-only repair approach can reduce trainable parameters by 3× and optimizer-state memory by 3×.75% similarUnverifiedZero-shot prediction results are cached and reused across random seeds, so this part does not constitute truly independent repeated experiments.75% similarUnverifiedThe experiments used three training random seeds paired with a fixed test set split, along with a limited hyperparameter tuning budget.73% similarUnverifiedThe results of a single-task evaluation are subject to randomness, and the same model may produce noticeably different answers under different prompts and different random seeds.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/990510API
curl https://kongchang.com/api/v1/knowledge/claims/990510MCP
get_claim(id=990510)