Unverified50% confidenceFactExact time
The method of directly averaging weights sometimes works when two models are fine-tuned from the same initialization, but typically yields poor-performing results for networks independently randomly initialized and trained from scratch.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe effectiveness of Task Arithmetic is premised on multiple fine-tuned models sharing the same pre-training starting point and cannot be directly applied to two networks independently trained from random initialization.76% similarUnverifiedThe study found that jointly training both directions within a single network outperforms training two separate specialized models.74% similarUnverified实验表明按由易到难顺序喂给模型训练样本能加速收敛并学到更鲁棒的特征表示，性能优于随机打乱顺序的训练73% similarUnverified对于MoE模型，路由层中两个专家得分极为接近时，微小数值差异可能翻转专家选择结果，导致分布式训练崩溃71% similarUnverifiedTraining from scratch means not using any pretrained weights, but instead completing the full training process on a new dataset with randomly initialized parameters.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/968584API
curl https://kongchang.com/api/v1/knowledge/claims/968584MCP
get_claim(id=968584)