Unverified50% confidenceFactExact time
Contrastive learning is the dominant paradigm for training text embedding models, and the InfoNCE loss function is the most commonly used training objective.
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
Related Claims
UnverifiedCLIP采用对比学习框架，使用InfoNCE损失函数，在互联网上约4亿图文对上进行大规模预训练，无需人工标注73% similarUnverified对比学习是 Sentence-BERT、OpenAI text-embedding 系列等主流 embedding 模型的核心训练范式70% similarUnverifiedPre-training data on the internet is nearly exhausted, and the contextual data generated by users in AI programming tools is the key fuel for training the next generation of models.70% similarUnverifiedDuring SFT training, the model only computes loss on the assistant portion (loss masking technique), meaning it only learns how to answer, not how to ask questions.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/758574API
curl https://kongchang.com/api/v1/knowledge/claims/758574MCP
get_claim(id=758574)