Unverified60% confidenceFactTime unknown
The original SAM version released by Meta in 2023 was trained on the SA-1B dataset of over 1 billion masks
2
Sources
60%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLlama 3于2024年发布，参数规格为8B、70B和405B，训练数据量超过15万亿tokens，采用分组查询注意力（GQA）技术61% similarVerifiedSAM (Segment Anything Model) is a foundation model released in 2023 capable of zero-shot segmentation of any object in an image; SAM 2, released in 2024, supports video segmentation.61% similarUnverified2021年的GLaM验证了64个专家的MoE架构在1.2万亿参数下的高效训练方法59% similarUnverifiedThe largest challenge number successfully factored to date is RSA-250 (829 binary bits), completed by a research team in 2020 using the equivalent of 2,700 core-years of computational resources.59% similarUnverifiedMasked autoencoders (MAE) is a self-supervised pretraining framework proposed by He et al. in 2022.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/10205API
curl https://kongchang.com/api/v1/knowledge/claims/10205MCP
get_claim(id=10205)