Feed Entities Podcasts Deep Dives Developers About

[KongchangAI]

Real-time AI knowledge base. Every claim is traceable to its source, with data freshness under 24 hours.

Powered by 240+ first-hand AI sources

Knowledge

Feed
Hot Entities
Podcasts

Developers

CLI
MCP Server
REST API
About

© 2026 KongchangAI. The freshest, most accurate AI knowledge

Unverified50% confidenceFactExact time

DINOv2 Giant参数量约11亿，嵌入维度为1536

1

Sources

50%

Confidence

Long-term

Relevance

7/10/2026

First Seen

Sources

DINOv2 vs SigLIP：k-NN分类为何差距悬殊？视觉编码器选型避坑指南

redditr/MachineLearning7/8/2026

Related Claims

UnverifiedDINOv2 Giant参数量约11亿、嵌入维度1536，均显著大于SigLIP2 SO400M（约4亿参数、维度1152）和CLIP ViT-L（约3亿参数、维度768），但k-NN任务表现最差85% similar UnverifiedDINOv2 Giant输出嵌入维度高达1536维，在仅175张训练样本下k-NN面临维度灾难挑战73% similar UnverifiedLLaMA-2 70B 总参数约 700 亿均匀分布于 80 层，每层约 8.75 亿参数，FP16 精度下单层约占 1.75GB 显存67% similar Partially Verified嵌入模型生成的向量通常为768维或1536维66% similar Unverified运行700亿参数级别开源模型在 FP16 精度下至少需要约 140GB 显存，即 8 张 H100 并行65% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/469428

API

curl https://kongchang.com/api/v1/knowledge/claims/469428

MCP

get_claim(id=469428)