Unverified50% confidenceFactExact time
SDXL将U-Net参数量提升至35亿，采用双文本编码器（OpenCLIP ViT-bigG和CLIP ViT-L）以及两阶段生成架构（base + refiner）
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSDXL的参数量从SD1.5的约0.9B增长到6.6B63% similarUnverifiedvLLM通过PagedAttention技术使同等硬件下可并发服务用户数提升2-4倍，吞吐量提升最高24倍59% similarUnverifiedSD 1.5 模型体积约 2GB，是最成熟的早期架构，生态插件最丰富但分辨率和细节表现有上限58% similarUnverifiedSDXL 基础模型仅加载就需要 6-8GB 显存58% similarUnverified自带双2.5GbE网口和两个M.2 NVMe插槽，硬件扩展性在同价位4盘位机型中属于较高水平，适合希望未来升级网络或用SSD做缓存加速的用户57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/895246API
curl https://kongchang.com/api/v1/knowledge/claims/895246MCP
get_claim(id=895246)