Unverified50% confidenceBenchmarkExact time
报告测出DiffusionGemma平均每步生成token数达到19.74个，而自回归方法每步只能生成1个
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/27/2026
First Seen
Valid until: 11/25/2026
Sources
Related Entities
Related Claims
Unverified在同一张英伟达H100显卡上，DiffusionGemma每秒可生成约1500个token，而上一代自回归模型仅能生成约300个，快了近5倍66% similarUnverified扩散模型推理时通过采样器算法通常只需20-30步迭代即可获得高质量图像生成结果，远少于训练时的1000步62% similarVerified如果每步准确率为95%，经过20步循环后整体准确率会降至约36%62% similarUnverified模型输入与输出Token的比例从4.52:1飙升至13:160% similarUnverifiedMeta的LLaMA论文披露其训练数据中约67%来自Common Crawl，经过多轮过滤后仅保留原始数据的约15%59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/807341API
curl https://kongchang.com/api/v1/knowledge/claims/807341MCP
get_claim(id=807341)