Unverified50% confidenceFactExact time
The model is a Latent Flow Transformer with only 2.4 to 4 million parameters, quantized to int8 precision.
1
Sources
50%
Confidence
Long-term
Relevance
8/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDiT（Diffusion Transformer）架构的参数量已达数十亿级别，单次推理的中间激活值在高分辨率下可占满80GB A100的全部显存66% similarUnverifiedBlackwell's second-generation Transformer Engine adds FP4 precision support, which in theory can improve Transformer model training throughput by 2–4x compared to the H100.65% similarUnverifiedFlux is a new-generation model based on the Diffusion Transformer (DiT) architecture.63% similarUnverified现代商用NPU执行Transformer模型推理时的能效比可达GPU的5至10倍63% similarUnverifiedGoogle 2017年提出的Sparsely-Gated MoE首次将稀疏门控引入Transformer规模的序列模型，可在不显著增加计算量前提下将模型容量扩大1000倍以上61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/825586API
curl https://kongchang.com/api/v1/knowledge/claims/825586MCP
get_claim(id=825586)