Unverified50% confidenceFactExact time
NVIDIA 的 Conformer 系列将卷积模块嵌入 Transformer 层以捕捉局部音频模式
1
Sources
50%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
VerifiedTransformer已从NLP渗透至计算机视觉（ViT）、语音识别（Whisper）、多模态模型（CLIP、GPT-4V）乃至蛋白质结构预测（AlphaFold2）66% similarUnverifiedNVIDIA Ampere 架构引入了异步内存拷贝和稀疏矩阵加速单元，Hopper 架构引入了 Transformer Engine 专用硬件66% similarUnverifiedNVIDIA随H100推出Transformer Engine，为每个张量维护历史激活值最大绝对值统计并动态调整缩放因子65% similarUnverifiedTransformers框架已支持数百种模型架构，覆盖NLP、CV、Audio、Multimodal等方向64% similarPartially VerifiedSora采用DiT（Diffusion Transformer）架构，将视频压缩为时空patch序列后输入Transformer，以Transformer替代传统U-Net作为去噪骨干网络63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/582181API
curl https://kongchang.com/api/v1/knowledge/claims/582181MCP
get_claim(id=582181)