热门标签
更多>
搜索结果
查询Tags标签: selfAttention,共有 2条记录-
小黑算法成长日记23:selfAttention与multiHeadAttention
SelfAttention操作 从单个字的角度: qi=hiWQ,kj=hjWK,vj=hjWVq_i = h_iW_Q,k_j = h_jW_K,v_j = h_jW_Vqi=hiWQ,kj=hjWK,vj=hjWV eij=qikjTe_{ij} = q_ik_j^Teij=qikjT αi=Softmax([ei,1,...,ei,T])\alpha_i = Softmax([e_{i,1},...,e_{i,T}])αi…
2022/1/2 12:37:34 人评论 次浏览 -
小黑算法成长日记23:selfAttention与multiHeadAttention
SelfAttention操作 从单个字的角度: qi=hiWQ,kj=hjWK,vj=hjWVq_i = h_iW_Q,k_j = h_jW_K,v_j = h_jW_Vqi=hiWQ,kj=hjWK,vj=hjWV eij=qikjTe_{ij} = q_ik_j^Teij=qikjT αi=Softmax([ei,1,...,ei,T])\alpha_i = Softmax([e_{i,1},...,e_{i,T}])αi…
2022/1/2 12:37:34 人评论 次浏览