打开微信,使用扫一扫进入页面后,点击右上角菜单,
点击“发送给朋友”或“分享到朋友圈”完成分享
MLU270 V1.7.0开发包 框架:Tensorflow
模型输入:1*1000*22
模型输出:1*4
生成离线模型时日志显示:
W tensorflow/stream_executor/mlu/mlu_api/fusion_op/fusion_op.cc:215][26383] Only one offline model is supported, but now there are multiple fusion segments. So only the offline model generated by the last fusion segment is saved.
offline_model.cambricon_twins文件里信息:Shape(dim): 1 256
This is the Twins File of "offline_model.cambricon". Function number: 1 ---------------kernel graph-------------------- Function #0 { Kernel num:1 Cache mode:0 Name: subnet0 Input number: 1 Input #0. Mask: 338036233 Shape(dim): 1 256 Name: Id: 216 Data type: CNRT_FLOAT32 Dim Order: CNRT_NHWC Quantize position: 1886413614 Quantize scale: -0.000000 Output number: 1 Output #0. Mask: 338036233 Shape(dim): 1 4 Name: Id: 223 Data type: CNRT_FLOAT32 Dim Order: CNRT_NHWC Quantize position: 32700
问题:
Shape(dim): 1 256为什么是1*256? 应该是1*1000*22才对;
请问这种情况应该如何处理?
热门帖子
精华帖子