chenpinganan
Results
3
comments of
chenpinganan
+1 for OpenBMB/MiniCPM-V-2
TensorrRT 8.4.12 not support bf16/uin8/int64/fp8 dtype, but TensorRT-LLM use them and not do code isolation by different cuda/tensortrt versions
TensorRT-LLM use some tensorrt API in runtimeBuffers.cpp & tllmRuntime.cpp, but this API only be supported by tensortrt9.1.0, not be supported in tensort8.4.1, such as class nvinfer1::ICudaEngine::getTensorDataType etc. so can not...