Conversation
add ${TVM_RUNTIME_LINKER_LIBS} in tvm_rpc
load from relay ir tune torch models using pickle mod and binary params
infer face images with different input name
export relay from onnx add export from paddlepaddle
use pickle relay mod export arm opencl export paddle ocr rec export to desired folder goon tune when break
add rk3588 opencl support
add float16 export and build
add script to start tracker and query add _opt for optimised export
make tsim correct
|
This is indeed an interesting change and it would be great if we could get this in. On the other hand, I would love to see the PR being organized a bit more :-) |
# Conflicts: # python/tvm/driver/build_module.py
|
Thank you! I 've already fixed conflicts I will learning how to make a better PR |
…r to compile ptx code remove debug log add --Wl,--exclude-all-symbols only in mingw compiler
|
Thank you so much! I'm happy to shepherd this PR :-) |
.gitignore
Outdated
| /python/images/ | ||
| /python/models/ | ||
| /python/depoly/ | ||
| /python/build/ |
There was a problem hiding this comment.
These lines may not be relevant - let's get them removed
|
Generally speaking, there are couple of rules we might want to follow to draft a PR:
Let's polish the code quality and get it in ASAP :-) |
# Conflicts: # python/auto_tune_utils.py # python/tune_config.py
CMakeLists.txt
Outdated
| MESSAGE(STATUS "c compiler : ${CMAKE_COMPILER_IS_GNUCC}") | ||
| message(STATUS "system name : ${CMAKE_SYSTEM_NAME}") |
There was a problem hiding this comment.
thanks for the PR! just wanted to point out that we have Summary.cmake: https://github.com/apache/tvm/blob/main/cmake/utils/Summary.cmake#L47 (which you can enable with -DSUMMARIZE=1 or set(SUMMARIZE ON) in config.cmake), so these info messages could be moved there
remove test scripts
|
Please continue in #13079 |
Add support for MinGW compiler on windows platform, so You can use std c++17 or LLVM higher version(13.x ...) without upgrade MSVC compilers, It's also easy to deploy on windows platform