Closed lum1n0us closed 4 months ago
Please follow the REAMD in the repo to convert the model. python -c 'import xfastertransformer as xft; xft.LlamaConvert().convert("${HF_DATASET_DIR}","${OUTPUT_DIR}")' You can replace the LlamaConvert with OptConvert.
Thanks. I have passed model conversion. But still being blocked by compiling error about missing headers
now, it is
src/utils/numa_allocator.cpp:17:10: fatal error: numa.h: No such file or directory 17 | #include
After using yum install numactl-devel
, it has been resolved.
Now the only problem is missing mkl.h
In file included from xFasterTransformer/src/searchers/sample_search.cpp:16: xFasterTransformer/src/utils/decoder_util.h:19:10: fatal error: mkl.h: No such file or directory 19 | #include
Recently, I met some problems when following step by step tutorial. Hope to get second look and clarification.
All below commands are running in a container from cesg-prc-registry.cn-beijing.cr.aliyuncs.com/xfastertransformer/xfastertransformer:dev-ubuntu22.04(as tutorial's suggestion)
There is actually no ./tools/opt_convert.py in https://github.com/intel/xFasterTransformer.git. Although I located another opt_convert.py in src/xfastertransformer/tools/opt_convert.py, it is also not working.
I've tried latest commit (3b2e8b15b24bc0d80f5a8c15c223b5f8f92e84fe), v1.5.0 tag and v1.4.0 tag. There are same compiling error in all three: