[Question]: How to compile lightseq

[FrostML]

I try to compile lightseq by using build.sh, but run into the following problem:

lightseq/csrc/proto/bert_weight.cc:451:15: error: ‘class Bert’ has no member named ‘ParseFromIstream’; did you mean ‘ParseFromString’?
     if (!bert.ParseFromIstream(&raw_input)) {
               ^~~~~~~~~~~~~~~~
               ParseFromString

lightseq/csrc/proto/bert_crf_weight.cc:38:37: error: no match for ‘operator[]’ (operand types are ‘const google::protobuf::RepeatedPtrField<BertCrfEncoderLayer>’ and ‘int’)
   _inner_size = bert.encoder_stack()[0].ffn_first_kernel_size() / _hidden_size;

Branch master and tag v3.0.1 both failed.

Did I miss something? How can I manage to compile this project?

[FrostML]

I try to follow the instruction of this doc docs/build.md and another error occur.

/usr/local/bin/ld: /usr/lib/x86_64-linux-gnu/hdf5/serial/libhdf5.a(H5.o): relocation R_X86_64_PC32 against symbol `stderr@@GLIBC_2.2.5' can not be used when making a shared object; recompile with -fPIC
/usr/local/bin/ld: final link failed: bad value
collect2: error: ld returned 1 exit status

HDF5 was installed by:

$ curl -O -L -C - https://github.com/HDFGroup/hdf5/archive/refs/tags/hdf5-1_12_0.tar.gz
$ tar -xzvf hdf5-1_12_0.tar.gz
$ cd hdf5-hdf5-1_12_0
$ ./configure --prefix=/usr/local/hdf5 "CFLAGS=-fPIC" "CXXFLAGS=-fPIC"
$ make
$ make install
$ cd ..

[Taka152]

Have you added the hdf5 path to $PATH before building?

[FrostML]

There might be something wrong with my env. I used docker container and fixed this error. Thanks a lot @Taka152 .

Two more questions, how can I reproduce this speedup of Transformer-Big mentioned in here. What are the batch size, sequence length or beam size you used?
And did you fix the GPU's frequency (Memory and Graphics clocks) when testing?

[Taka152]

1. We compared with huggingface model with batch_size [1, 8, 32], sequence length [16, 32, 64], and beam_size=4.
2. We didn't change the frequency.

[FrostML]

Thanks a lot.