OrgLion-ML
/
dlib
mirror of https://github.com/davisking/dlib.git
1
0
Fork 0
Code Issues Releases Wiki Activity

fix it by duplicating most code in send_job (works on my machine) 

I will probably need to find a way to reuse the code
This commit is contained in:
Adrià Arrufat 2021-09-26 12:15:02 +09:00
parent ceed02479e
commit 4851d54f0f
1 changed files with 39 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
dlib/dnn/trainer.h
View File

@ -1211,8 +1211,7 @@ namespace dlib
if (start < stop)
{
devices[i]->net.to_tensor(dbegin+start, dbegin+stop, job.t[i]);
if (job.labels[i].size() != 0)
job.labels[i].assign(lbegin+start, lbegin+stop);
job.labels[i].assign(lbegin+start, lbegin+stop);
job.have_data[i] = true;
}
else
@ -1238,8 +1237,44 @@ namespace dlib
data_iterator dend
)
{
typename std::vector<training_label_type>::iterator nothing;
send_job(test_only, dbegin, dend, nothing);
propagate_exception();
size_t num = std::distance(dbegin, dend);
size_t devs = devices.size();
job.t.resize(devs);
job.have_data.resize(devs);
job.test_only = test_only;
// chop the data into devs blocks, each of about block_size elements.
const double block_size = num / static_cast<double>(devs);
const auto prev_dev = dlib::cuda::get_device();
double j = 0;
for (size_t i = 0; i < devs; ++i)
{
dlib::cuda::set_device(devices[i]->device_id);
const size_t start = static_cast<size_t>(std::round(j));
const size_t stop = static_cast<size_t>(std::round(j + block_size));
if (start < stop)
{
devices[i]->net.to_tensor(dbegin+start, dbegin+stop, job.t[i]);
job.have_data[i] = true;
}
else
{
job.have_data[i] = false;
}
j += block_size;
}
DLIB_ASSERT(std::fabs(j - num) < 1e-10);
dlib::cuda::set_device(prev_dev);
job_pipe.enqueue(job);
}
void print_progress()