Skip to content

Added auto transform to beam_search_decode_op #10286

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
May 4, 2018
Merged

Added auto transform to beam_search_decode_op #10286

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c1e044b
Added auto transform to beam_search_decode_op
Apr 28, 2018
27b6a67
Added some comment
Apr 28, 2018
35c6d22
Added unittest for beam_search_decode_op on GPU
May 4, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
58 changes: 51 additions & 7 deletions paddle/fluid/operators/beam_search_decode_op.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,25 +23,69 @@ struct BeamSearchDecodeFunctor {
BeamSearchDecodeFunctor(const LoDTensorArray& step_ids,
const LoDTensorArray& step_scores,
LoDTensor* id_tensor, LoDTensor* score_tensor)
: step_ids_(step_ids),
step_scores_(step_scores),
: step_ids_origin_(step_ids),
step_scores_origin_(step_scores),
id_tensor_(id_tensor),
score_tensor_(score_tensor) {}
score_tensor_(score_tensor) {
tensor_on_gpu_ = false;
// First make a copy of GPU data on CPU
if (platform::is_gpu_place(step_ids_origin_[0].place())) {
tensor_on_gpu_ = true;
platform::DeviceContextPool& pool =
platform::DeviceContextPool::Instance();
auto* dev_ctx = pool.Get(step_ids_origin_[0].place());
// Copy all tensors in the input tensor array
for (auto& step_id : step_ids_origin_) {
framework::LoDTensor out;
dev_ctx->Wait();
framework::TensorCopy(step_id, platform::CPUPlace(), *dev_ctx, &out);
dev_ctx->Wait();

out.set_lod(step_id.lod());
step_ids_.push_back(out);
}
}
if (platform::is_gpu_place(step_scores_origin_[0].place())) {
tensor_on_gpu_ = true;
platform::DeviceContextPool& pool =
platform::DeviceContextPool::Instance();
auto* dev_ctx = pool.Get(step_scores_origin_[0].place());
// Copy all tensors in the input tensor array
for (auto& step_score : step_scores_origin_) {
framework::LoDTensor out;
dev_ctx->Wait();
framework::TensorCopy(step_score, platform::CPUPlace(), *dev_ctx, &out);
dev_ctx->Wait();

out.set_lod(step_score.lod());
step_scores_.push_back(out);
}
}
}

template <typename T>
void operator()() const;

const LoDTensorArray& step_ids_;
const LoDTensorArray& step_scores_;
bool tensor_on_gpu_;
const LoDTensorArray& step_ids_origin_;
const LoDTensorArray& step_scores_origin_;
LoDTensorArray step_ids_ = LoDTensorArray();
LoDTensorArray step_scores_ = LoDTensorArray();
LoDTensor* id_tensor_;
LoDTensor* score_tensor_;
};

template <typename T>
void BeamSearchDecodeFunctor::operator()() const {
BeamSearchDecoder<T> beam_search_decoder;
beam_search_decoder.PackAllSteps(step_ids_, step_scores_, id_tensor_,
score_tensor_);
// Check if the tensor is on GPU. If so, use the CPU copy instead
if (tensor_on_gpu_) {
beam_search_decoder.PackAllSteps(step_ids_, step_scores_, id_tensor_,
score_tensor_);
} else {
beam_search_decoder.PackAllSteps(step_ids_origin_, step_scores_origin_,
id_tensor_, score_tensor_);
}
}

template <>
Expand Down
12 changes: 9 additions & 3 deletions python/paddle/fluid/tests/unittests/test_beam_search_decode_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,12 +22,12 @@
class TestBeamSearchDecodeOp(unittest.TestCase):
def setUp(self):
self.scope = core.Scope()
self.cpu_place = core.CPUPlace()
self.place = core.CPUPlace()

def append_lod_tensor(self, tensor_array, lod, data):
lod_tensor = core.LoDTensor()
lod_tensor.set_lod(lod)
lod_tensor.set(data, self.cpu_place)
lod_tensor.set(data, self.place)
tensor_array.append(lod_tensor)

def test_get_set(self):
Expand Down Expand Up @@ -71,7 +71,7 @@ def test_get_set(self):
SentenceIds="sentence_ids",
SentenceScores="sentence_scores")

beam_search_decode_op.run(self.scope, self.cpu_place)
beam_search_decode_op.run(self.scope, self.place)

expected_lod = [[0, 4, 8], [0, 1, 3, 6, 9, 10, 13, 16, 19]]
self.assertEqual(sentence_ids.lod(), expected_lod)
Expand All @@ -84,5 +84,11 @@ def test_get_set(self):
np.array_equal(np.array(sentence_scores), expected_data))


class TestBeamSearchDecodeOpGPU(TestBeamSearchDecodeOp):
def setUp(self):
self.scope = core.Scope()
self.place = core.CUDAPlace(0)


if __name__ == '__main__':
unittest.main()