Skip to content
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

[GPU] Fix beam search accuracy/functional errors #27833

Merged
merged 2 commits into from
Dec 4, 2024
Merged

[GPU] Fix beam search accuracy/functional errors #27833

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
14f77ad
[GPU] Fix beam search accuracy/functional errors
sshlyapn Nov 29, 2024
2e3eceb
Merge branch 'master' into fix_beam_search
ceciliapeng2011 Dec 4, 2024
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
42 changes: 28 additions & 14 deletions src/plugins/intel_gpu/src/graph/impls/ocl/scaled_dot_product_attention.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,20 +57,34 @@ struct scaled_dot_product_attention_impl : multi_stage_primitive<scaled_dot_prod

protected:
std::vector<layout> get_internal_buffer_layouts_impl() const override {
// TODO: current implementation is supposed to have the same kernel version for both indirect/default paths,
// considering this, we may assume that both indirect/default kernels have absolutely the same intermediate
// buffers number and its' sizes (since update_dispatch_data is called for both kernels too), and
// do not double memory allocations during reallocate_if_needed() function call
// Look for the first sdpa_opt kernel entry. Currently, it can be used as default sdpa, indirect sdpa, or for both default
// and indirect cases. All of sdpa_opt kernels use the same internal buffers, so we can find the first sdpa_opt and
// use its` internal buffers configuration. The following scenarios are possible:
// 1) _kernels_data[0] - micro_sdpa (default)
// => internal buffers are not needed
// 2) _kernels_data[0] - sdpa_opt (default)
// => use internal buffers from [0] kernel
// 2) _kernels_data[0] - sdpa_opt (default)
// _kernels_data[1] - sdpa_opt (indirect)
// => use internal buffers from [0] kernel
// 3) _kernels_data[0] - micro_sdpa (default)
// _kernels_data[1] - sdpa_opt (indirect)
// => use internal buffers from [1] kernel
size_t kernel_idx = _kernels_data.size();
if (_kernels_data.size() >= 1 && !_kernels_data[0].internalBufferSizes.empty()) {
kernel_idx = 0;
} else if (_kernels_data.size() >= 2 && !_kernels_data[1].internalBufferSizes.empty()) {
kernel_idx = 1;
}

std::vector<layout> layouts;
for (size_t i = 0; i < _kernels_data.size(); i++) {
if (!_kernels_data[i].internalBufferSizes.empty()) {
auto dtype = from_data_type(_kernels_data[i].internalBufferDataType);
const auto bpp = data_type_traits::size_of(dtype);
for (auto size : _kernels_data[i].internalBufferSizes) {
layout inbuf_layout = {dtype, format::bfyx, // simple linear format (flattern to x channel)
{1, 1, 1, (tensor::value_type)(size / bpp)}};
layouts.push_back(inbuf_layout);
}
if (kernel_idx < _kernels_data.size()) {
auto dtype = from_data_type(_kernels_data[kernel_idx].internalBufferDataType);
const auto bpp = data_type_traits::size_of(dtype);
for (auto size : _kernels_data[kernel_idx].internalBufferSizes) {
layout inbuf_layout = {dtype, format::bfyx, // simple linear format (flattern to x channel)
{1, 1, 1, (tensor::value_type)(size / bpp)}};
layouts.push_back(inbuf_layout);
}
}

Expand Down Expand Up @@ -332,7 +346,7 @@ struct scaled_dot_product_attention_impl : multi_stage_primitive<scaled_dot_prod
}

if (indirect && has_indirect_inputs(impl_param)) {
params.beam_table.SetDynamicShapeOffset(get_beam_table_id(desc));
params.beam_table.SetDynamicShapeOffset(in_offsets_map.at(get_beam_table_id(desc)));
}

return params;
Expand Down
4 changes: 2 additions & 2 deletions src/plugins/intel_gpu/src/plugin/multi_tensor_variable_state.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,10 +87,10 @@ static void rearrange_cache(cldnn::memory::ptr kv_in_mem, cldnn::memory::ptr bt_
for (size_t f = 0; f < kv_shape[1]; f++) {
for (size_t y = 0; y < kv_shape[2]; y++) {
for (size_t x = 0; x < kv_shape[3]; x++) {
size_t b_kv = bt_in_ptr[b* kv_shape[concat_axis] + y];
auto out_idx = std::vector<int>{static_cast<int>(b), static_cast<int>(f), static_cast<int>(y), static_cast<int>(x)};

size_t b_kv = bt_in_ptr[b * kv_shape[concat_axis] + out_idx[concat_axis]]; // bt_idx = b * total_seq_len + seq_len_idx
auto in_idx = std::vector<int>{static_cast<int>(b_kv), static_cast<int>(f), static_cast<int>(y), static_cast<int>(x)};
auto out_idx = std::vector<int>{static_cast<int>(b), static_cast<int>(f), static_cast<int>(y), static_cast<int>(x)};

cldnn::tensor in(cldnn::format::bfyx, in_idx, 0);
cldnn::tensor out(cldnn::format::bfyx, out_idx, 0);
Expand Down
4 changes: 2 additions & 2 deletions src/plugins/intel_gpu/src/runtime/ocl/ocl_stream.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,7 +125,7 @@ void set_arguments_impl(ocl_kernel_type& kernel,
switch (scalar.t) {
case scalar_t::UINT8:
status = kernel.setArg(i, scalar.v.u8);
GPU_DEBUG_TRACE_DETAIL << "kernel: " << kernel.get() << " set scalar " << i << " (u8): " << scalar.v.u8 << "\n";
GPU_DEBUG_TRACE_DETAIL << "kernel: " << kernel.get() << " set scalar " << i << " (u8): " << static_cast<int>(scalar.v.u8) << "\n";
break;
case scalar_t::UINT16:
status = kernel.setArg(i, scalar.v.u16);
Expand All @@ -141,7 +141,7 @@ void set_arguments_impl(ocl_kernel_type& kernel,
break;
case scalar_t::INT8:
status = kernel.setArg(i, scalar.v.s8);
GPU_DEBUG_TRACE_DETAIL << "kernel: " << kernel.get() << " set scalar " << i << " (s8): " << scalar.v.s8 << "\n";
GPU_DEBUG_TRACE_DETAIL << "kernel: " << kernel.get() << " set scalar " << i << " (s8): " << static_cast<int>(scalar.v.s8) << "\n";
break;
case scalar_t::INT16:
status = kernel.setArg(i, scalar.v.s16);
Expand Down
4 changes: 4 additions & 0 deletions src/plugins/intel_gpu/tests/functional/subgraph_tests/dynamic/kv_cache_sdpa.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -343,6 +343,10 @@ std::vector<Params> get_test_params() {
p.push_back({with_rearrange, with_mask, !with_scale, !causal, !compressed, 1, ov::element::Type_t::f16, 10, 4, 1, {0, 2, 1, 3}});
p.push_back({!with_rearrange, with_mask, !with_scale, !causal, !compressed, 1, ov::element::Type_t::f16, 10, 4, 1, {0, 2, 1, 3}});

// Beam search
p.push_back({with_rearrange, !with_mask, !with_scale, !causal, !compressed, 2, ov::element::Type_t::f16, 10, 4, 1, {0, 1, 2, 3}});
p.push_back({with_rearrange, !with_mask, !with_scale, !causal, !compressed, 4, ov::element::Type_t::f16, 5, 16, 1, {0, 2, 1, 3}});

// Compressed
p.push_back({with_rearrange, with_mask, !with_scale, !causal, compressed, 1, ov::element::Type_t::f16, 10, 1, 1, {0, 1, 2, 3}});
p.push_back({with_rearrange, with_mask, !with_scale, !causal, compressed, 1, ov::element::Type_t::f16, 10, 4, 1, {0, 2, 1, 3}});
Expand Down
Loading