Skip to content

Commit

Permalink
delete debug prints
Browse files Browse the repository at this point in the history
  • Loading branch information
@itikhono
itikhono committed Dec 17, 2024
1 parent bc54ab8 commit 879ae7a
Show file tree
Hide file tree
Showing 9 changed files with 24 additions and 59 deletions.
6 changes: 0 additions & 6 deletions ...mon/transformations/src/transformations/sdpa_to_paged_attention/position_ids_replacer.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,6 @@ ov::pass::PositionIDsReplacer::PositionIDsReplacer(const Output<Node>& position_
auto add = pattern::wrap_type<v1::Add>({mul, position_embed});

ov::matcher_pass_callback callback = [=](ov::pass::pattern::Matcher& m) {
// std::cout << "XXXXXX PositionIDsReplacer" << std::endl;
const auto& pattern_map = m.get_pattern_value_map();
replace_node(pattern_map.at(position_ids_pattern).get_node_shared_ptr(), position_ids.get_node_shared_ptr());
return true;
Expand All @@ -62,7 +61,6 @@ ov::pass::PositionIDsReplacerQwen::PositionIDsReplacerQwen(const Output<Node>& p
const auto& pattern_map = m.get_pattern_value_map();
auto max_context_len = pattern_map.at(max_context_len_pattern).get_node_shared_ptr();
if (max_context_len->get_friendly_name() != "max_context_len") {
// std::cout << "XXXX return false;" << std::endl;
return false;
}

Expand All @@ -80,10 +78,6 @@ ov::pass::PositionIDsReplacerQwen::PositionIDsReplacerQwen(const Output<Node>& p
replace_node(slice_2, reshape);

gather->validate_and_infer_types();
/* std::cout << "slice_2 in(0) " << slice_2->input(0).get_partial_shape() << std::endl;
std::cout << "slice_2 out " << slice_2->output(0).get_partial_shape() << std::endl;
std::cout << "gather in " << gather->input(0).get_partial_shape() << std::endl;
std::cout << "gather out " << gather->output(0).get_partial_shape() << std::endl;*/
return true;
};

Expand Down
1 change: 0 additions & 1 deletion ...nsformations/src/transformations/sdpa_to_paged_attention/prev_sequence_length_pattern.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,6 @@ ov::pass::PrevSequenceLengthPattern::PrevSequenceLengthPattern(std::shared_ptr<o
auto seq = pattern::wrap_type<v8::Gather>({kv_shape, pattern::any_input(), pattern::any_input()});

ov::matcher_pass_callback callback = [=](ov::pass::pattern::Matcher& m) {
std::cout << "XXXXXX PrevSequenceLengthPattern" << std::endl;
// TODO: Check that seq has axis that really takes sequence len but not any other dimension -- use symbolics or
// look at the constant input
// Detect the case by taking initialization expression for ReadValue and compare it with the second gather index
Expand Down
17 changes: 7 additions & 10 deletions ...sformations/src/transformations/sdpa_to_paged_attention/total_sequence_length_pattern.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,6 @@ ov::pass::TotalSequenceLengthPattern::TotalSequenceLengthPattern(
ov::matcher_pass_callback callback = [=](ov::pass::pattern::Matcher& m) {
// TODO: Check that seq has axis that really takes sequence len but not any other dimension --
// use symbolic infra or look at the constant input
std::cout << "XXXXXX TotalSequenceLengthPattern" << std::endl;
const auto& pattern_map = m.get_pattern_value_map();

auto concat = ov::as_type_ptr<v0::Concat>(pattern_map.at(kv_concat).get_node_shared_ptr());
Expand Down Expand Up @@ -109,24 +108,22 @@ ov::pass::TotalSequenceLengthPatternQwen::TotalSequenceLengthPatternQwen(
const std::shared_ptr<ov::op::v0::Parameter>& max_context_len) {
MATCHER_SCOPE(TotalSequenceLengthPatternQwen);

auto kv_past = pattern::wrap_type<v6::ReadValue>({pattern::any_input()});
auto kv_gather = pattern::wrap_type<v8::Gather>({kv_past, pattern::any_input(), pattern::any_input()});
auto kv_shape = pattern::wrap_type<v3::ShapeOf>({kv_gather});
auto seq_past = pattern::wrap_type<v8::Gather>({kv_shape, pattern::any_input(), pattern::any_input()});
auto prev_max_seq_len = pattern::wrap_type<v0::Parameter>();
auto opt_convert_1 = pattern::optional<v0::Convert>(prev_max_seq_len);
auto opt_reshape_1 = pattern::optional<v1::Reshape>({opt_convert_1, pattern::any_input()});

auto input_ids = pattern::wrap_type<v0::Parameter>();
auto unsqueeze = pattern::wrap_type<v0::Unsqueeze>({input_ids, pattern::any_input()});
auto optional_reshape = pattern::optional<v1::Reshape>({unsqueeze, pattern::any_input()});
auto optional_convert = pattern::optional<v0::Convert>(optional_reshape);
auto kv_shape_current = pattern::wrap_type<v3::ShapeOf>({optional_convert});
auto opt_reshape_2 = pattern::optional<v1::Reshape>({unsqueeze, pattern::any_input()});
auto opt_convert_2 = pattern::optional<v0::Convert>(opt_reshape_2);
auto kv_shape_current = pattern::wrap_type<v3::ShapeOf>({opt_convert_2});
auto seq_current = pattern::wrap_type<v8::Gather>({kv_shape_current, pattern::any_input(), pattern::any_input()});

auto pattern_total_seq = pattern::wrap_type<v1::Add>({seq_current, seq_past});
auto pattern_total_seq = pattern::wrap_type<v1::Add>({seq_current, opt_reshape_1});

ov::matcher_pass_callback callback = [=](ov::pass::pattern::Matcher& m) {
// TODO: Check that seq has axis that really takes sequence len but not any other dimension --
// use symbolic infra or look at the constant input
std::cout << "XXXXXX TotalSequenceLengthPatternQwen" << std::endl;
const auto& pattern_map = m.get_pattern_value_map();
auto total_seq = pattern_map.at(pattern_total_seq).get_node_shared_ptr();

Expand Down
4 changes: 2 additions & 2 deletions src/core/include/openvino/core/version.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,8 +19,8 @@
* @brief Defines OpenVINO patch version
*/

#define OPENVINO_VERSION_MAJOR 2024
#define OPENVINO_VERSION_MINOR 5
#define OPENVINO_VERSION_MAJOR 2025
#define OPENVINO_VERSION_MINOR 0
#define OPENVINO_VERSION_PATCH 0

namespace ov {
Expand Down
4 changes: 2 additions & 2 deletions src/core/src/pass/sdpa_to_paged_attention.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,6 @@
#include "openvino/op/subtract.hpp"
#include "openvino/op/unsqueeze.hpp"
#include "openvino/pass/manager.hpp"
#include "openvino/pass/visualize_tree.hpp"
#include "transformations/sdpa_to_paged_attention/position_ids_replacer.hpp"
#include "transformations/sdpa_to_paged_attention/prev_sequence_length_pattern.hpp"
#include "transformations/sdpa_to_paged_attention/state_management_pattern.hpp"
Expand Down Expand Up @@ -128,9 +127,10 @@ bool ov::pass::SDPAToPagedAttention::run_on_model(const std::shared_ptr<ov::Mode
score_results,
m_use_block_indices_inputs,
m_use_score_outputs);
manager.register_pass<TotalSequenceLengthPatternQwen>(max_context_len);

manager.register_pass<PrevSequenceLengthPattern>(prev_max_seq_len, batch_dim);
manager.register_pass<TotalSequenceLengthPattern>(max_context_len);
manager.register_pass<TotalSequenceLengthPatternQwen>(max_context_len);
manager.register_pass<PositionIDsReplacer>(unsqueezed_position_ids->output(0));
manager.register_pass<PositionIDsReplacerQwen>(unsqueezed_position_ids->output(0));
manager.run_passes(model);
Expand Down
18 changes: 9 additions & 9 deletions src/core/src/pass/visualize_tree.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -216,7 +216,7 @@ static void collect_symbol_print_values(const std::shared_ptr<ov::Model>& m,
bool ov::pass::VisualizeTree::run_on_model(const std::shared_ptr<ov::Model>& f) {
RUN_ON_MODEL_SCOPE(VisualizeTree);

static const bool ovasp = true;
static const bool ovasp = ov::util::getenv_bool("OV_VISUALIZE_APPLY_SYMBOLIC_PROPAGATION");
if (ovasp) {
std::cerr << "Warning: OV_VISUALIZE_APPLY_SYMBOLIC_PROPAGATION enabled. ov::pass::SymbolicPropagation will be "
"triggered"
Expand Down Expand Up @@ -561,11 +561,11 @@ std::string ov::pass::VisualizeTree::get_attributes(std::shared_ptr<Node> node)
std::stringstream label;
label << "label=\"" << get_node_name(node);

static const bool nvtos = true;
static const bool nvtot = true;
static const bool nvtio = true;
static const bool nvtrti = true;
static const bool ovpvl = true;
static const bool nvtos = ov::util::getenv_bool("OV_VISUALIZE_TREE_OUTPUT_SHAPES");
static const bool nvtot = ov::util::getenv_bool("OV_VISUALIZE_TREE_OUTPUT_TYPES");
static const bool nvtio = ov::util::getenv_bool("OV_VISUALIZE_TREE_IO");
static const bool nvtrti = ov::util::getenv_bool("OV_VISUALIZE_TREE_RUNTIME_INFO");
static const bool ovpvl = ov::util::getenv_bool("OV_VISUALIZE_PARTIAL_VALUES_AND_LABELS");

if (nvtos || nvtot || nvtio) {
if (nvtio) {
Expand Down Expand Up @@ -618,7 +618,7 @@ std::string ov::pass::VisualizeTree::get_attributes(std::shared_ptr<Node> node)
}

std::string ov::pass::VisualizeTree::get_node_name(std::shared_ptr<Node> node) {
static const bool nvtmn = true;
static const bool nvtmn = ov::util::getenv_bool("OV_VISUALIZE_TREE_MEMBERS_NAME");
std::string rc = (nvtmn ? std::string("friendly_name: ") : "") + node->get_friendly_name();
if (node->get_friendly_name() != node->get_name()) {
rc += "\\n" + (nvtmn ? std::string("name: ") : "") + node->get_name();
Expand All @@ -627,7 +627,7 @@ std::string ov::pass::VisualizeTree::get_node_name(std::shared_ptr<Node> node) {
rc += "\\n" + (nvtmn ? std::string("type_name: ") : "") + std::string(type_info.version_id) +
"::" + std::string(type_info.name);

static const bool nvttn = true;
static const bool nvttn = ov::util::getenv_bool("OV_VISUALIZE_TREE_TENSORS_NAME");
if (nvttn) {
auto to_string = [](const std::unordered_set<std::string>& names) {
std::stringstream ss;
Expand Down Expand Up @@ -665,7 +665,7 @@ std::string ov::pass::VisualizeTree::get_node_name(std::shared_ptr<Node> node) {
}
}

static const bool nvtrti = true;
static const bool nvtrti = ov::util::getenv_bool("OV_VISUALIZE_TREE_RUNTIME_INFO");
if (nvtrti) {
const auto& rt = node->get_rt_info();
if (!rt.empty()) {
Expand Down
4 changes: 0 additions & 4 deletions src/frontends/ir/src/frontend.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,8 +11,6 @@
#include "input_model.hpp"
#include "openvino/core/any.hpp"
#include "openvino/core/so_extension.hpp"
#include "openvino/pass/sdpa_to_paged_attention.hpp"
#include "openvino/pass/visualize_tree.hpp"
#include "openvino/runtime/aligned_buffer.hpp"
#include "openvino/runtime/shared_buffer.hpp"
#include "openvino/util/file_util.hpp"
Expand Down Expand Up @@ -304,8 +302,6 @@ std::string FrontEnd::get_name() const {
void FrontEnd::normalize(const std::shared_ptr<ov::Model>& model) const {
ov::pass::Manager manager("Frontend:IR:normalize");
manager.register_pass<pass::ResolveNameCollisions>();
// manager.register_pass<pass::VisualizeTree>("baichuan_sdpa.svg");
// manager.register_pass<pass::SDPAToPagedAttention>();
manager.run_passes(model);
}

Expand Down
22 changes: 3 additions & 19 deletions src/plugins/intel_cpu/src/nodes/kernels/scaled_attn/executor_pa.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1775,26 +1775,15 @@ struct AttentionExecutor : public PagedAttentionExecutor {
auto H = q.size(1) / S;
auto h_each_group_len = 1;
if (Hk != H) {
h_each_group_len = H / Hk;
h_each_group_len = H / Hk;
}
auto B_seq = past_lens.size(0);
std::cout << "1 assert " << std::endl;
q.assert_dims({B_token, H * S});

std::cout << "b token" << B_token << std::endl;
std::cout << "Hk " << Hk << std::endl;
std::cout << "S " << S << std::endl;
std::cout << "SV " << SV << std::endl;
std::cout << "3 assert " << std::endl;
v.assert_dims({B_token, Hk * SV});
std::cout << "2 assert " << std::endl;
q.assert_dims({B_token, H * S});
k.assert_dims({B_token, Hk * S});

std::cout << "4 assert " << std::endl;
v.assert_dims({B_token, Hk * SV});
q = q.reshape({B_token, H, 1, S});
std::cout << "5 assert " << std::endl;
k = k.reshape({B_token, Hk, 1, S});
std::cout << "6 assert " << std::endl;
v = v.reshape({B_token, Hk, 1, SV});
if (k_cache.m_dt == ov::element::Type_t::u8) {
k_cache.assert_dims({0, Hk, block_size, S + sizeof(float) * 2}, true);
Expand All @@ -1803,20 +1792,15 @@ struct AttentionExecutor : public PagedAttentionExecutor {
k_cache.assert_dims({0, Hk, block_size, S}, true);
v_cache.assert_dims({k_cache.m_dims[0], Hk, block_size, SV});
}
std::cout << "7 assert " << std::endl;
past_lens.assert_dims({B_seq});
std::cout << "8 assert " << std::endl;
subsequence_begins.assert_dims({B_seq + 1});
std::cout << "9 assert " << std::endl;
block_indices.assert_dims({0}, true);
std::cout << "10 assert " << std::endl;
block_indices_begins.assert_dims({B_seq + 1});
if (scale == 0.0f)
scale = 1.0f / sqrt(S);
if (alibi_slopes) {
alibi_slopes.assert_dims({H});
}
std::cout << "11 assert " << std::endl;
output_emb.assert_dims({B_token, H * SV});
output_emb = output_emb.reshape({B_token, 1, H * SV});

Expand Down
7 changes: 1 addition & 6 deletions src/plugins/intel_cpu/src/nodes/scaled_attn.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -973,11 +973,7 @@ struct ScaledDotProductAttention::AttentionExecutor : public ScaledDotProductAtt
SV = v_input.size(3);
L0 = present_key.size(2) - L1;
auto Hk = k_input.size(1);
std::cout << "B: " << B << std::endl;
std::cout << "Hk: " << Hk << std::endl;
std::cout << "S: " << S << std::endl;
std::cout << "L1: " << L1 << std::endl;
std::cout << "SV: " << SV << std::endl;

if (fuse_concat) {
k_input.assert_dims({B, Hk, L1, S});
v_input.assert_dims({B, Hk, L1, SV});
Expand Down Expand Up @@ -1199,7 +1195,6 @@ void ScaledDotProductAttention::createPrimitive() {
}

void ScaledDotProductAttention::execute(dnnl::stream strm) {
std::cout << "XXXXXXX SDPA 2" << std::endl;
auto orginSDPInputNumber = getOriginalInputsNumber() - (m_config.config.fuse_concat ? 3 : 0);
std::vector<MemoryPtr> inputs(orginSDPInputNumber);
auto output = getDstMemoryAtPort(0);
Expand Down

0 comments on commit 879ae7a

Please sign in to comment.