happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
llama.cpp/ggml/src/ggml-openvino/utils.cpp

507 lines
21 KiB
C++
Raw Blame History

#include "utils.h"
#include <algorithm>
#include <cassert>
#include <cmath>
#include <cstddef>
#include <cstdint>
#include <cstdlib>
#include <iomanip>
#include <iostream>
#include <memory>
#include <mutex>
#include <openvino/core/any.hpp>
#include <openvino/core/graph_util.hpp>
#include <openvino/core/type/float16.hpp>
#include <openvino/frontend/manager.hpp>
#include <openvino/openvino.hpp>
#include <openvino/runtime/compiled_model.hpp>
#include <openvino/runtime/infer_request.hpp>
#include <openvino/runtime/intel_npu/properties.hpp>
#include <openvino/runtime/properties.hpp>
#include <openvino/runtime/tensor.hpp>
#include <unordered_map>
#include <vector>
#include "ggml-impl.h"
#include "ggml-openvino/ggml-decoder.h"
#include "ggml.h"
#include "openvino/frontend.hpp"
#include "openvino/input_model.hpp"
ov::Tensor convert_ggml_input_to_ov(std::shared_ptr<GgmlOvDecoder> ggml_decoder, const std::string& name) {
const auto* ggml_tensor = ggml_decoder->get_input_ggml_tensor(name);
auto* input_data = ggml_tensor->data;
ov::Shape input_shape;
if (name.find("cache_k") == 0 || name.find("cache_v") == 0) {
input_shape = ggml_decoder->get_graph_input_shape(ggml_tensor).to_shape();
} else if (ggml_tensor->op == GGML_OP_VIEW) {
// This case is added to make test-backend-ops work
input_shape = ggml_decoder->get_graph_input_shape(ggml_tensor->view_src).to_shape();
} else {
input_shape = ggml_decoder->get_input_shape(name).to_shape();
}
auto input_tensor = ov::Tensor(ggml_decoder->get_input_type(name), input_shape, input_data);
return input_tensor;
}
std::map<std::string, void*> get_ggml_graph_output_dst(std::shared_ptr<GgmlOvDecoder> ggml_decoder) {
std::map<std::string, void*> output_tensors;
auto output_names = ggml_decoder->get_model_output_names();
for (size_t inp = 0; inp < output_names.size(); ++inp) {
auto name = output_names[inp];
const auto* tensor = ggml_decoder->get_output_ggml_tensor(name);
auto* output_data = tensor->view_src ? tensor->view_src->data : tensor->data;
output_tensors[name] = output_data;
}
return output_tensors;
}
static ov::frontend::FrontEnd::Ptr get_ggml_frontend() {
auto fem = ov::frontend::FrontEndManager();
auto front_end = fem.load_by_framework("ggml");
return front_end;
}
enum ggml_status openvino_frontend_compute(ggml_backend_t backend, struct ggml_cgraph* cgraph) {
static ov::Core core;
static std::string device = getenv("GGML_OPENVINO_DEVICE") ? getenv("GGML_OPENVINO_DEVICE") : "";
if (device.empty()) {
const std::vector<std::string> preferred_device = { "GPU", "CPU", "NPU" };
const auto available_devices = core.get_available_devices();
for (const auto& dev : preferred_device) {
if (std::find(available_devices.begin(), available_devices.end(), dev) != available_devices.end()) {
device = dev;
break;
}
}
}
bool is_static = device == "NPU" ? true : false;
ov::AnyMap config;
if (device == "GPU") {
config = {
{"GPU_ENABLE_SDPA_OPTIMIZATION", "0"}
};
}
if (getenv("GGML_OPENVINO_DUMP_CGRAPH")) {
std::string filename = "cgraph.txt";
GgmlOvDecoder::dump_cgraph(cgraph, filename);
}
if (is_naive(cgraph)) {
return naive_compute(cgraph, core, device, config);
}
auto start_time = ggml_time_us();
auto* cache_dir = getenv("GGML_OPENVINO_CACHE_DIR");
if (cache_dir && !is_static) {
core.set_property(ov::cache_dir(cache_dir));
}
static std::mutex cache_mutex;
static std::unordered_map<struct ggml_cgraph*, std::shared_ptr<ov::InferRequest>> infer_request_cache;
static std::unordered_map<struct ggml_cgraph*, std::vector<std::string>> ov_input_names_cache;
static std::unordered_map<struct ggml_cgraph*, std::vector<std::string>> ov_output_names_cache;
// For NPU, store the kvcache model, since we cannot create two infer_request
static std::unordered_map<struct ggml_cgraph*, ov::CompiledModel> compiled_model_cache;
std::shared_ptr<GgmlOvDecoder> ggml_decoder;
ov::InferRequest infer_request;
int64_t decoder_end_time;
int64_t conversion_end_time;
int64_t compile_end_time;
{
std::lock_guard<std::mutex> lock(cache_mutex);
auto it = infer_request_cache.find(cgraph);
if (it != infer_request_cache.end()) {
std::map<std::string, std::shared_ptr<ov::Node>> model_weights;
ggml_decoder = std::make_shared<GgmlOvDecoder>(cgraph, model_weights, is_static, false);
decoder_end_time = ggml_time_us();
// For NPU for the first time we call kvcache modle, pop the compiled kvcache model from cache
if (is_static && compiled_model_cache.find(cgraph) != compiled_model_cache.end()) {
infer_request_cache[cgraph] =
std::make_shared<ov::InferRequest>(compiled_model_cache[cgraph].create_infer_request());
compiled_model_cache.erase(cgraph);
}
infer_request = *infer_request_cache[cgraph];
conversion_end_time = ggml_time_us();
compile_end_time = conversion_end_time;
} else {
std::shared_ptr<ov::Model> model;
auto model_weights = GgmlOvDecoder::create_weight_nodes(cgraph, get_types_to_requant(device));
if (is_static) {
ggml_decoder = std::make_shared<GgmlOvDecoder>(cgraph, model_weights, is_static, true);
auto ggml_decoder_kvcache = std::make_shared<GgmlOvDecoder>(cgraph, model_weights, is_static, false);
decoder_end_time = ggml_time_us();
auto input_model = std::make_shared<ov::frontend::ggml::InputModel>(ggml_decoder);
auto input_model_kvcache = std::make_shared<ov::frontend::ggml::InputModel>(ggml_decoder_kvcache);
model = ov::frontend::ggml::FrontEnd::convert(input_model);
ggml_decoder->clear_model_weights();
auto model_kvcache = ov::frontend::ggml::FrontEnd::convert(input_model_kvcache);
ggml_decoder_kvcache->clear_model_weights();
conversion_end_time = ggml_time_us();
if (getenv("GGML_OPENVINO_DUMP_IR")) {
char timestamped_filename[64];
auto timestamp = (long long) ggml_time_us();
snprintf(timestamped_filename, sizeof(timestamped_filename), "model_prefill_%lld.xml", timestamp);
ov::serialize(model, timestamped_filename);
snprintf(timestamped_filename, sizeof(timestamped_filename), "model_kvcache_%lld.xml", timestamp);
ov::serialize(model_kvcache, timestamped_filename);
}
auto compiled_model = core.compile_model(model, device, get_npu_prefill_config());
auto compiled_model_kvcache = core.compile_model(model_kvcache, device, get_npu_generate_config());
compiled_model_cache[cgraph] = compiled_model_kvcache;
compile_end_time = ggml_time_us();
infer_request_cache[cgraph] = std::make_shared<ov::InferRequest>(compiled_model.create_infer_request());
infer_request = *infer_request_cache[cgraph];
compiled_model_cache[cgraph] = compiled_model_kvcache;
} else {
ggml_decoder = std::make_shared<GgmlOvDecoder>(cgraph, model_weights, is_static, true);
decoder_end_time = ggml_time_us();
auto input_model = std::make_shared<ov::frontend::ggml::InputModel>(ggml_decoder);
model = ov::frontend::ggml::FrontEnd::convert(input_model);
ggml_decoder->clear_model_weights();
conversion_end_time = ggml_time_us();
if (getenv("GGML_OPENVINO_DUMP_IR")) {
char timestamped_filename[64];
auto timestamp = (long long) ggml_time_us();
snprintf(timestamped_filename, sizeof(timestamped_filename), "model_%lld.xml", timestamp);
ov::serialize(model, timestamped_filename);
}
auto compiled_model = core.compile_model(model, device, config);
compile_end_time = ggml_time_us();
infer_request_cache[cgraph] = std::make_shared<ov::InferRequest>(compiled_model.create_infer_request());
infer_request = *infer_request_cache[cgraph];
}
std::vector<std::string> ov_input_names;
std::vector<std::string> ov_output_names;
for (const auto& ov_param : model->get_parameters()) {
ov_input_names.push_back(ov_param->get_friendly_name());
}
for (const auto& ov_output : model->get_results()) {
ov_output_names.push_back(ov_output->get_friendly_name());
}
ov_input_names_cache[cgraph] = ov_input_names;
ov_output_names_cache[cgraph] = ov_output_names;
}
}
auto ov_input_names = ov_input_names_cache[cgraph];
auto ov_output_names = ov_output_names_cache[cgraph];
for (size_t i = 0; i < ov_input_names.size(); i++) {
auto param_name = ov_input_names[i];
auto input_tensor = get_ov_input_tensor(ggml_decoder, param_name);
infer_request.set_input_tensor(i, input_tensor);
if (getenv("GGML_OPENVINO_DEBUG_INPUT")) {
print_input_tensor_info(param_name, input_tensor);
}
}
auto input_end_time = ggml_time_us();
infer_request.infer();
auto infer_end_time = ggml_time_us();
auto gguf_tensor_addrs = get_ggml_graph_output_dst(ggml_decoder);
for (size_t i = 0; i < ov_output_names.size(); i++) {
auto& result_name = ov_output_names[i];
const auto output_tensor = infer_request.get_output_tensor(i);
std::memcpy(gguf_tensor_addrs[result_name], output_tensor.data(), output_tensor.get_byte_size());
if (getenv("GGML_OPENVINO_DEBUG_OUTPUT")) {
print_output_tensor_info(result_name, output_tensor, gguf_tensor_addrs);
}
}
auto end_time = ggml_time_us();
if (getenv("GGML_OPENVINO_PROFILING")) {
GGML_LOG_INFO("GGML OpenVINO Backend: \n");
GGML_LOG_INFO(" - Graph decoder Time: %ld ms \n", (decoder_end_time - start_time) / 1000);
GGML_LOG_INFO(" - Graph conversion Time: %ld ms \n", (conversion_end_time - decoder_end_time) / 1000);
GGML_LOG_INFO(" - Graph compile Time: %ld ms \n", (compile_end_time - conversion_end_time) / 1000);
GGML_LOG_INFO(" - Graph Input Time: %ld ms \n", (input_end_time - compile_end_time) / 1000);
GGML_LOG_INFO(" - Graph Inference Time: %ld ms \n", (infer_end_time - input_end_time) / 1000);
GGML_LOG_INFO(" - Graph Output Time: %ld ms \n", (end_time - infer_end_time) / 1000);
}
return GGML_STATUS_SUCCESS;
GGML_UNUSED(backend);
}
namespace {
ov::AnyMap get_npu_base_config() {
return {
{"NPU_COMPILATION_MODE_PARAMS", "compute-layers-with-higher-precision=Sqrt,Power,ReduceMean,Add_RMSNorm" },
{"NPU_COMPILER_DYNAMIC_QUANTIZATION", "YES" },
{"NPU_USE_NPUW", "YES" },
{"NPUW_DEVICES", "NPU" },
{"NPUW_FOLD", "YES" },
{"NPUW_WEIGHTS_BANK", "shared" },
{"NPUW_FUNCALL_FOR_ALL", "YES" },
{"NPUW_FUNCALL_ASYNC", "YES" },
{"NPUW_DQ", "YES" },
{"NPUW_DQ_FULL", "NO" },
{"NPUW_CACHE_DIR", getenv("GGML_OPENVINO_CACHE_DIR") ? getenv("GGML_OPENVINO_CACHE_DIR") : ""},
};
}
} // namespace
ov::AnyMap get_npu_prefill_config() {
auto config = get_npu_base_config();
return config;
}
ov::AnyMap get_npu_generate_config() {
auto config = get_npu_base_config();
return config;
}
std::map<ggml_type, ExtraQuantType> get_types_to_requant(const std::string& device) {
if (device == "NPU") {
return {
{GGML_TYPE_Q4_0, ExtraQuantType::Q4_0_128},
{GGML_TYPE_Q4_1, ExtraQuantType::Q4_0_128},
{GGML_TYPE_Q4_K, ExtraQuantType::Q4_0_128},
{GGML_TYPE_Q6_K, ExtraQuantType::F16 },
};
}
if (device == "GPU") {
return {
// CVS-166739
{GGML_TYPE_Q6_K, ExtraQuantType::Q8_1_C},
};
}
return {};
}
bool is_naive(struct ggml_cgraph* cgraph) {
constexpr int naive_graph_size_threshold = 20;
return cgraph->n_nodes < naive_graph_size_threshold;
}
enum ggml_status naive_compute(struct ggml_cgraph* cgraph,
ov::Core& core,
const std::string& device,
const ov::AnyMap& config) {
if (cgraph->n_nodes == 1 && (cgraph->nodes[0]->op == GGML_OP_NONE || cgraph->nodes[0]->op == GGML_OP_VIEW)) {
return GGML_STATUS_SUCCESS;
}
if (cgraph->nodes[0]->op == GGML_OP_FLASH_ATTN_EXT) {
return GGML_STATUS_FAILED;
}
auto model_weights = GgmlOvDecoder::create_weight_nodes(cgraph);
auto decoder = std::make_shared<GgmlOvDecoder>(cgraph, model_weights);
auto input_model = std::make_shared<ov::frontend::ggml::InputModel>(decoder);
auto naive = true;
auto model = ov::frontend::ggml::FrontEnd::convert(input_model, naive);
if (getenv("GGML_OPENVINO_DUMP_IR")) {
ov::serialize(model, "IR_naive.xml");
}
auto infer_request = core.compile_model(model, device, config).create_infer_request();
auto ov_params = model->get_parameters();
for (size_t i = 0; i < ov_params.size(); i++) {
auto param_name = ov_params[i]->get_friendly_name();
auto input_tensor = get_ov_input_tensor(decoder, param_name);
infer_request.set_input_tensor(i, input_tensor);
}
infer_request.infer();
auto gguf_tensor_addrs = get_ggml_graph_output_dst(decoder);
auto ov_results = model->get_results();
for (size_t i = 0; i < ov_results.size(); i++) {
auto result_name = ov_results[i]->get_friendly_name();
const auto output_tensor = infer_request.get_output_tensor(i);
std::memcpy(gguf_tensor_addrs[result_name], output_tensor.data(), output_tensor.get_byte_size());
}
return GGML_STATUS_SUCCESS;
}
ov::Tensor get_ov_input_tensor(std::shared_ptr<GgmlOvDecoder> ggml_decoder, const std::string& param_name) {
bool is_static = ggml_decoder->is_static();
bool is_first_token = ggml_decoder->is_first_token();
ov::Tensor input_tensor;
if (ggml_decoder->get_model_extra_inputs().find(param_name) != ggml_decoder->get_model_extra_inputs().end()) {
input_tensor = *ggml_decoder->get_model_extra_input_values().at(param_name);
} else if (!is_static) {
input_tensor = convert_ggml_input_to_ov(ggml_decoder, param_name);
} else {
if (param_name == "inp_tokens" || param_name == "inp_pos") {
if (is_first_token) {
size_t context_size = ggml_decoder->get_context_size();
const auto* input_tensor_ggml = ggml_decoder->get_input_ggml_tensor(param_name);
std::vector<int32_t> padded_data = pad_input<int32_t>(input_tensor_ggml, 1, context_size, 0);
input_tensor = ov::Tensor(ov::element::i32, ov::Shape{1, 1, context_size});
auto* data_ptr = input_tensor.data<int32_t>();
std::copy(padded_data.begin(), padded_data.end(), data_ptr);
} else {
input_tensor = convert_ggml_input_to_ov(ggml_decoder, param_name);
}
} else if (param_name.find("KQ_mask") == 0) {
size_t context_size = ggml_decoder->get_context_size();
const auto* input_tensor_ggml = ggml_decoder->get_input_ggml_tensor(param_name);
if (is_first_token) {
std::vector<float> padded_data =
pad_input<float>(input_tensor_ggml, context_size, context_size, -INFINITY);
set_zero_diagonal(padded_data, context_size);
input_tensor = ov::Tensor(ov::element::f32, ov::Shape{1, context_size, context_size});
auto* data_ptr = input_tensor.data<float>();
std::copy(padded_data.begin(), padded_data.end(), data_ptr);
} else {
std::vector<float> padded_data = pad_input<float>(input_tensor_ggml, 1, context_size, -INFINITY);
input_tensor = ov::Tensor(ov::element::f32, ov::Shape{1, 1, context_size});
auto* data_ptr = input_tensor.data<float>();
std::copy(padded_data.begin(), padded_data.end(), data_ptr);
}
} else if (const auto* op = ggml_decoder->get_tensor_used_op(ggml_decoder->get_tensor_from_name(param_name));
op && op->op == GGML_OP_SET_ROWS && is_static && is_first_token) {
input_tensor = ov::Tensor(ov::element::i64, ov::Shape{1, 1, 1});
} else {
input_tensor = convert_ggml_input_to_ov(ggml_decoder, param_name);
}
}
return input_tensor;
}
size_t checksum(const void* data, size_t size) {
const uint8_t* bytes = static_cast<const uint8_t*>(data);
size_t sum = 0;
for (size_t i = 0; i < size; ++i) {
sum += (uint8_t) i;
sum += bytes[i];
}
return sum;
}
// Suppress deprecation warning for ov::Tensor::data()
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
void print_input_tensor_info(const std::string& name, const ov::Tensor& tensor) {
std::cout << "Input name: " << name << ", Input shape: " << tensor.get_shape() << ", Address: " << tensor.data()
<< std::endl;
switch (tensor.get_element_type()) {
case ov::element::f32:
std::cout << *(tensor.data<float>()) << std::endl;
break;
case ov::element::f16:
std::cout << *(tensor.data<ov::float16>()) << std::endl;
break;
case ov::element::i32:
for (size_t i = 0; i < tensor.get_size(); ++i) {
std::cout << tensor.data<int32_t>()[i] << " ";
}
std::cout << std::endl;
break;
case ov::element::i64:
std::cout << *(tensor.data<int64_t>()) << std::endl;
break;
default:
break;
}
}
void print_output_tensor_info(const std::string& name, const ov::Tensor& tensor,
std::map<std::string, void*>& output_dst) {
std::cout << "Output name: " << name << ", Output shape: " << tensor.get_shape()
<< ", Address: " << output_dst[name] << std::endl;
auto print_float_stats = [](const std::string& type_name, size_t size, auto get_value) {
if (size == 0) {
return;
}
float first = get_value(0);
float min = first;
float max = first;
double sum = first;
for (size_t i = 1; i < size; ++i) {
float v = get_value(i);
if (v < min) {
min = v;
}
if (v > max) {
max = v;
}
sum += v;
}
double mean = sum / size;
std::cout << std::right << std::setw(6) << type_name << std::right << std::setw(12) << "First" << std::setw(12)
<< "Min" << std::setw(12) << "Max" << std::setw(12) << "Mean" << std::endl;
std::cout << std::right << std::setw(6) << "" << std::right << std::setw(12) << first << std::setw(12) << min
<< std::setw(12) << max << std::setw(12) << mean << std::endl;
};
switch (tensor.get_element_type()) {
case ov::element::f32: {
const float* data = tensor.data<float>();
size_t size = tensor.get_size();
print_float_stats("[f32]", size, [data](size_t i) { return data[i]; });
break;
}
case ov::element::f16: {
const ov::float16* data = tensor.data<ov::float16>();
size_t size = tensor.get_size();
print_float_stats("[f16]", size, [data](size_t i) { return static_cast<float>(data[i]); });
break;
}
default:
break;
}
}
#pragma GCC diagnostic pop
void set_zero_diagonal(std::vector<float>& matrix, size_t dim) {
for (size_t i = 0; i < dim; ++i) {
matrix[i * dim + i] = 0.0f;
}
}
bool is_prefill(struct ggml_cgraph* cgraph) {
for (int i = 0; i < cgraph->n_nodes; ++i) {
auto* op = cgraph->nodes[i];
for (int j = 0; j < GGML_MAX_SRC; ++j) {
auto* src = op->src[j];
if (src == nullptr) {
break;
}
if (std::string(src->name) == "inp_tokens") {
return src->ne[0] != 1;
}
}
}
GGML_LOG_ERROR("is_prefill: inp_tokens not found in cgraph");
throw std::runtime_error("is_prefill: inp_tokens not found in cgraph");
}
Reference in New Issue View Git Blame Copy Permalink