happyz
/
gemma.cpp
mirror of https://github.com/google/gemma.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'dev' into deinterleave-vecs

This commit is contained in:
Sam Kaufman 2024-04-30 15:54:19 -07:00
parent 2829ef17ad 12fb2f05cf
commit 564937ede6
12 changed files with 226 additions and 213 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
BUILD.bazel
View File

@ -46,8 +46,10 @@ cc_test(
deps = [ deps = [
":ops", ":ops",
"@googletest//:gtest_main", "@googletest//:gtest_main",
"//compression:compress",
"@hwy//:hwy", "@hwy//:hwy",
"@hwy//:hwy_test_util", "@hwy//:hwy_test_util",
"@hwy//:thread_pool",
], ],
) )

11
debug_prompt.cc
View File

@ -7,6 +7,7 @@
#include "nlohmann/json.hpp" #include "nlohmann/json.hpp"
#include "util/app.h" #include "util/app.h"
#include "util/args.h" #include "util/args.h"
#include "hwy/contrib/thread_pool/thread_pool.h"
using json = nlohmann::json; using json = nlohmann::json;
@ -27,8 +28,8 @@ class PromptArgs : public gcpp::ArgsBase<PromptArgs> {
std::pair<std::string, int> QueryModel( std::pair<std::string, int> QueryModel(
gcpp::Gemma& model, gcpp::InferenceArgs& args, gcpp::AppArgs& app, gcpp::Gemma& model, gcpp::InferenceArgs& args, gcpp::AppArgs& app,
gcpp::KVCache& kv_cache, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, gcpp::KVCache& kv_cache, hwy::ThreadPool& pool, const std::string& input,
const std::string& input, gcpp::LayersOutputT* layers_output) { gcpp::LayersOutputT* layers_output) {
std::vector<int> prompt; std::vector<int> prompt;
HWY_ASSERT(model.Tokenizer()->Encode(input, &prompt)); HWY_ASSERT(model.Tokenizer()->Encode(input, &prompt));
@ -55,8 +56,7 @@ std::pair<std::string, int> QueryModel(
} }
GenerateGemma(model, args.max_tokens, args.max_generated_tokens, GenerateGemma(model, args.max_tokens, args.max_generated_tokens,
args.temperature, prompt, /*abs_pos=*/0, kv_cache, pool, args.temperature, prompt, /*abs_pos=*/0, kv_cache, pool,
inner_pool, stream_token, accept_token, gen, app.verbosity, stream_token, accept_token, gen, app.verbosity, layers_output);
layers_output);
return {res, total_tokens}; return {res, total_tokens};
} }
@ -92,7 +92,6 @@ int main(int argc, char** argv) {
gcpp::LayersOutputT* layers_output = gcpp::LayersOutputT* layers_output =
log_layers_output ? &json_logger.layers_output_log_f : nullptr; log_layers_output ? &json_logger.layers_output_log_f : nullptr;
hwy::ThreadPool inner_pool(0);
hwy::ThreadPool pool(app.num_threads); hwy::ThreadPool pool(app.num_threads);
// For many-core, pinning threads to cores helps. // For many-core, pinning threads to cores helps.
if (app.num_threads > 10) { if (app.num_threads > 10) {
@ -112,7 +111,7 @@ int main(int argc, char** argv) {
return EXIT_FAILURE; return EXIT_FAILURE;
} }
const auto [answer, token_count] = QueryModel( const auto [answer, token_count] = QueryModel(
model, args, app, kv_cache, inner_pool, pool, prompt, layers_output); model, args, app, kv_cache, pool, prompt, layers_output);
std::cout << answer.substr(prompt.size()) << "\n" << std::flush; std::cout << answer.substr(prompt.size()) << "\n" << std::flush;
if (log_layers_output) { if (log_layers_output) {

37
gemma/benchmark.cc
View File

@ -58,8 +58,7 @@ void LogSpeedStats(const double time_start, size_t total_tokens) {
std::pair<std::string, int> QueryModel( std::pair<std::string, int> QueryModel(
gcpp::Gemma& model, gcpp::InferenceArgs& args, gcpp::AppArgs& app, gcpp::Gemma& model, gcpp::InferenceArgs& args, gcpp::AppArgs& app,
gcpp::KVCache& kv_cache, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, gcpp::KVCache& kv_cache, hwy::ThreadPool& pool, const std::string& input) {
const std::string& input) {
std::vector<int> prompt; std::vector<int> prompt;
HWY_ASSERT(model.Tokenizer()->Encode(input, &prompt)); HWY_ASSERT(model.Tokenizer()->Encode(input, &prompt));
@ -90,7 +89,7 @@ std::pair<std::string, int> QueryModel(
} }
GenerateGemma(model, args.max_tokens, args.max_generated_tokens, GenerateGemma(model, args.max_tokens, args.max_generated_tokens,
args.temperature, prompt, /*abs_pos=*/0, kv_cache, pool, args.temperature, prompt, /*abs_pos=*/0, kv_cache, pool,
inner_pool, stream_token, accept_token, gen, app.verbosity); stream_token, accept_token, gen, app.verbosity);
if (app.verbosity >= 1) { if (app.verbosity >= 1) {
LogSpeedStats(time_start, total_tokens); LogSpeedStats(time_start, total_tokens);
} }
@ -131,8 +130,7 @@ std::string ReadFile(const gcpp::Path& path) {
int BenchmarkGoldens(gcpp::Gemma& model, gcpp::InferenceArgs& args, int BenchmarkGoldens(gcpp::Gemma& model, gcpp::InferenceArgs& args,
gcpp::AppArgs& app, gcpp::KVCache& kv_cache, gcpp::AppArgs& app, gcpp::KVCache& kv_cache,
hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, hwy::ThreadPool& pool, const std::string& golden_path) {
const std::string& golden_path) {
const std::vector<std::pair<std::string, std::string>> queries_answers = const std::vector<std::pair<std::string, std::string>> queries_answers =
load_goldens(golden_path); load_goldens(golden_path);
int correct_answers = 0; int correct_answers = 0;
@ -140,7 +138,7 @@ int BenchmarkGoldens(gcpp::Gemma& model, gcpp::InferenceArgs& args,
const double time_start = hwy::platform::Now(); const double time_start = hwy::platform::Now();
for (const auto& [question, expected_answer] : queries_answers) { for (const auto& [question, expected_answer] : queries_answers) {
const auto [answer, token_count] = const auto [answer, token_count] =
QueryModel(model, args, app, kv_cache, inner_pool, pool, question); QueryModel(model, args, app, kv_cache, pool, question);
total_tokens += token_count; total_tokens += token_count;
if (answer.find(expected_answer) != std::string::npos) { if (answer.find(expected_answer) != std::string::npos) {
correct_answers++; correct_answers++;
@ -164,14 +162,13 @@ int BenchmarkGoldens(gcpp::Gemma& model, gcpp::InferenceArgs& args,
int BenchmarkSummary(gcpp::Gemma& model, gcpp::InferenceArgs& args, int BenchmarkSummary(gcpp::Gemma& model, gcpp::InferenceArgs& args,
gcpp::AppArgs& app, gcpp::KVCache& kv_cache, gcpp::AppArgs& app, gcpp::KVCache& kv_cache,
hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, hwy::ThreadPool& pool, const gcpp::Path& text) {
const gcpp::Path& text) {
std::string prompt("Here is some text to summarize:\n"); std::string prompt("Here is some text to summarize:\n");
prompt.append(ReadFile(text)); prompt.append(ReadFile(text));
prompt.append("\nSummarize this text.\n"); prompt.append("\nSummarize this text.\n");
const double time_start = hwy::platform::Now(); const double time_start = hwy::platform::Now();
const auto [answer, token_count] = const auto [answer, token_count] =
QueryModel(model, args, app, kv_cache, inner_pool, pool, prompt); QueryModel(model, args, app, kv_cache, pool, prompt);
std::cout << answer.substr(prompt.size()) << "\n" << std::flush; std::cout << answer.substr(prompt.size()) << "\n" << std::flush;
LogSpeedStats(time_start, token_count); LogSpeedStats(time_start, token_count);
return EXIT_SUCCESS; return EXIT_SUCCESS;
@ -179,8 +176,8 @@ int BenchmarkSummary(gcpp::Gemma& model, gcpp::InferenceArgs& args,
int BenchmarkCrossEntropy(gcpp::Gemma& model, gcpp::Model model_type, int BenchmarkCrossEntropy(gcpp::Gemma& model, gcpp::Model model_type,
gcpp::InferenceArgs& args, gcpp::AppArgs& app, gcpp::InferenceArgs& args, gcpp::AppArgs& app,
hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, hwy::ThreadPool& pool, const gcpp::Path& text,
const gcpp::Path& text, size_t batch_tokens) { size_t batch_tokens) {
std::string input = ReadFile(text); std::string input = ReadFile(text);
std::vector<int> prompt; std::vector<int> prompt;
HWY_ASSERT(model.Tokenizer()->Encode(input, &prompt)); HWY_ASSERT(model.Tokenizer()->Encode(input, &prompt));
@ -197,7 +194,7 @@ int BenchmarkCrossEntropy(gcpp::Gemma& model, gcpp::Model model_type,
auto kv_cache = CreateKVCache(model_type); auto kv_cache = CreateKVCache(model_type);
float entropy = float entropy =
ComputeCrossEntropy(model, num_tokens, prompt_slice, kv_cache, pool, ComputeCrossEntropy(model, num_tokens, prompt_slice, kv_cache, pool,
inner_pool, app.verbosity); app.verbosity);
total_entropy += entropy; total_entropy += entropy;
LogSpeedStats(time_start, pos + num_tokens); LogSpeedStats(time_start, pos + num_tokens);
std::string text_slice; std::string text_slice;
@ -211,8 +208,8 @@ int BenchmarkCrossEntropy(gcpp::Gemma& model, gcpp::Model model_type,
int BenchmarkTriviaQA(gcpp::Gemma& model, gcpp::InferenceArgs& args, int BenchmarkTriviaQA(gcpp::Gemma& model, gcpp::InferenceArgs& args,
gcpp::AppArgs& app, gcpp::KVCache& kv_cache, gcpp::AppArgs& app, gcpp::KVCache& kv_cache,
hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, hwy::ThreadPool& pool, const gcpp::Path& json_file,
const gcpp::Path& json_file, size_t max_questions) { size_t max_questions) {
std::ifstream trivia_file(json_file.path); std::ifstream trivia_file(json_file.path);
if (!trivia_file) { if (!trivia_file) {
std::cout << "Could not load file: " << json_file.path << "\n" std::cout << "Could not load file: " << json_file.path << "\n"
@ -225,7 +222,7 @@ int BenchmarkTriviaQA(gcpp::Gemma& model, gcpp::InferenceArgs& args,
while (std::getline(trivia_file, line)) { while (std::getline(trivia_file, line)) {
json data = json::parse(line); json data = json::parse(line);
const auto [answer, token_count] = QueryModel( const auto [answer, token_count] = QueryModel(
model, args, app, kv_cache, inner_pool, pool, data["question"]); model, args, app, kv_cache, pool, data["question"]);
std::cout << answer << "\n"; std::cout << answer << "\n";
bool correct = false; bool correct = false;
for (const std::string expected : data["answer"]["aliases"]) { for (const std::string expected : data["answer"]["aliases"]) {
@ -263,7 +260,6 @@ int main(int argc, char** argv) {
HWY_ABORT("\nInvalid inference args: %s", error); HWY_ABORT("\nInvalid inference args: %s", error);
} }
hwy::ThreadPool inner_pool(0);
hwy::ThreadPool pool(app.num_threads); hwy::ThreadPool pool(app.num_threads);
// For many-core, pinning threads to cores helps. // For many-core, pinning threads to cores helps.
if (app.num_threads > 10) { if (app.num_threads > 10) {
@ -280,17 +276,16 @@ int main(int argc, char** argv) {
if (!benchmark_args.goldens.path.empty()) { if (!benchmark_args.goldens.path.empty()) {
const std::string golden_path = const std::string golden_path =
benchmark_args.goldens.path + "/" + loader.model_type_str + ".txt"; benchmark_args.goldens.path + "/" + loader.model_type_str + ".txt";
return BenchmarkGoldens(model, args, app, kv_cache, inner_pool, pool, return BenchmarkGoldens(model, args, app, kv_cache, pool, golden_path);
golden_path);
} else if (!benchmark_args.summarize_text.path.empty()) { } else if (!benchmark_args.summarize_text.path.empty()) {
return BenchmarkSummary(model, args, app, kv_cache, inner_pool, pool, return BenchmarkSummary(model, args, app, kv_cache, pool,
benchmark_args.summarize_text); benchmark_args.summarize_text);
} else if (!benchmark_args.cross_entropy.path.empty()) { } else if (!benchmark_args.cross_entropy.path.empty()) {
return BenchmarkCrossEntropy(model, loader.ModelType(), args, app, return BenchmarkCrossEntropy(model, loader.ModelType(), args, app,
inner_pool, pool, benchmark_args.cross_entropy, pool, benchmark_args.cross_entropy,
benchmark_args.batch_tokens); benchmark_args.batch_tokens);
} else if (!benchmark_args.trivia_qa.path.empty()) { } else if (!benchmark_args.trivia_qa.path.empty()) {
return BenchmarkTriviaQA(model, args, app, kv_cache, inner_pool, pool, return BenchmarkTriviaQA(model, args, app, kv_cache, pool,
benchmark_args.trivia_qa, benchmark_args.trivia_qa,
benchmark_args.max_questions); benchmark_args.max_questions);
} }

4
gemma/compress_weights.cc
View File

@ -70,7 +70,7 @@ struct Args : public ArgsBase<Args> {
template <class Visitor> template <class Visitor>
void ForEach(const Visitor& visitor) { void ForEach(const Visitor& visitor) {
visitor(weights, "weights", Path(), visitor(weights, "weights", Path(),
"Path name of model weights (.sbs) file.\n" "Path to model weights (.bin) file.\n"
" Required argument."); " Required argument.");
visitor(model_type_str, "model", std::string(), visitor(model_type_str, "model", std::string(),
"Model type\n 2b-it = 2B parameters, instruction-tuned\n " "Model type\n 2b-it = 2B parameters, instruction-tuned\n "
@ -80,7 +80,7 @@ struct Args : public ArgsBase<Args> {
"gr2b-pt = griffin 2B parameters, pretrained\n " "gr2b-pt = griffin 2B parameters, pretrained\n "
" Required argument."); " Required argument.");
visitor(compressed_weights, "compressed_weights", Path(), visitor(compressed_weights, "compressed_weights", Path(),
"Path name where compressed weights file will be written.\n" "Path name where compressed weights (.sbs) file will be written.\n"
" Required argument."); " Required argument.");
visitor(num_threads, "num_threads", visitor(num_threads, "num_threads",
kDefaultNumThreads, // see ChooseNumThreads kDefaultNumThreads, // see ChooseNumThreads

6
gemma/configs.h
View File

@ -28,6 +28,11 @@
#define GEMMA_TOPK 1 #define GEMMA_TOPK 1
#endif // !GEMMA_TOPK #endif // !GEMMA_TOPK
// Allow changing upper bound on threads as a compiler flag
#ifndef GEMMA_MAX_THREADS
#define GEMMA_MAX_THREADS 128
#endif // !GEMMA_MAX_THREADS
#include <stddef.h> #include <stddef.h>
#include <array> #include <array>
@ -45,6 +50,7 @@ namespace gcpp {
static constexpr size_t kSeqLen = GEMMA_MAX_SEQLEN; static constexpr size_t kSeqLen = GEMMA_MAX_SEQLEN;
static constexpr size_t kTopK = GEMMA_TOPK; static constexpr size_t kTopK = GEMMA_TOPK;
static constexpr size_t kMaxThreads = GEMMA_MAX_THREADS;
enum class LayerAttentionType { enum class LayerAttentionType {
kGemma, kGemma,

262
gemma/gemma.cc
View File

@ -399,9 +399,9 @@ struct Activations {
static constexpr size_t kQKVDim = TConfig::kQKVDim; static constexpr size_t kQKVDim = TConfig::kQKVDim;
static constexpr size_t kHeads = TConfig::kHeads; static constexpr size_t kHeads = TConfig::kHeads;
static constexpr size_t kKVHeads = TConfig::kKVHeads; static constexpr size_t kKVHeads = TConfig::kKVHeads;
static constexpr size_t kCacheLayerSize = kKVHeads * kQKVDim * 2;
static constexpr size_t kCachePosSize = static constexpr size_t kCachePosSize =
TConfig::kGemmaLayers * kKVHeads * kQKVDim; TConfig::kGemmaLayers * kCacheLayerSize;
static constexpr size_t kCacheLayerSize = kKVHeads * kQKVDim;
std::array<float, kBatchSize * kModelDim> x; // input std::array<float, kBatchSize * kModelDim> x; // input
std::array<float, kBatchSize * kModelDim> pre_att_rms_out; std::array<float, kBatchSize * kModelDim> pre_att_rms_out;
@ -421,6 +421,10 @@ struct Activations {
std::array<float, kBatchSize * kModelDim> ffw_out; std::array<float, kBatchSize * kModelDim> ffw_out;
std::array<float, kBatchSize * TConfig::kVocabSize> logits; std::array<float, kBatchSize * TConfig::kVocabSize> logits;
// For bf16/f32 vectors * bf16 matrix: faster to unpack once beforehand, into
// per-thread storage.
std::array<float, kModelDim * kMaxThreads> even_odd;
// Griffin layer internal activations // Griffin layer internal activations
static constexpr size_t kGriffinDim = static constexpr size_t kGriffinDim =
TConfig::kGriffinLayers > 0 ? kModelDim : 0; TConfig::kGriffinLayers > 0 ? kModelDim : 0;
@ -440,15 +444,13 @@ struct GemmaInterface {
virtual void Generate(size_t max_tokens, size_t max_generated_tokens, virtual void Generate(size_t max_tokens, size_t max_generated_tokens,
float temperature, const std::vector<int>& prompt, float temperature, const std::vector<int>& prompt,
size_t start_pos, KVCache& kv_cache, size_t start_pos, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, const StreamFunc& stream_token,
const StreamFunc& stream_token,
const AcceptFunc& accept_token, std::mt19937& gen, const AcceptFunc& accept_token, std::mt19937& gen,
int verbosity, LayersOutputT* layers_output) = 0; int verbosity, LayersOutputT* layers_output) = 0;
virtual float ComputeCrossEntropy(size_t max_tokens, virtual float ComputeCrossEntropy(size_t max_tokens,
const std::vector<int>& prompt, const std::vector<int>& prompt,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool,
int verbosity) = 0; int verbosity) = 0;
}; };
@ -535,13 +537,12 @@ struct GemmaImpl : public GemmaInterface {
void Generate(size_t max_tokens, size_t max_generated_tokens, void Generate(size_t max_tokens, size_t max_generated_tokens,
float temperature, const std::vector<int>& prompt, float temperature, const std::vector<int>& prompt,
size_t start_pos, KVCache& kv_cache, hwy::ThreadPool& pool, size_t start_pos, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool, const StreamFunc& stream_token, const StreamFunc& stream_token, const AcceptFunc& accept_token,
const AcceptFunc& accept_token, std::mt19937&, int verbosity, std::mt19937&, int verbosity,
LayersOutputT* layers_output) override; LayersOutputT* layers_output) override;
float ComputeCrossEntropy(size_t max_tokens, const std::vector<int>& prompt, float ComputeCrossEntropy(size_t max_tokens, const std::vector<int>& prompt,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool,
int verbosity) override; int verbosity) override;
GemmaTokenizerImpl tokenizer; GemmaTokenizerImpl tokenizer;
@ -578,13 +579,14 @@ HWY_NOINLINE void GriffinRecurrent(
gcpp::Activations<TConfig, kBatchSize>::kModelDim; gcpp::Activations<TConfig, kBatchSize>::kModelDim;
static constexpr size_t kConv1dWidth = TConfig::kConv1dWidth; static constexpr size_t kConv1dWidth = TConfig::kConv1dWidth;
static constexpr size_t kHeads = TConfig::kHeads; static constexpr size_t kHeads = TConfig::kHeads;
static constexpr bool kAdd = true;
const size_t batch_offset = batch_idx * kModelDim; const size_t batch_offset = batch_idx * kModelDim;
const size_t pos = batch_start + batch_idx; const size_t pos = batch_start + batch_idx;
// X / Y linear layers. // X / Y linear layers.
float* HWY_RESTRICT y = activations.griffin_y.data() + batch_offset; float* HWY_RESTRICT y = activations.griffin_y.data() + batch_offset;
float* HWY_RESTRICT x = activations.griffin_x.data() + batch_offset; float* HWY_RESTRICT x = activations.griffin_x.data() + batch_offset;
TwoMatVecAdd<true, kModelDim, kModelDim>( TwoMatVecAdd<kAdd, kModelDim, kModelDim>(
layer_weights->griffin.linear_x_w, layer_weights->griffin.linear_y_w, 0, layer_weights->griffin.linear_x_w, layer_weights->griffin.linear_y_w, 0,
activations.pre_att_rms_out.data() + batch_offset, activations.pre_att_rms_out.data() + batch_offset,
/*add0=*/layer_weights->griffin.linear_x_biases.data(), /*add0=*/layer_weights->griffin.linear_x_biases.data(),
@ -634,7 +636,7 @@ HWY_NOINLINE void GriffinRecurrent(
constexpr size_t kHeadDim = kModelDim / kHeads; constexpr size_t kHeadDim = kModelDim / kHeads;
constexpr size_t kMatrixSize = kHeadDim * kHeadDim; constexpr size_t kMatrixSize = kHeadDim * kHeadDim;
size_t head_offset = head * kHeadDim; size_t head_offset = head * kHeadDim;
TwoOfsMatVecAddLoop<true, kHeadDim, kHeadDim>( TwoOfsMatVecAddLoop<kAdd, kHeadDim, kHeadDim>(
layer_weights->griffin.gate_w, kMatrixSize * head, layer_weights->griffin.gate_w, kMatrixSize * head,
kMatrixSize * (kHeads + head), x + head_offset, kMatrixSize * (kHeads + head), x + head_offset,
/*add0=*/layer_weights->griffin.gate_biases.data() + head_offset, /*add0=*/layer_weights->griffin.gate_biases.data() + head_offset,
@ -673,9 +675,10 @@ HWY_NOINLINE void GriffinRecurrent(
// Final linear layer. // Final linear layer.
float* out_ptr = activations.att_post2.data() + batch_idx * kModelDim; float* out_ptr = activations.att_post2.data() + batch_idx * kModelDim;
MatVecAdd<true, kModelDim, kModelDim>( MatVecAdd<kAdd, kModelDim, kModelDim>(
layer_weights->griffin.linear_out_w, 0, x, layer_weights->griffin.linear_out_w, 0, x,
layer_weights->griffin.linear_out_biases.data(), out_ptr, pool); layer_weights->griffin.linear_out_biases.data(),
activations.even_odd.data(), out_ptr, pool);
} }
template <size_t kBatchSize, typename LayerT, class TConfig> template <size_t kBatchSize, typename LayerT, class TConfig>
@ -707,26 +710,7 @@ HWY_NOINLINE void Attention(size_t batch_start, size_t batch_idx, size_t layer,
float* x = activations.pre_att_rms_out.data() + batch_idx * kModelDim; float* x = activations.pre_att_rms_out.data() + batch_idx * kModelDim;
auto ProjQ = [&](uint64_t head, size_t head_offset) HWY_ATTR { auto Attn = [&](uint64_t head, size_t head_offset, size_t thread) HWY_ATTR {
float* HWY_RESTRICT q =
activations.q.data() + head * kQKVDim + batch_idx * kHeads * kQKVDim;
MatVecLoop<kQKVDim, kModelDim>(layer_weights->qkv_einsum_w,
head_offset + 0 * kQKVDim * kModelDim, x, q);
};
auto ProjKV = [&](size_t k_offset, size_t v_offset,
size_t kv_offset) HWY_ATTR {
float* HWY_RESTRICT k = kv_cache.key_cache.get() + kv_offset;
float* HWY_RESTRICT v = kv_cache.value_cache.get() + kv_offset;
TwoOfsMatVecLoop<kQKVDim, kModelDim>(layer_weights->qkv_einsum_w, k_offset,
v_offset, x, k, v);
Rope(k, TConfig::kUseHalfRope ? kQKVDim / 2 : kQKVDim, pos);
};
auto Attn = [&](uint64_t head, size_t head_offset) HWY_ATTR {
// Calculate scores // Calculate scores
float* HWY_RESTRICT q = float* HWY_RESTRICT q =
activations.q.data() + head * kQKVDim + batch_idx * kHeads * kQKVDim; activations.q.data() + head * kQKVDim + batch_idx * kHeads * kQKVDim;
@ -741,7 +725,7 @@ HWY_NOINLINE void Attention(size_t batch_start, size_t batch_idx, size_t layer,
for (size_t pos2 = 0; pos2 < cache_num; ++pos2) { for (size_t pos2 = 0; pos2 < cache_num; ++pos2) {
const size_t cache_offset = const size_t cache_offset =
pos2 * kCachePosSize + layer * kCacheLayerSize + head_offset; pos2 * kCachePosSize + layer * kCacheLayerSize + head_offset;
const float* HWY_RESTRICT k2 = kv_cache.key_cache.get() + cache_offset; const float* HWY_RESTRICT k2 = kv_cache.kv_cache.get() + cache_offset;
const float score = Dot(q, k2, kQKVDim); const float score = Dot(q, k2, kQKVDim);
head_att[pos2] = score; head_att[pos2] = score;
} }
@ -754,7 +738,7 @@ HWY_NOINLINE void Attention(size_t batch_start, size_t batch_idx, size_t layer,
for (size_t pos2 = 0; pos2 < cache_num; ++pos2) { for (size_t pos2 = 0; pos2 < cache_num; ++pos2) {
const size_t cache_offset = const size_t cache_offset =
pos2 * kCachePosSize + layer * kCacheLayerSize + head_offset; pos2 * kCachePosSize + layer * kCacheLayerSize + head_offset;
float* HWY_RESTRICT v2 = kv_cache.value_cache.get() + cache_offset; float* HWY_RESTRICT v2 = kv_cache.kv_cache.get() + cache_offset + kQKVDim;
MulByConstAndAdd(head_att[pos2], v2, att_out, kQKVDim); MulByConstAndAdd(head_att[pos2], v2, att_out, kQKVDim);
} }
// linear projection from kQKVDim back to kModelDim, sum projections // linear projection from kQKVDim back to kModelDim, sum projections
@ -763,20 +747,21 @@ HWY_NOINLINE void Attention(size_t batch_start, size_t batch_idx, size_t layer,
head == 0 head == 0
? activations.att_post2.data() + batch_idx * kModelDim ? activations.att_post2.data() + batch_idx * kModelDim
: activations.att_post1.data() + head * kBatchSize * kModelDim; : activations.att_post1.data() + head * kBatchSize * kModelDim;
float* even_odd = activations.even_odd.data() + thread * kQKVDim;
if (head == 0) { if (head == 0) {
MatVecAddLoop<TConfig::kSoftmaxAttnOutputBiases, kModelDim, kQKVDim>( MatVecAddLoop<TConfig::kSoftmaxAttnOutputBiases, kModelDim, kQKVDim>(
layer_weights->attn_vec_einsum_w, head * kModelDim * kQKVDim, att_out, layer_weights->attn_vec_einsum_w, head * kModelDim * kQKVDim, att_out,
layer_weights->attention_output_biases.data(), head_out); layer_weights->attention_output_biases.data(), even_odd, head_out);
} else { } else {
MatVecLoop<kModelDim, kQKVDim>(layer_weights->attn_vec_einsum_w, MatVecLoop<kModelDim, kQKVDim>(layer_weights->attn_vec_einsum_w,
head * kModelDim * kQKVDim, att_out, head * kModelDim * kQKVDim, att_out,
head_out); even_odd, head_out);
} }
}; };
if constexpr (kHeads == kKVHeads) { if constexpr (kHeads == kKVHeads) {
// Multi-Head Attention // Multi-Head Attention
pool.Run(0, kHeads, [&](const uint64_t head, size_t /*thread*/) HWY_ATTR { pool.Run(0, kHeads, [&](const uint64_t head, size_t thread) HWY_ATTR {
// linear projections to QKV // linear projections to QKV
const size_t head_offset = TConfig::kInterleaveQKV const size_t head_offset = TConfig::kInterleaveQKV
? 3 * kQKVDim * kModelDim ? 3 * kQKVDim * kModelDim
@ -787,28 +772,41 @@ HWY_NOINLINE void Attention(size_t batch_start, size_t batch_idx, size_t layer,
const size_t k_offset = head * head_offset + 1 * mat_offset; const size_t k_offset = head * head_offset + 1 * mat_offset;
const size_t v_offset = head * head_offset + 2 * mat_offset; const size_t v_offset = head * head_offset + 2 * mat_offset;
ProjQ(head, q_offset); // ProjQ
float* HWY_RESTRICT q =
activations.q.data() + head * kQKVDim + batch_idx * kHeads * kQKVDim;
MatVecLoop<kQKVDim, kModelDim>(
layer_weights->qkv_einsum_w, q_offset + 0 * kQKVDim * kModelDim, x,
activations.even_odd.data() + thread * kModelDim, q);
const size_t kv_offset = // ProjKV
cache_pos * kCachePosSize + layer * kCacheLayerSize + head * kQKVDim; const size_t kv_offset = cache_pos * kCachePosSize +
layer * kCacheLayerSize + head * kQKVDim * 2;
float* HWY_RESTRICT k = kv_cache.kv_cache.get() + kv_offset;
float* HWY_RESTRICT v = k + kQKVDim;
TwoOfsMatVecLoop<kQKVDim, kModelDim>(layer_weights->qkv_einsum_w,
k_offset, v_offset, x, k, v);
Rope(k, TConfig::kUseHalfRope ? kQKVDim / 2 : kQKVDim, pos);
ProjKV(k_offset, v_offset, kv_offset); Attn(head, head * kQKVDim * 2, thread);
Attn(head, head * kQKVDim);
}); });
} else { } else {
// Multi-Query Attention // Multi-Query Attention
constexpr const size_t q_offset = kHeads * kQKVDim * kModelDim; float* HWY_RESTRICT q = activations.q.data() + batch_idx * kHeads * kQKVDim;
constexpr const size_t k_offset = q_offset + 0 * kQKVDim * kModelDim; MatVec<kHeads * kQKVDim, kModelDim>(layer_weights->qkv_einsum_w, 0, x,
constexpr const size_t v_offset = q_offset + 1 * kQKVDim * kModelDim; activations.even_odd.data(), q, pool);
const size_t kv_offset =
cache_pos * kCachePosSize + layer * kCacheLayerSize;
ProjKV(k_offset, v_offset, kv_offset); float* HWY_RESTRICT kv = kv_cache.kv_cache.get() +
cache_pos * kCachePosSize +
layer * kCacheLayerSize;
MatVec<kQKVDim * 2, kModelDim>(layer_weights->qkv_einsum_w,
kHeads * kQKVDim * kModelDim, x,
activations.even_odd.data(), kv, pool);
pool.Run(0, kHeads, [&](const uint64_t head, size_t /*thread*/) HWY_ATTR { Rope(kv, TConfig::kUseHalfRope ? kQKVDim / 2 : kQKVDim, pos);
ProjQ(head, head * kQKVDim * kModelDim);
Attn(head, 0); pool.Run(0, kHeads, [&](const uint64_t head, size_t thread) HWY_ATTR {
Attn(head, 0, thread);
}); });
} }
@ -828,6 +826,7 @@ HWY_NOINLINE void FFW(Activations<TConfig, kBatchSize>& activations,
static constexpr size_t kModelDim = TConfig::kModelDim; static constexpr size_t kModelDim = TConfig::kModelDim;
static constexpr size_t kFFHiddenDim = TConfig::kFFHiddenDim; static constexpr size_t kFFHiddenDim = TConfig::kFFHiddenDim;
const size_t hidden_offset = batch_idx * kFFHiddenDim * 2; const size_t hidden_offset = batch_idx * kFFHiddenDim * 2;
float* HWY_RESTRICT even_odd = activations.even_odd.data();
{ {
PROFILER_ZONE("Gen.FFW.GatedGELU"); PROFILER_ZONE("Gen.FFW.GatedGELU");
@ -836,15 +835,15 @@ HWY_NOINLINE void FFW(Activations<TConfig, kBatchSize>& activations,
float* HWY_RESTRICT out = activations.ffw_hidden.data() + hidden_offset; float* HWY_RESTRICT out = activations.ffw_hidden.data() + hidden_offset;
float* HWY_RESTRICT out_mul = out + kFFHiddenDim; float* HWY_RESTRICT out_mul = out + kFFHiddenDim;
// Same matrix, first and second half of rows. Could fuse into one MatVec, // Same matrix, first and second half of rows. Could fuse into one MatVec.
// but separating them could help on NUMA e.g. multiple sockets.
MatVecAdd<TConfig::kFFBiases, kFFHiddenDim, kModelDim>( MatVecAdd<TConfig::kFFBiases, kFFHiddenDim, kModelDim>(
layer_weights->gating_einsum_w, kFFHiddenDim * kModelDim, vec, layer_weights->gating_einsum_w, kFFHiddenDim * kModelDim, vec,
layer_weights->ffw_gating_biases.data() + kFFHiddenDim, out_mul, pool); layer_weights->ffw_gating_biases.data() + kFFHiddenDim, even_odd,
out_mul, pool);
// Gate, will go through the nonlinearity. // Gate, will go through the nonlinearity.
MatVecAdd<TConfig::kFFBiases, kFFHiddenDim, kModelDim>( MatVecAdd<TConfig::kFFBiases, kFFHiddenDim, kModelDim>(
layer_weights->gating_einsum_w, 0, vec, layer_weights->gating_einsum_w, 0, vec,
layer_weights->ffw_gating_biases.data(), out, pool); layer_weights->ffw_gating_biases.data(), even_odd, out, pool);
namespace hn = hwy::HWY_NAMESPACE; namespace hn = hwy::HWY_NAMESPACE;
using DF = hn::ScalableTag<float>; using DF = hn::ScalableTag<float>;
@ -857,7 +856,7 @@ HWY_NOINLINE void FFW(Activations<TConfig, kBatchSize>& activations,
PROFILER_ZONE("Gen.FFW\\GatedGELU"); PROFILER_ZONE("Gen.FFW\\GatedGELU");
MatVecAdd<TConfig::kFFBiases, kModelDim, kFFHiddenDim>( MatVecAdd<TConfig::kFFBiases, kModelDim, kFFHiddenDim>(
layer_weights->linear_w, 0, activations.ffw_hidden.data() + hidden_offset, layer_weights->linear_w, 0, activations.ffw_hidden.data() + hidden_offset,
layer_weights->ffw_output_biases.data(), layer_weights->ffw_output_biases.data(), even_odd,
activations.ffw_out.data() + batch_idx * kModelDim, pool); activations.ffw_out.data() + batch_idx * kModelDim, pool);
} }
@ -880,8 +879,7 @@ template <size_t kBatchSize, typename WeightArrayT, typename TConfig>
HWY_NOINLINE void Prefill(const int* tokens, size_t num_tokens, size_t pos, HWY_NOINLINE void Prefill(const int* tokens, size_t num_tokens, size_t pos,
const WeightArrayT& weights, const WeightArrayT& weights,
Activations<TConfig, kBatchSize>& activations, Activations<TConfig, kBatchSize>& activations,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool) {
hwy::ThreadPool& inner_pool) {
PROFILER_ZONE("Gen.Prefill\\Att\\FFW"); PROFILER_ZONE("Gen.Prefill\\Att\\FFW");
static constexpr size_t kModelDim = TConfig::kModelDim; static constexpr size_t kModelDim = TConfig::kModelDim;
GEMMA_CONSTEXPR_EMBSCALING const float kEmbScaling = GEMMA_CONSTEXPR_EMBSCALING const float kEmbScaling =
@ -924,19 +922,17 @@ HWY_NOINLINE void Prefill(const int* tokens, size_t num_tokens, size_t pos,
} }
// TODO: sink the loop into these functions, i.e. make them MatMul. // TODO: sink the loop into these functions, i.e. make them MatMul.
pool.Run( for (size_t token_idx = 0; token_idx < num_tokens; ++token_idx) {
0, num_tokens, AddFrom(activations.att_post2.data() + token_idx * kModelDim,
[&](const uint64_t token_idx, size_t thread_id) HWY_ATTR { activations.x.data() + token_idx * kModelDim, kModelDim);
AddFrom(activations.att_post2.data() + token_idx * kModelDim, RMSNorm(activations.x.data() + token_idx * kModelDim,
activations.x.data() + token_idx * kModelDim, kModelDim); layer_weights->pre_ffw_norm_scale.data(),
RMSNorm(activations.x.data() + token_idx * kModelDim, activations.bf_pre_ffw_rms_out.data() + token_idx * kModelDim,
layer_weights->pre_ffw_norm_scale.data(), kModelDim);
activations.bf_pre_ffw_rms_out.data() + token_idx * kModelDim, FFW<kBatchSize>(activations, token_idx, layer_weights, pool);
kModelDim); AddFrom(activations.ffw_out.data() + token_idx * kModelDim,
FFW<kBatchSize>(activations, token_idx, layer_weights, inner_pool); activations.x.data() + token_idx * kModelDim, kModelDim);
AddFrom(activations.ffw_out.data() + token_idx * kModelDim, }
activations.x.data() + token_idx * kModelDim, kModelDim);
});
} // foreach layer } // foreach layer
pool.Run( pool.Run(
@ -950,8 +946,7 @@ HWY_NOINLINE void Prefill(const int* tokens, size_t num_tokens, size_t pos,
template <typename WeightArrayT, class TConfig> template <typename WeightArrayT, class TConfig>
void Transformer(int token, size_t pos, const WeightArrayT& weights, void Transformer(int token, size_t pos, const WeightArrayT& weights,
Activations<TConfig, 1>& activations, KVCache& kv_cache, Activations<TConfig, 1>& activations, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, LayersOutputT* layers_output) {
LayersOutputT* layers_output) {
if (layers_output != nullptr) { if (layers_output != nullptr) {
float token_f = token; float token_f = token;
(*layers_output)(pos, "Tokens", &token_f, 1); (*layers_output)(pos, "Tokens", &token_f, 1);
@ -1033,8 +1028,7 @@ template <class TConfig>
void GenerateImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens, void GenerateImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens,
size_t max_generated_tokens, float temperature, size_t max_generated_tokens, float temperature,
const std::vector<int>& prompt, size_t pos, KVCache& kv_cache, const std::vector<int>& prompt, size_t pos, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, const StreamFunc& stream_token,
const StreamFunc& stream_token,
const AcceptFunc& accept_token, std::mt19937& gen, const AcceptFunc& accept_token, std::mt19937& gen,
int verbosity, LayersOutputT* layers_output) { int verbosity, LayersOutputT* layers_output) {
static constexpr size_t kVocabSize = TConfig::kVocabSize; static constexpr size_t kVocabSize = TConfig::kVocabSize;
@ -1077,7 +1071,7 @@ void GenerateImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens,
HWY_DASSERT(pos_offset + batch_size <= prompt_size - 1); HWY_DASSERT(pos_offset + batch_size <= prompt_size - 1);
const int* batch_tokens = prompt.data() + pos_offset; const int* batch_tokens = prompt.data() + pos_offset;
Prefill<kPrefillBatchSize>(batch_tokens, batch_size, pos, weights, Prefill<kPrefillBatchSize>(batch_tokens, batch_size, pos, weights,
prefill_activations, kv_cache, pool, inner_pool); prefill_activations, kv_cache, pool);
for (size_t idx = 0; idx < batch_size; ++idx) { for (size_t idx = 0; idx < batch_size; ++idx) {
if (!stream_token(batch_tokens[idx], 0.0f)) return; if (!stream_token(batch_tokens[idx], 0.0f)) return;
} }
@ -1105,7 +1099,7 @@ void GenerateImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens,
pos < max_tokens && generate_pos < max_generated_tokens; pos < max_tokens && generate_pos < max_generated_tokens;
++pos, ++pos_offset, ++generate_pos) { ++pos, ++pos_offset, ++generate_pos) {
const bool is_generating_phase = pos_offset >= prompt_size - 1; const bool is_generating_phase = pos_offset >= prompt_size - 1;
Transformer(token, pos, weights, activations, kv_cache, pool, inner_pool, Transformer(token, pos, weights, activations, kv_cache, pool,
layers_output); layers_output);
float* final_activation = activations.x.data(); float* final_activation = activations.x.data();
// The condition below is always true if we are doing Prefill above. // The condition below is always true if we are doing Prefill above.
@ -1114,9 +1108,9 @@ void GenerateImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens,
if (is_generating_phase) { if (is_generating_phase) {
PROFILER_ZONE("Gen.Embedding"); PROFILER_ZONE("Gen.Embedding");
// Generation phase // Generation phase
MatVec<kVocabSize, TConfig::kModelDim>(weights.embedder_input_embedding, MatVec<kVocabSize, TConfig::kModelDim>(
0, final_activation, weights.embedder_input_embedding, 0, final_activation,
activations.logits.data(), pool); activations.even_odd.data(), activations.logits.data(), pool);
// Barrier: must have all logits so we can subtract max. // Barrier: must have all logits so we can subtract max.
Softmax(activations.logits.data(), kVocabSize); Softmax(activations.logits.data(), kVocabSize);
token = SampleTopK<TConfig::kTopK>(activations.logits.data(), kVocabSize, token = SampleTopK<TConfig::kTopK>(activations.logits.data(), kVocabSize,
@ -1171,8 +1165,7 @@ void LogTopK(GemmaImpl<TConfig>& gemma, float* logits, float* dist, size_t len,
template <class TConfig> template <class TConfig>
float ComputeCrossEntropyImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens, float ComputeCrossEntropyImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens,
const std::vector<int>& prompt, KVCache& kv_cache, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& pool, int verbosity) {
hwy::ThreadPool& inner_pool, int verbosity) {
static constexpr size_t kModelDim = TConfig::kModelDim; static constexpr size_t kModelDim = TConfig::kModelDim;
static constexpr size_t kVocabSize = TConfig::kVocabSize; static constexpr size_t kVocabSize = TConfig::kVocabSize;
Activations<TConfig, 1>& activations = *gemma.state.get(); Activations<TConfig, 1>& activations = *gemma.state.get();
@ -1196,11 +1189,11 @@ float ComputeCrossEntropyImpl(GemmaImpl<TConfig>& gemma, size_t max_tokens,
printf("Processed %zu tokens, cross-entropy per token: %f\n", pos + 1, printf("Processed %zu tokens, cross-entropy per token: %f\n", pos + 1,
total_entropy / std::log(2.0) / (pos + 1)); total_entropy / std::log(2.0) / (pos + 1));
} }
Transformer(token, pos, weights, activations, kv_cache, pool, inner_pool, Transformer(token, pos, weights, activations, kv_cache, pool,
/*layers_output=*/nullptr); /*layers_output=*/nullptr);
MatVec<kVocabSize, kModelDim>(weights.embedder_input_embedding, 0, MatVec<kVocabSize, kModelDim>(
activations.x.data(), weights.embedder_input_embedding, 0, activations.x.data(),
activations.logits.data(), pool); activations.even_odd.data(), activations.logits.data(), pool);
LogitsSoftCap(30.0f, activations.logits.data(), kVocabSize); LogitsSoftCap(30.0f, activations.logits.data(), kVocabSize);
memcpy(logits.data(), activations.logits.data(), memcpy(logits.data(), activations.logits.data(),
kVocabSize * sizeof(logits[0])); kVocabSize * sizeof(logits[0]));
@ -1215,62 +1208,59 @@ void Generate2B(GemmaImpl<ConfigGemma2B>& gemma, size_t max_tokens,
size_t max_generated_tokens, float temperature, size_t max_generated_tokens, float temperature,
const std::vector<int>& prompt, size_t start_pos, const std::vector<int>& prompt, size_t start_pos,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool, const StreamFunc& stream_token, const StreamFunc& stream_token, const AcceptFunc& accept_token,
const AcceptFunc& accept_token, std::mt19937& gen, std::mt19937& gen, int verbosity,
int verbosity, LayersOutputT* layers_output) { LayersOutputT* layers_output) {
GenerateImpl(gemma, max_tokens, max_generated_tokens, temperature, prompt, GenerateImpl(gemma, max_tokens, max_generated_tokens, temperature, prompt,
start_pos, kv_cache, pool, inner_pool, stream_token, start_pos, kv_cache, pool, stream_token, accept_token, gen,
accept_token, gen, verbosity, layers_output); verbosity, layers_output);
} }
void Generate7B(GemmaImpl<ConfigGemma7B>& gemma, size_t max_tokens, void Generate7B(GemmaImpl<ConfigGemma7B>& gemma, size_t max_tokens,
size_t max_generated_tokens, float temperature, size_t max_generated_tokens, float temperature,
const std::vector<int>& prompt, size_t start_pos, const std::vector<int>& prompt, size_t start_pos,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool, const StreamFunc& stream_token, const StreamFunc& stream_token, const AcceptFunc& accept_token,
const AcceptFunc& accept_token, std::mt19937& gen, std::mt19937& gen, int verbosity,
int verbosity, LayersOutputT* layers_output) { LayersOutputT* layers_output) {
GenerateImpl(gemma, max_tokens, max_generated_tokens, temperature, prompt, GenerateImpl(gemma, max_tokens, max_generated_tokens, temperature, prompt,
start_pos, kv_cache, pool, inner_pool, stream_token, start_pos, kv_cache, pool, stream_token, accept_token, gen,
accept_token, gen, verbosity, layers_output); verbosity, layers_output);
} }
void GenerateGriffin2B(GemmaImpl<ConfigGriffin2B>& gemma, size_t max_tokens, void GenerateGriffin2B(GemmaImpl<ConfigGriffin2B>& gemma, size_t max_tokens,
size_t max_generated_tokens, float temperature, size_t max_generated_tokens, float temperature,
const std::vector<int>& prompt, size_t start_pos, const std::vector<int>& prompt, size_t start_pos,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool,
const StreamFunc& stream_token, const StreamFunc& stream_token,
const AcceptFunc& accept_token, std::mt19937& gen, const AcceptFunc& accept_token, std::mt19937& gen,
int verbosity, LayersOutputT* layers_output) { int verbosity, LayersOutputT* layers_output) {
GenerateImpl(gemma, max_tokens, max_generated_tokens, temperature, prompt, GenerateImpl(gemma, max_tokens, max_generated_tokens, temperature, prompt,
start_pos, kv_cache, pool, inner_pool, stream_token, start_pos, kv_cache, pool, stream_token, accept_token, gen,
accept_token, gen, verbosity, layers_output); verbosity, layers_output);
} }
float ComputeCrossEntropy2B(GemmaImpl<ConfigGemma2B>& gemma, size_t max_tokens, float ComputeCrossEntropy2B(GemmaImpl<ConfigGemma2B>& gemma, size_t max_tokens,
const std::vector<int>& prompt, KVCache& kv_cache, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, int verbosity) {
int verbosity) {
return ComputeCrossEntropyImpl(gemma, max_tokens, prompt, kv_cache, pool, return ComputeCrossEntropyImpl(gemma, max_tokens, prompt, kv_cache, pool,
inner_pool, verbosity); verbosity);
} }
float ComputeCrossEntropy7B(GemmaImpl<ConfigGemma7B>& gemma, size_t max_tokens, float ComputeCrossEntropy7B(GemmaImpl<ConfigGemma7B>& gemma, size_t max_tokens,
const std::vector<int>& prompt, KVCache& kv_cache, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, int verbosity) {
int verbosity) {
return ComputeCrossEntropyImpl(gemma, max_tokens, prompt, kv_cache, pool, return ComputeCrossEntropyImpl(gemma, max_tokens, prompt, kv_cache, pool,
inner_pool, verbosity); verbosity);
} }
float ComputeCrossEntropyGriffin2B(GemmaImpl<ConfigGriffin2B>& gemma, float ComputeCrossEntropyGriffin2B(GemmaImpl<ConfigGriffin2B>& gemma,
size_t max_tokens, size_t max_tokens,
const std::vector<int>& prompt, const std::vector<int>& prompt,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool, int verbosity) { int verbosity) {
return ComputeCrossEntropyImpl(gemma, max_tokens, prompt, kv_cache, pool, return ComputeCrossEntropyImpl(gemma, max_tokens, prompt, kv_cache, pool,
inner_pool, verbosity); verbosity);
} }
// Calls func(name, float*, CompressedArray&) for each tensor. float* is null // Calls func(name, float*, CompressedArray&) for each tensor. float* is null
@ -1477,9 +1467,8 @@ KVCache CreateKVCache(size_t size_cache_pos, size_t seq_len,
size_t conv1d_cache_size, size_t rglru_cache_size) { size_t conv1d_cache_size, size_t rglru_cache_size) {
KVCache kv_cache = {}; KVCache kv_cache = {};
if (size_cache_pos != 0) { if (size_cache_pos != 0) {
kv_cache.key_cache = hwy::AllocateAligned<float>(seq_len * size_cache_pos); kv_cache.kv_cache =
kv_cache.value_cache = hwy::AllocateAligned<float>(seq_len * size_cache_pos * 2);
hwy::AllocateAligned<float>(seq_len * size_cache_pos);
} }
if (conv1d_cache_size != 0) { if (conv1d_cache_size != 0) {
kv_cache.conv1d_cache = hwy::AllocateAligned<float>(conv1d_cache_size); kv_cache.conv1d_cache = hwy::AllocateAligned<float>(conv1d_cache_size);
@ -1507,12 +1496,12 @@ template <>
void GemmaImpl<ConfigGemma2B>::Generate( void GemmaImpl<ConfigGemma2B>::Generate(
size_t max_tokens, size_t max_generated_tokens, float temperature, size_t max_tokens, size_t max_generated_tokens, float temperature,
const std::vector<int>& prompt, size_t start_pos, KVCache& kv_cache, const std::vector<int>& prompt, size_t start_pos, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, const StreamFunc& stream_token,
const StreamFunc& stream_token, const AcceptFunc& accept_token, const AcceptFunc& accept_token, std::mt19937& gen, int verbosity,
std::mt19937& gen, int verbosity, LayersOutputT* layers_output) { LayersOutputT* layers_output) {
HWY_DYNAMIC_DISPATCH(Generate2B) HWY_DYNAMIC_DISPATCH(Generate2B)
(*this, max_tokens, max_generated_tokens, temperature, prompt, start_pos, (*this, max_tokens, max_generated_tokens, temperature, prompt, start_pos,
kv_cache, pool, inner_pool, stream_token, accept_token, gen, verbosity, kv_cache, pool, stream_token, accept_token, gen, verbosity,
layers_output); layers_output);
} }
@ -1520,50 +1509,49 @@ template <>
void GemmaImpl<ConfigGemma7B>::Generate( void GemmaImpl<ConfigGemma7B>::Generate(
size_t max_tokens, size_t max_generated_tokens, float temperature, size_t max_tokens, size_t max_generated_tokens, float temperature,
const std::vector<int>& prompt, size_t start_pos, KVCache& kv_cache, const std::vector<int>& prompt, size_t start_pos, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, const StreamFunc& stream_token,
const StreamFunc& stream_token, const AcceptFunc& accept_token, const AcceptFunc& accept_token, std::mt19937& gen, int verbosity,
std::mt19937& gen, int verbosity, LayersOutputT* layers_output) { LayersOutputT* layers_output) {
HWY_DYNAMIC_DISPATCH(Generate7B) HWY_DYNAMIC_DISPATCH(Generate7B)
(*this, max_tokens, max_generated_tokens, temperature, prompt, start_pos, (*this, max_tokens, max_generated_tokens, temperature, prompt, start_pos,
kv_cache, pool, inner_pool, stream_token, accept_token, gen, verbosity, kv_cache, pool, stream_token, accept_token, gen, verbosity, layers_output);
layers_output);
} }
template <> template <>
void GemmaImpl<ConfigGriffin2B>::Generate( void GemmaImpl<ConfigGriffin2B>::Generate(
size_t max_tokens, size_t max_generated_tokens, float temperature, size_t max_tokens, size_t max_generated_tokens, float temperature,
const std::vector<int>& prompt, size_t start_pos, KVCache& kv_cache, const std::vector<int>& prompt, size_t start_pos, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, const StreamFunc& stream_token,
const StreamFunc& stream_token, const AcceptFunc& accept_token, const AcceptFunc& accept_token, std::mt19937& gen, int verbosity,
std::mt19937& gen, int verbosity, LayersOutputT* layers_output) { LayersOutputT* layers_output) {
HWY_DYNAMIC_DISPATCH(GenerateGriffin2B) HWY_DYNAMIC_DISPATCH(GenerateGriffin2B)
(*this, max_tokens, max_generated_tokens, temperature, prompt, start_pos, (*this, max_tokens, max_generated_tokens, temperature, prompt, start_pos,
kv_cache, pool, inner_pool, stream_token, accept_token, gen, verbosity, kv_cache, pool, stream_token, accept_token, gen, verbosity,
layers_output); layers_output);
} }
template <> template <>
float GemmaImpl<ConfigGemma2B>::ComputeCrossEntropy( float GemmaImpl<ConfigGemma2B>::ComputeCrossEntropy(
size_t max_tokens, const std::vector<int>& prompt, KVCache& kv_cache, size_t max_tokens, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, int verbosity) { hwy::ThreadPool& pool, int verbosity) {
return HWY_DYNAMIC_DISPATCH(ComputeCrossEntropy2B)( return HWY_DYNAMIC_DISPATCH(ComputeCrossEntropy2B)(
*this, max_tokens, prompt, kv_cache, pool, inner_pool, verbosity); *this, max_tokens, prompt, kv_cache, pool, verbosity);
} }
template <> template <>
float GemmaImpl<ConfigGemma7B>::ComputeCrossEntropy( float GemmaImpl<ConfigGemma7B>::ComputeCrossEntropy(
size_t max_tokens, const std::vector<int>& prompt, KVCache& kv_cache, size_t max_tokens, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, int verbosity) { hwy::ThreadPool& pool, int verbosity) {
return HWY_DYNAMIC_DISPATCH(ComputeCrossEntropy7B)( return HWY_DYNAMIC_DISPATCH(ComputeCrossEntropy7B)(
*this, max_tokens, prompt, kv_cache, pool, inner_pool, verbosity); *this, max_tokens, prompt, kv_cache, pool, verbosity);
} }
template <> template <>
float GemmaImpl<ConfigGriffin2B>::ComputeCrossEntropy( float GemmaImpl<ConfigGriffin2B>::ComputeCrossEntropy(
size_t max_tokens, const std::vector<int>& prompt, KVCache& kv_cache, size_t max_tokens, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, int verbosity) { hwy::ThreadPool& pool, int verbosity) {
return HWY_DYNAMIC_DISPATCH(ComputeCrossEntropyGriffin2B)( return HWY_DYNAMIC_DISPATCH(ComputeCrossEntropyGriffin2B)(
*this, max_tokens, prompt, kv_cache, pool, inner_pool, verbosity); *this, max_tokens, prompt, kv_cache, pool, verbosity);
} }
Gemma::Gemma(const Path& tokenizer_path, const Path& weights, Model model_type, Gemma::Gemma(const Path& tokenizer_path, const Path& weights, Model model_type,
@ -1607,13 +1595,13 @@ const GemmaTokenizer* Gemma::Tokenizer() const { return impl_->Tokenizer(); }
void GenerateGemma(Gemma& gemma, size_t max_tokens, size_t max_generated_tokens, void GenerateGemma(Gemma& gemma, size_t max_tokens, size_t max_generated_tokens,
float temperature, const std::vector<int>& prompt, float temperature, const std::vector<int>& prompt,
size_t start_pos, KVCache& kv_cache, hwy::ThreadPool& pool, size_t start_pos, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool, const StreamFunc& stream_token, const StreamFunc& stream_token,
const AcceptFunc& accept_token, std::mt19937& gen, const AcceptFunc& accept_token, std::mt19937& gen,
int verbosity, LayersOutputT* layers_output) { int verbosity, LayersOutputT* layers_output) {
pool.SetWaitMode(hwy::PoolWaitMode::kSpin); pool.SetWaitMode(hwy::PoolWaitMode::kSpin);
gemma.impl_->Generate(max_tokens, max_generated_tokens, temperature, prompt, gemma.impl_->Generate(max_tokens, max_generated_tokens, temperature, prompt,
start_pos, kv_cache, pool, inner_pool, stream_token, start_pos, kv_cache, pool, stream_token, accept_token,
accept_token, gen, verbosity, layers_output); gen, verbosity, layers_output);
pool.SetWaitMode(hwy::PoolWaitMode::kBlock); pool.SetWaitMode(hwy::PoolWaitMode::kBlock);
} }
@ -1621,10 +1609,9 @@ void GenerateGemma(Gemma& gemma, RuntimeConfig runtime_config,
const std::vector<int>& prompt, size_t start_pos, const std::vector<int>& prompt, size_t start_pos,
KVCache& kv_cache, hwy::ThreadPool& pool, KVCache& kv_cache, hwy::ThreadPool& pool,
const StreamFunc& stream_token, std::mt19937& gen) { const StreamFunc& stream_token, std::mt19937& gen) {
hwy::ThreadPool inner_pool(0);
GenerateGemma( GenerateGemma(
gemma, runtime_config.max_tokens, runtime_config.max_generated_tokens, gemma, runtime_config.max_tokens, runtime_config.max_generated_tokens,
runtime_config.temperature, prompt, start_pos, kv_cache, pool, inner_pool, runtime_config.temperature, prompt, start_pos, kv_cache, pool,
stream_token, [](int) { return true; }, gen, runtime_config.verbosity, stream_token, [](int) { return true; }, gen, runtime_config.verbosity,
/*layers_output=*/nullptr); /*layers_output=*/nullptr);
} }
@ -1637,11 +1624,10 @@ void CompressWeights(gcpp::Model model, const Path& weights,
float ComputeCrossEntropy(Gemma& gemma, size_t max_tokens, float ComputeCrossEntropy(Gemma& gemma, size_t max_tokens,
const std::vector<int>& prompt, KVCache& kv_cache, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, int verbosity) {
int verbosity) {
pool.SetWaitMode(hwy::PoolWaitMode::kSpin); pool.SetWaitMode(hwy::PoolWaitMode::kSpin);
const float result = gemma.impl_->ComputeCrossEntropy( const float result = gemma.impl_->ComputeCrossEntropy(
max_tokens, prompt, kv_cache, pool, inner_pool, verbosity); max_tokens, prompt, kv_cache, pool, verbosity);
pool.SetWaitMode(hwy::PoolWaitMode::kBlock); pool.SetWaitMode(hwy::PoolWaitMode::kBlock);
return result; return result;
} }

10
gemma/gemma.h
View File

@ -44,9 +44,7 @@ constexpr bool kSystemPrompt = false;
struct KVCache { struct KVCache {
hwy::AlignedFreeUniquePtr<float[]> hwy::AlignedFreeUniquePtr<float[]>
key_cache; // kSeqLen * kGemmaLayers * kKVHeads * kQKVDim kv_cache; // kSeqLen * kGemmaLayers * kKVHeads * kQKVDim * 2
hwy::AlignedFreeUniquePtr<float[]>
value_cache; // kSeqLen * kGemmaLayers * kKVHeads * kQKVDim
hwy::AlignedFreeUniquePtr<float[]> hwy::AlignedFreeUniquePtr<float[]>
conv1d_cache; // (kConv1dWidth - 1) * kModelDim * kGriffinLayers conv1d_cache; // (kConv1dWidth - 1) * kModelDim * kGriffinLayers
hwy::AlignedFreeUniquePtr<float[]> hwy::AlignedFreeUniquePtr<float[]>
@ -104,13 +102,12 @@ using AcceptFunc = std::function<bool(int)>;
void GenerateGemma(Gemma& gemma, size_t max_tokens, size_t max_generated_tokens, void GenerateGemma(Gemma& gemma, size_t max_tokens, size_t max_generated_tokens,
float temperature, const std::vector<int>& prompt, float temperature, const std::vector<int>& prompt,
size_t start_pos, KVCache& kv_cache, hwy::ThreadPool& pool, size_t start_pos, KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool, const StreamFunc& stream_token, const StreamFunc& stream_token,
const AcceptFunc& accept_token, std::mt19937& gen, const AcceptFunc& accept_token, std::mt19937& gen,
int verbosity, LayersOutputT* layers_output = nullptr); int verbosity, LayersOutputT* layers_output = nullptr);
// Convenience function for the common case: // Convenience function for the common case:
// - Bundle runtime parameters as RuntimeConfig // - Bundle runtime parameters as RuntimeConfig
// - No ThreadPool within ThreadPool (inner_pool = dummy)
// - All tokens accepted // - All tokens accepted
void GenerateGemma(Gemma& gemma, RuntimeConfig runtime_config, void GenerateGemma(Gemma& gemma, RuntimeConfig runtime_config,
const std::vector<int>& prompt, size_t start_pos, const std::vector<int>& prompt, size_t start_pos,
@ -122,8 +119,7 @@ void CompressWeights(gcpp::Model model, const Path& weights,
float ComputeCrossEntropy(Gemma& gemma, size_t max_tokens, float ComputeCrossEntropy(Gemma& gemma, size_t max_tokens,
const std::vector<int>& prompt, KVCache& kv_cache, const std::vector<int>& prompt, KVCache& kv_cache,
hwy::ThreadPool& pool, hwy::ThreadPool& inner_pool, hwy::ThreadPool& pool, int verbosity);
int verbosity);
constexpr int EOS_ID = 1; constexpr int EOS_ID = 1;

9
gemma/gemma_test.cc
View File

@ -36,7 +36,6 @@ class GemmaTest : public ::testing::Test {
: weights("./2b-it-mqa.sbs"), : weights("./2b-it-mqa.sbs"),
tokenizer("./tokenizer.spm"), tokenizer("./tokenizer.spm"),
pool(std::min<int>(20, (std::thread::hardware_concurrency() - 1) / 2)), pool(std::min<int>(20, (std::thread::hardware_concurrency() - 1) / 2)),
inner_pool(0),
model_type(gcpp::Model::GEMMA_2B), model_type(gcpp::Model::GEMMA_2B),
model(tokenizer, weights, model_type, pool) { model(tokenizer, weights, model_type, pool) {
kv_cache = CreateKVCache(model_type); kv_cache = CreateKVCache(model_type);
@ -60,8 +59,8 @@ class GemmaTest : public ::testing::Test {
gcpp::GenerateGemma( gcpp::GenerateGemma(
model, /*max_tokens=*/3072, /*max_generated_tokens=*/2048, model, /*max_tokens=*/3072, /*max_generated_tokens=*/2048,
/*temperature=*/1.0, prompt, /*start_pos=*/0, kv_cache, pool, /*temperature=*/1.0, prompt, /*start_pos=*/0, kv_cache, pool,
inner_pool, stream_token, stream_token, /*accept=*/[](int) { return true; }, gen,
/*accept=*/[](int) { return true; }, gen, /*verbosity=*/0); /*verbosity=*/0);
std::string response_text; std::string response_text;
HWY_ASSERT(model.Tokenizer()->Decode(response, &response_text)); HWY_ASSERT(model.Tokenizer()->Decode(response, &response_text));
return response_text; return response_text;
@ -71,8 +70,7 @@ class GemmaTest : public ::testing::Test {
std::vector<int> prompt; std::vector<int> prompt;
HWY_ASSERT(model.Tokenizer()->Encode(prompt_string, &prompt)); HWY_ASSERT(model.Tokenizer()->Encode(prompt_string, &prompt));
return gcpp::ComputeCrossEntropy(model, /*max_tokens=*/3072, prompt, return gcpp::ComputeCrossEntropy(model, /*max_tokens=*/3072, prompt,
kv_cache, pool, inner_pool, kv_cache, pool, /*verbosity=*/0) /
/*verbosity=*/0) /
prompt_string.size(); prompt_string.size();
} }
@ -89,7 +87,6 @@ class GemmaTest : public ::testing::Test {
gcpp::Path tokenizer; gcpp::Path tokenizer;
gcpp::KVCache kv_cache; gcpp::KVCache kv_cache;
hwy::ThreadPool pool; hwy::ThreadPool pool;
hwy::ThreadPool inner_pool;
gcpp::Model model_type = {}; gcpp::Model model_type = {};
gcpp::Gemma model; gcpp::Gemma model;
}; };

58
gemma/ops.h
View File

@ -129,11 +129,13 @@ HWY_INLINE void ToEvenOddF32(
} }
// Simple version without tiling nor threading. // Simple version without tiling nor threading.
// even_odd is precomputed for the current thread.
template <bool kAdd, size_t kOuter, size_t kInner, typename ArrayT, template <bool kAdd, size_t kOuter, size_t kInner, typename ArrayT,
typename VecT, typename AddT> typename VecT, typename AddT>
HWY_INLINE void MatVecAddLoop(const ArrayT& mat, const size_t mat_ofs, HWY_INLINE void MatVecAddLoop(const ArrayT& mat, const size_t mat_ofs,
const VecT* HWY_RESTRICT vec_aligned, const VecT* HWY_RESTRICT vec_aligned,
const AddT* HWY_RESTRICT add, const AddT* HWY_RESTRICT add,
float* HWY_RESTRICT even_odd,
float* HWY_RESTRICT out) { float* HWY_RESTRICT out) {
PROFILER_ZONE("MatVecAddLoop"); PROFILER_ZONE("MatVecAddLoop");
const hn::ScalableTag<float> df; const hn::ScalableTag<float> df;
@ -149,7 +151,6 @@ HWY_INLINE void MatVecAddLoop(const ArrayT& mat, const size_t mat_ofs,
} }
} }
#if !defined(HWY_NATIVE_DOT_BF16) || !HWY_NATIVE_DOT_BF16 #if !defined(HWY_NATIVE_DOT_BF16) || !HWY_NATIVE_DOT_BF16
template <bool kAdd, size_t kOuter, size_t kInner, typename VecT, typename AddT, template <bool kAdd, size_t kOuter, size_t kInner, typename VecT, typename AddT,
size_t kCapacity> size_t kCapacity>
@ -157,33 +158,39 @@ HWY_INLINE void MatVecAddLoop(
const CompressedArray<hwy::bfloat16_t, kCapacity>& mat, const CompressedArray<hwy::bfloat16_t, kCapacity>& mat,
const size_t mat_ofs, const size_t mat_ofs,
const VecT* HWY_RESTRICT vec_aligned, const VecT* HWY_RESTRICT vec_aligned,
const AddT* HWY_RESTRICT add, const AddT* HWY_RESTRICT add, float* HWY_RESTRICT even_odd,
float* HWY_RESTRICT out) { float* HWY_RESTRICT out) {
PROFILER_ZONE("MatVecAddLoop"); PROFILER_ZONE("MatVecAddLoop");
// Sanity check: we can write without race conditions.
if (HWY_IS_TSAN) {
even_odd[0] = hwy::ConvertScalarTo<float>(vec_aligned[0]);
even_odd[kInner - 1] = -even_odd[0];
}
const hn::ScalableTag<float> df; const hn::ScalableTag<float> df;
ToEvenOddF32(vec_aligned, kInner, even_odd);
const auto vec_dequant = hwy::AllocateAligned<float>(kInner);
ToEvenOddF32(vec_aligned, kInner, vec_dequant.get());
for (size_t idx_row = 0; idx_row < kOuter; ++idx_row) { for (size_t idx_row = 0; idx_row < kOuter; ++idx_row) {
const size_t row_ofs = mat_ofs + idx_row * kInner; const size_t row_ofs = mat_ofs + idx_row * kInner;
if constexpr (kAdd) { if constexpr (kAdd) {
out[idx_row] = hwy::ConvertScalarTo<float>(add[idx_row]) + out[idx_row] = hwy::ConvertScalarTo<float>(add[idx_row]) +
Dot<true>(df, mat, row_ofs, vec_dequant.get(), kInner); Dot<true>(df, mat, row_ofs, even_odd, kInner);
} else { } else {
out[idx_row] = Dot<true>(df, mat, row_ofs, vec_dequant.get(), kInner); out[idx_row] = Dot<true>(df, mat, row_ofs, even_odd, kInner);
} }
} }
} }
#endif #endif
// even_odd is precomputed for the current thread.
template <size_t kOuter, size_t kInner, typename ArrayT, typename VecT> template <size_t kOuter, size_t kInner, typename ArrayT, typename VecT>
HWY_INLINE void MatVecLoop(const ArrayT& mat, const size_t mat_ofs, HWY_INLINE void MatVecLoop(const ArrayT& mat, const size_t mat_ofs,
const VecT* HWY_RESTRICT vec_aligned, const VecT* HWY_RESTRICT vec_aligned,
float* HWY_RESTRICT even_odd,
float* HWY_RESTRICT out) { float* HWY_RESTRICT out) {
MatVecAddLoop<false, kOuter, kInner>( MatVecAddLoop</*kAdd=*/false, kOuter, kInner>(
mat, mat_ofs, vec_aligned, /*add=*/static_cast<VecT*>(nullptr), out); mat, mat_ofs, vec_aligned, /*add=*/static_cast<VecT*>(nullptr), even_odd,
out);
} }
// Simple version without tiling nor threading, but two offsets/outputs. // Simple version without tiling nor threading, but two offsets/outputs.
@ -221,7 +228,7 @@ HWY_INLINE void TwoOfsMatVecLoop(const ArrayT& mat, const size_t mat_ofs0,
const VecT* HWY_RESTRICT vec_aligned, const VecT* HWY_RESTRICT vec_aligned,
float* HWY_RESTRICT out0, float* HWY_RESTRICT out0,
float* HWY_RESTRICT out1) { float* HWY_RESTRICT out1) {
TwoOfsMatVecAddLoop<false, kOuter, kInner, ArrayT, VecT, VecT>( TwoOfsMatVecAddLoop</*kAdd=*/false, kOuter, kInner, ArrayT, VecT, VecT>(
mat, mat_ofs0, mat_ofs1, vec_aligned, /*add0=*/nullptr, /*add1=*/nullptr, mat, mat_ofs0, mat_ofs1, vec_aligned, /*add0=*/nullptr, /*add1=*/nullptr,
out0, out1); out0, out1);
} }
@ -307,11 +314,21 @@ template <bool kVecIsEvenOdd, bool kAdd, size_t kOuter, size_t kInner,
HWY_INLINE void MatVecAddInner(const ArrayT& mat, const size_t mat_ofs, HWY_INLINE void MatVecAddInner(const ArrayT& mat, const size_t mat_ofs,
const VecT* HWY_RESTRICT const vec_aligned, const VecT* HWY_RESTRICT const vec_aligned,
const AddT* HWY_RESTRICT const add, const AddT* HWY_RESTRICT const add,
float* HWY_RESTRICT even_odd,
float* HWY_RESTRICT out, hwy::ThreadPool& pool) { float* HWY_RESTRICT out, hwy::ThreadPool& pool) {
const hn::ScalableTag<float> df; const hn::ScalableTag<float> df;
constexpr size_t kRowsPerStrip = RowsPerStrip<kOuter>(); constexpr size_t kRowsPerStrip = RowsPerStrip<kOuter>();
constexpr size_t kNumStrips = kOuter / kRowsPerStrip; constexpr size_t kNumStrips = kOuter / kRowsPerStrip;
// Sanity check: each thread can write without race conditions.
if (HWY_IS_TSAN) {
pool.Run(
0, pool.NumWorkers(), [even_odd](uint64_t /*task*/, size_t thread) {
even_odd[thread * kInner] = -static_cast<float>(thread);
even_odd[thread * kInner + kInner - 1] = static_cast<float>(thread);
});
}
// For each entire strip. // For each entire strip.
pool.Run(0, kNumStrips, [&](const uint64_t strip, size_t thread) HWY_ATTR { pool.Run(0, kNumStrips, [&](const uint64_t strip, size_t thread) HWY_ATTR {
PROFILER_ZONE("MatVec.lambda"); PROFILER_ZONE("MatVec.lambda");
@ -340,6 +357,7 @@ template <bool kAdd, size_t kOuter, size_t kInner, typename ArrayT,
HWY_INLINE void MatVecAdd(const ArrayT& mat, const size_t mat_ofs, HWY_INLINE void MatVecAdd(const ArrayT& mat, const size_t mat_ofs,
const VecT* HWY_RESTRICT const vec_aligned, const VecT* HWY_RESTRICT const vec_aligned,
const AddT* HWY_RESTRICT const add, const AddT* HWY_RESTRICT const add,
float* HWY_RESTRICT even_odd,
float* HWY_RESTRICT out, hwy::ThreadPool& pool) { float* HWY_RESTRICT out, hwy::ThreadPool& pool) {
PROFILER_ZONE("MatVecAdd"); PROFILER_ZONE("MatVecAdd");
@ -352,25 +370,25 @@ HWY_INLINE void MatVecAdd(const ArrayT& mat, const size_t mat_ofs,
CompressTraits<typename ArrayT::value_type>::kSupportsEvenOdd CompressTraits<typename ArrayT::value_type>::kSupportsEvenOdd
&& hwy::IsSameEither<VecT, float, hwy::bfloat16_t>() && hwy::IsSameEither<VecT, float, hwy::bfloat16_t>()
) { ) {
const auto vec_dequant = hwy::AllocateAligned<float>(kInner); ToEvenOddF32(vec_aligned, kInner, even_odd);
ToEvenOddF32(vec_aligned, kInner, vec_dequant.get());
detail::MatVecAddInner<true, kAdd, kOuter, kInner>( detail::MatVecAddInner<true, kAdd, kOuter, kInner>(
mat, mat_ofs, vec_dequant.get(), add, out, pool); mat, mat_ofs, even_odd, add, even_odd, out, pool);
return; return;
} }
#endif #endif
detail::MatVecAddInner<false, kAdd, kOuter, kInner>( detail::MatVecAddInner<false, kAdd, kOuter, kInner>(
mat, mat_ofs, vec_aligned, add, out, pool); mat, mat_ofs, vec_aligned, add, even_odd, out, pool);
} }
template <size_t kOuter, size_t kInner, typename ArrayT, typename VecT> template <size_t kOuter, size_t kInner, typename ArrayT, typename VecT>
HWY_INLINE void MatVec(const ArrayT& mat, const size_t mat_ofs, HWY_INLINE void MatVec(const ArrayT& mat, const size_t mat_ofs,
const VecT* HWY_RESTRICT const vec_aligned, const VecT* HWY_RESTRICT const vec_aligned,
float* HWY_RESTRICT out, hwy::ThreadPool& pool) { float* HWY_RESTRICT even_odd, float* HWY_RESTRICT out,
MatVecAdd<false, kOuter, kInner>( hwy::ThreadPool& pool) {
mat, mat_ofs, vec_aligned, /*add=*/static_cast<VecT*>(nullptr), out, MatVecAdd</*kAdd=*/false, kOuter, kInner>(
pool); mat, mat_ofs, vec_aligned, /*add=*/static_cast<VecT*>(nullptr), even_odd,
out, pool);
} }
template <class D, HWY_IF_F32_D(D)> template <class D, HWY_IF_F32_D(D)>
@ -523,7 +541,7 @@ HWY_NOINLINE void TwoMatVec(const ArrayT& mat0, const ArrayT& mat1,
const VecT* HWY_RESTRICT vec_aligned, const VecT* HWY_RESTRICT vec_aligned,
float* HWY_RESTRICT out0, float* HWY_RESTRICT out1, float* HWY_RESTRICT out0, float* HWY_RESTRICT out1,
hwy::ThreadPool& pool) { hwy::ThreadPool& pool) {
TwoMatVecAdd<false, kOuter, kInner, ArrayT, VecT, VecT>( TwoMatVecAdd</*kAdd=*/false, kOuter, kInner, ArrayT, VecT, VecT>(
mat0, mat1, mat_ofs, vec_aligned, /*add0=*/nullptr, /*add1=*/nullptr, mat0, mat1, mat_ofs, vec_aligned, /*add0=*/nullptr, /*add1=*/nullptr,
out0, out1, pool); out0, out1, pool);
} }

24
gemma/ops_test.cc
View File

@ -17,11 +17,15 @@
#define HWY_DISABLED_TARGETS HWY_SCALAR #define HWY_DISABLED_TARGETS HWY_SCALAR
#endif #endif
#include <algorithm>
#include <array> #include <array>
#include <random> #include <random>
#include <vector>
#include "compression/compress.h"
#include "hwy/aligned_allocator.h" #include "hwy/aligned_allocator.h"
#include "hwy/base.h" #include "hwy/base.h"
#include "hwy/contrib/thread_pool/thread_pool.h"
// clang-format off // clang-format off
#undef HWY_TARGET_INCLUDE #undef HWY_TARGET_INCLUDE
@ -375,6 +379,7 @@ CompressedArray<float, kOuter * kInner> GenerateMat(size_t offset) {
template <size_t length> template <size_t length>
hwy::AlignedFreeUniquePtr<float[]> GenerateVec(size_t offset) { hwy::AlignedFreeUniquePtr<float[]> GenerateVec(size_t offset) {
hwy::AlignedFreeUniquePtr<float[]> vec = hwy::AllocateAligned<float>(length); hwy::AlignedFreeUniquePtr<float[]> vec = hwy::AllocateAligned<float>(length);
HWY_ASSERT(vec);
for (size_t idx = 0; idx < length; idx++) { for (size_t idx = 0; idx < length; idx++) {
vec[idx] = static_cast<float>(idx + offset); vec[idx] = static_cast<float>(idx + offset);
} }
@ -388,8 +393,9 @@ hwy::AlignedFreeUniquePtr<float[]> SimpleMatVecAdd(
const hwy::AlignedFreeUniquePtr<float[]>& add) { const hwy::AlignedFreeUniquePtr<float[]>& add) {
hwy::AlignedFreeUniquePtr<float[]> uncompressed_mat = hwy::AlignedFreeUniquePtr<float[]> uncompressed_mat =
hwy::AllocateAligned<float>(kOuter * kInner); hwy::AllocateAligned<float>(kOuter * kInner);
Decompress(mat, 0, uncompressed_mat.get(), kOuter * kInner);
hwy::AlignedFreeUniquePtr<float[]> out = hwy::AllocateAligned<float>(kOuter); hwy::AlignedFreeUniquePtr<float[]> out = hwy::AllocateAligned<float>(kOuter);
HWY_ASSERT(uncompressed_mat && out);
Decompress(mat, 0, uncompressed_mat.get(), kOuter * kInner);
for (size_t idx_row = 0; idx_row < kOuter; idx_row++) { for (size_t idx_row = 0; idx_row < kOuter; idx_row++) {
out[idx_row] = add[idx_row]; out[idx_row] = add[idx_row];
for (size_t idx_col = 0; idx_col < kInner; idx_col++) { for (size_t idx_col = 0; idx_col < kInner; idx_col++) {
@ -418,12 +424,15 @@ void TestMatVecAdd() {
CompressedArray<float, kOuter * kInner> mat = GenerateMat<kOuter, kInner>(0); CompressedArray<float, kOuter * kInner> mat = GenerateMat<kOuter, kInner>(0);
hwy::AlignedFreeUniquePtr<float[]> vec = GenerateVec<kInner>(0); hwy::AlignedFreeUniquePtr<float[]> vec = GenerateVec<kInner>(0);
hwy::AlignedFreeUniquePtr<float[]> add = GenerateVec<kOuter>(0); hwy::AlignedFreeUniquePtr<float[]> add = GenerateVec<kOuter>(0);
hwy::AlignedFreeUniquePtr<float[]> even_odd =
hwy::AllocateAligned<float>(kInner * pool.NumWorkers());
hwy::AlignedFreeUniquePtr<float[]> expected_out = hwy::AlignedFreeUniquePtr<float[]> expected_out =
SimpleMatVecAdd<kOuter, kInner>(mat, vec, add); SimpleMatVecAdd<kOuter, kInner>(mat, vec, add);
hwy::AlignedFreeUniquePtr<float[]> actual_out = hwy::AlignedFreeUniquePtr<float[]> actual_out =
hwy::AllocateAligned<float>(kOuter); hwy::AllocateAligned<float>(kOuter);
MatVecAdd<true, kOuter, kInner>(mat, 0, vec.get(), add.get(), HWY_ASSERT(vec && add && even_odd && expected_out && actual_out);
actual_out.get(), pool); MatVecAdd</*kAdd=*/true, kOuter, kInner>(
mat, 0, vec.get(), add.get(), even_odd.get(), actual_out.get(), pool);
AssertClose<kOuter>(actual_out, expected_out); AssertClose<kOuter>(actual_out, expected_out);
} }
@ -433,12 +442,15 @@ void TestMatVecAddLoop() {
CompressedArray<float, kOuter * kInner> mat = GenerateMat<kOuter, kInner>(0); CompressedArray<float, kOuter * kInner> mat = GenerateMat<kOuter, kInner>(0);
hwy::AlignedFreeUniquePtr<float[]> vec = GenerateVec<kInner>(0); hwy::AlignedFreeUniquePtr<float[]> vec = GenerateVec<kInner>(0);
hwy::AlignedFreeUniquePtr<float[]> add = GenerateVec<kOuter>(0); hwy::AlignedFreeUniquePtr<float[]> add = GenerateVec<kOuter>(0);
hwy::AlignedFreeUniquePtr<float[]> even_odd =
hwy::AllocateAligned<float>(kInner);
hwy::AlignedFreeUniquePtr<float[]> expected_out = hwy::AlignedFreeUniquePtr<float[]> expected_out =
SimpleMatVecAdd<kOuter, kInner>(mat, vec, add); SimpleMatVecAdd<kOuter, kInner>(mat, vec, add);
hwy::AlignedFreeUniquePtr<float[]> actual_out = hwy::AlignedFreeUniquePtr<float[]> actual_out =
hwy::AllocateAligned<float>(kOuter); hwy::AllocateAligned<float>(kOuter);
HWY_ASSERT(vec && add && even_odd && expected_out && actual_out);
MatVecAddLoop<true, kOuter, kInner>(mat, 0, vec.get(), add.get(), MatVecAddLoop<true, kOuter, kInner>(mat, 0, vec.get(), add.get(),
actual_out.get()); even_odd.get(), actual_out.get());
AssertClose<kOuter>(actual_out, expected_out); AssertClose<kOuter>(actual_out, expected_out);
} }
@ -459,6 +471,8 @@ void TestTwoMatVecAdd() {
hwy::AllocateAligned<float>(kOuter); hwy::AllocateAligned<float>(kOuter);
hwy::AlignedFreeUniquePtr<float[]> actual_out1 = hwy::AlignedFreeUniquePtr<float[]> actual_out1 =
hwy::AllocateAligned<float>(kOuter); hwy::AllocateAligned<float>(kOuter);
HWY_ASSERT(vec && add0 && add1 && expected_out0 && actual_out0 &&
expected_out1 && actual_out1);
TwoMatVecAdd<true, kOuter, kInner>(mat0, mat1, 0, vec.get(), add0.get(), TwoMatVecAdd<true, kOuter, kInner>(mat0, mat1, 0, vec.get(), add0.get(),
add1.get(), actual_out0.get(), add1.get(), actual_out0.get(),
actual_out1.get(), pool); actual_out1.get(), pool);
@ -481,6 +495,8 @@ void TestTwoOfsMatVecAddLoop() {
hwy::AllocateAligned<float>(kOuter); hwy::AllocateAligned<float>(kOuter);
hwy::AlignedFreeUniquePtr<float[]> actual_out1 = hwy::AlignedFreeUniquePtr<float[]> actual_out1 =
hwy::AllocateAligned<float>(kOuter); hwy::AllocateAligned<float>(kOuter);
HWY_ASSERT(vec && add0 && add1 && expected_out0 && actual_out0 &&
expected_out1 && actual_out1);
TwoOfsMatVecAddLoop<true, kOuter, kInner>(mat, 0, 0, vec.get(), add0.get(), TwoOfsMatVecAddLoop<true, kOuter, kInner>(mat, 0, 0, vec.get(), add0.get(),
add1.get(), actual_out0.get(), add1.get(), actual_out0.get(),
actual_out1.get()); actual_out1.get());

11
gemma/run.cc
View File

@ -94,9 +94,8 @@ void ShowHelp(gcpp::LoaderArgs& loader, gcpp::InferenceArgs& inference,
void ReplGemma(gcpp::Gemma& model, ModelTraining training, void ReplGemma(gcpp::Gemma& model, ModelTraining training,
gcpp::KVCache& kv_cache, hwy::ThreadPool& pool, gcpp::KVCache& kv_cache, hwy::ThreadPool& pool,
hwy::ThreadPool& inner_pool, const InferenceArgs& args, const InferenceArgs& args, int verbosity,
int verbosity, const gcpp::AcceptFunc& accept_token, const gcpp::AcceptFunc& accept_token, std::string& eot_line) {
std::string& eot_line) {
PROFILER_ZONE("Gen.misc"); PROFILER_ZONE("Gen.misc");
size_t abs_pos = 0; // absolute token index over all turns size_t abs_pos = 0; // absolute token index over all turns
int current_pos = 0; // token index within the current turn int current_pos = 0; // token index within the current turn
@ -209,7 +208,7 @@ void ReplGemma(gcpp::Gemma& model, ModelTraining training,
const double time_start = hwy::platform::Now(); const double time_start = hwy::platform::Now();
GenerateGemma(model, args.max_tokens, args.max_generated_tokens, GenerateGemma(model, args.max_tokens, args.max_generated_tokens,
args.temperature, prompt, abs_pos, kv_cache, pool, inner_pool, args.temperature, prompt, abs_pos, kv_cache, pool,
stream_token, accept_token, gen, verbosity); stream_token, accept_token, gen, verbosity);
const double time_end = hwy::platform::Now(); const double time_end = hwy::platform::Now();
const double tok_sec = current_pos / (time_end - time_start); const double tok_sec = current_pos / (time_end - time_start);
@ -229,7 +228,6 @@ void ReplGemma(gcpp::Gemma& model, ModelTraining training,
void Run(LoaderArgs& loader, InferenceArgs& inference, AppArgs& app) { void Run(LoaderArgs& loader, InferenceArgs& inference, AppArgs& app) {
PROFILER_ZONE("Run.misc"); PROFILER_ZONE("Run.misc");
hwy::ThreadPool inner_pool(0);
hwy::ThreadPool pool(app.num_threads); hwy::ThreadPool pool(app.num_threads);
// For many-core, pinning threads to cores helps. // For many-core, pinning threads to cores helps.
if (app.num_threads > 10) { if (app.num_threads > 10) {
@ -271,8 +269,7 @@ void Run(LoaderArgs& loader, InferenceArgs& inference, AppArgs& app) {
} }
ReplGemma( ReplGemma(
model, loader.ModelTraining(), kv_cache, pool, inner_pool, inference, model, loader.ModelTraining(), kv_cache, pool, inference, app.verbosity,
app.verbosity,
/*accept_token=*/[](int) { return true; }, app.eot_line); /*accept_token=*/[](int) { return true; }, app.eot_line);
} }

5
util/app.h
View File

@ -96,8 +96,9 @@ class AppArgs : public ArgsBase<AppArgs> {
} }
static inline size_t GetSupportedThreadCount() { static inline size_t GetSupportedThreadCount() {
return static_cast<size_t>(std::clamp( return static_cast<size_t>(
static_cast<int>(std::thread::hardware_concurrency()) - 2, 1, 18)); std::clamp(static_cast<int>(std::thread::hardware_concurrency()) - 2, 1,
HWY_MIN(static_cast<int>(kMaxThreads), 18)));
} }
Path log; // output Path log; // output