Dual NNUE with L1-128 smallnet

Credit goes to @mstembera for:
- writing the code enabling dual NNUE:
  https://github.com/official-stockfish/Stockfish/pull/4898
- the idea of trying L1-128 trained exclusively on high simple eval
  positions

The L1-128 smallnet is:
- epoch 399 of a single-stage training from scratch
- trained only on positions from filtered data with high material
  difference
  - defined by abs(simple_eval) > 1000

```yaml
experiment-name: 128--S1-only-hse-v2

training-dataset:
  - /data/hse/S3/dfrc99-16tb7p-eval-filt-v2.min.high-simple-eval-1k.binpack
  - /data/hse/S3/leela96-filt-v2.min.high-simple-eval-1k.binpack
  - /data/hse/S3/test80-apr2022-16tb7p.min.high-simple-eval-1k.binpack

  - /data/hse/S7/test60-2020-2tb7p.v6-3072.high-simple-eval-1k.binpack
  - /data/hse/S7/test60-novdec2021-12tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack

  - /data/hse/S7/test77-nov2021-2tb7p.v6-3072.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test77-dec2021-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test77-jan2022-2tb7p.high-simple-eval-1k.binpack

  - /data/hse/S7/test78-jantomay2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test78-juntosep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack

  - /data/hse/S7/test79-apr2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test79-may2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack

  # T80 2022
  - /data/hse/S7/test80-may2022-16tb7p.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-jun2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-jul2022-16tb7p.v6-dd.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-aug2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-sep2022-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-oct2022-16tb7p.v6-dd.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-nov2022-16tb7p-v6-dd.min.high-simple-eval-1k.binpack

  # T80 2023
  - /data/hse/S7/test80-jan2023-3of3-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-feb2023-16tb7p-filter-v6-dd.min-mar2023.unmin.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-mar2023-2tb7p.v6-sk16.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-apr2023-2tb7p-filter-v6-sk16.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-may2023-2tb7p.v6.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-jun2023-2tb7p.v6-3072.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-jul2023-2tb7p.v6-3072.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-aug2023-2tb7p.v6.min.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-sep2023-2tb7p.high-simple-eval-1k.binpack
  - /data/hse/S7/test80-oct2023-2tb7p.high-simple-eval-1k.binpack

start-from-engine-test-net: False

nnue-pytorch-branch: linrock/nnue-pytorch/L1-128
engine-test-branch: linrock/Stockfish/L1-128-nolazy
engine-base-branch: linrock/Stockfish/L1-128

num-epochs: 500
lambda: 1.0
```

Experiment yaml configs converted to easy_train.sh commands with:
https://github.com/linrock/nnue-tools/blob/4339954/yaml_easy_train.py

Binpacks interleaved at training time with:
https://github.com/official-stockfish/nnue-pytorch/pull/259

Data filtered for high simple eval positions with:
https://github.com/linrock/nnue-data/blob/32d6a68/filter_high_simple_eval_plain.py
https://github.com/linrock/Stockfish/blob/61dbfe/src/tools/transform.cpp#L626-L655

Training data can be found at:
https://robotmoon.com/nnue-training-data/

Local elo at 25k nodes per move of
L1-128 smallnet (nnue-only eval) vs. L1-128 trained on standard S1 data:
nn-epoch399.nnue : -318.1 +/- 2.1

Passed STC:
https://tests.stockfishchess.org/tests/view/6574cb9d95ea6ba1fcd49e3b
LLR: 2.93 (-2.94,2.94) <0.00,2.00>
Total: 62432 W: 15875 L: 15521 D: 31036
Ptnml(0-2): 177, 7331, 15872, 7633, 203

Passed LTC:
https://tests.stockfishchess.org/tests/view/6575da2d4d789acf40aaac6e
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 64830 W: 16118 L: 15738 D: 32974
Ptnml(0-2): 43, 7129, 17697, 7497, 49

closes https://github.com/official-stockfish/Stockfish/pulls

Bench: 1330050

Co-Authored-By: mstembera <5421953+mstembera@users.noreply.github.com>
This commit is contained in:
Linmiao Xu
2023-12-02 17:50:32 -08:00
committed by Disservin
parent a5a76a6370
commit 584d9efedc
12 changed files with 293 additions and 194 deletions

View File

@@ -23,6 +23,7 @@
#include <cmath>
#include <cstdlib>
#include <fstream>
#include <initializer_list>
#include <iomanip>
#include <iostream>
#include <sstream>
@@ -31,6 +32,7 @@
#include "incbin/incbin.h"
#include "misc.h"
#include "nnue/evaluate_nnue.h"
#include "nnue/nnue_architecture.h"
#include "position.h"
#include "thread.h"
#include "types.h"
@@ -44,11 +46,15 @@
// const unsigned int gEmbeddedNNUESize; // the size of the embedded file
// Note that this does not work in Microsoft Visual Studio.
#if !defined(_MSC_VER) && !defined(NNUE_EMBEDDING_OFF)
INCBIN(EmbeddedNNUE, EvalFileDefaultName);
INCBIN(EmbeddedNNUEBig, EvalFileDefaultNameBig);
INCBIN(EmbeddedNNUESmall, EvalFileDefaultNameSmall);
#else
const unsigned char gEmbeddedNNUEData[1] = {0x0};
const unsigned char* const gEmbeddedNNUEEnd = &gEmbeddedNNUEData[1];
const unsigned int gEmbeddedNNUESize = 1;
const unsigned char gEmbeddedNNUEBigData[1] = {0x0};
const unsigned char* const gEmbeddedNNUEBigEnd = &gEmbeddedNNUEBigData[1];
const unsigned int gEmbeddedNNUEBigSize = 1;
const unsigned char gEmbeddedNNUESmallData[1] = {0x0};
const unsigned char* const gEmbeddedNNUESmallEnd = &gEmbeddedNNUESmallData[1];
const unsigned int gEmbeddedNNUESmallSize = 1;
#endif
@@ -56,7 +62,9 @@ namespace Stockfish {
namespace Eval {
std::string currentEvalFileName = "None";
std::string currentEvalFileName[2] = {"None", "None"};
const std::string EvFiles[2] = {"EvalFile", "EvalFileSmall"};
const std::string EvFileNames[2] = {EvalFileDefaultNameBig, EvalFileDefaultNameSmall};
// Tries to load a NNUE network at startup time, or when the engine
// receives a UCI command "setoption name EvalFile value nn-[a-z0-9]{12}.nnue"
@@ -67,84 +75,96 @@ std::string currentEvalFileName = "None";
// variable to have the engine search in a special directory in their distro.
void NNUE::init() {
std::string eval_file = std::string(Options["EvalFile"]);
if (eval_file.empty())
eval_file = EvalFileDefaultName;
for (NetSize netSize : {Big, Small})
{
// change after fishtest supports EvalFileSmall
std::string eval_file =
std::string(netSize == Small ? EvalFileDefaultNameSmall : Options[EvFiles[netSize]]);
if (eval_file.empty())
eval_file = EvFileNames[netSize];
#if defined(DEFAULT_NNUE_DIRECTORY)
std::vector<std::string> dirs = {"<internal>", "", CommandLine::binaryDirectory,
stringify(DEFAULT_NNUE_DIRECTORY)};
std::vector<std::string> dirs = {"<internal>", "", CommandLine::binaryDirectory,
stringify(DEFAULT_NNUE_DIRECTORY)};
#else
std::vector<std::string> dirs = {"<internal>", "", CommandLine::binaryDirectory};
std::vector<std::string> dirs = {"<internal>", "", CommandLine::binaryDirectory};
#endif
for (const std::string& directory : dirs)
if (currentEvalFileName != eval_file)
for (const std::string& directory : dirs)
{
if (directory != "<internal>")
if (currentEvalFileName[netSize] != eval_file)
{
std::ifstream stream(directory + eval_file, std::ios::binary);
if (NNUE::load_eval(eval_file, stream))
currentEvalFileName = eval_file;
}
if (directory != "<internal>")
{
std::ifstream stream(directory + eval_file, std::ios::binary);
if (NNUE::load_eval(eval_file, stream, netSize))
currentEvalFileName[netSize] = eval_file;
}
if (directory == "<internal>" && eval_file == EvalFileDefaultName)
{
// C++ way to prepare a buffer for a memory stream
class MemoryBuffer: public std::basic_streambuf<char> {
public:
MemoryBuffer(char* p, size_t n) {
setg(p, p, p + n);
setp(p, p + n);
}
};
if (directory == "<internal>" && eval_file == EvFileNames[netSize])
{
// C++ way to prepare a buffer for a memory stream
class MemoryBuffer: public std::basic_streambuf<char> {
public:
MemoryBuffer(char* p, size_t n) {
setg(p, p, p + n);
setp(p, p + n);
}
};
MemoryBuffer buffer(
const_cast<char*>(reinterpret_cast<const char*>(gEmbeddedNNUEData)),
size_t(gEmbeddedNNUESize));
(void) gEmbeddedNNUEEnd; // Silence warning on unused variable
MemoryBuffer buffer(
const_cast<char*>(reinterpret_cast<const char*>(
netSize == Small ? gEmbeddedNNUESmallData : gEmbeddedNNUEBigData)),
size_t(netSize == Small ? gEmbeddedNNUESmallSize : gEmbeddedNNUEBigSize));
(void) gEmbeddedNNUEBigEnd; // Silence warning on unused variable
(void) gEmbeddedNNUESmallEnd;
std::istream stream(&buffer);
if (NNUE::load_eval(eval_file, stream))
currentEvalFileName = eval_file;
std::istream stream(&buffer);
if (NNUE::load_eval(eval_file, stream, netSize))
currentEvalFileName[netSize] = eval_file;
}
}
}
}
}
// Verifies that the last net used was loaded successfully
void NNUE::verify() {
std::string eval_file = std::string(Options["EvalFile"]);
if (eval_file.empty())
eval_file = EvalFileDefaultName;
if (currentEvalFileName != eval_file)
for (NetSize netSize : {Big, Small})
{
// change after fishtest supports EvalFileSmall
std::string eval_file =
std::string(netSize == Small ? EvalFileDefaultNameSmall : Options[EvFiles[netSize]]);
if (eval_file.empty())
eval_file = EvFileNames[netSize];
std::string msg1 =
"Network evaluation parameters compatible with the engine must be available.";
std::string msg2 = "The network file " + eval_file + " was not loaded successfully.";
std::string msg3 = "The UCI option EvalFile might need to specify the full path, "
"including the directory name, to the network file.";
std::string msg4 = "The default net can be downloaded from: "
"https://tests.stockfishchess.org/api/nn/"
+ std::string(EvalFileDefaultName);
std::string msg5 = "The engine will be terminated now.";
if (currentEvalFileName[netSize] != eval_file)
{
std::string msg1 =
"Network evaluation parameters compatible with the engine must be available.";
std::string msg2 = "The network file " + eval_file + " was not loaded successfully.";
std::string msg3 = "The UCI option EvalFile might need to specify the full path, "
"including the directory name, to the network file.";
std::string msg4 = "The default net can be downloaded from: "
"https://tests.stockfishchess.org/api/nn/"
+ std::string(EvFileNames[netSize]);
std::string msg5 = "The engine will be terminated now.";
sync_cout << "info string ERROR: " << msg1 << sync_endl;
sync_cout << "info string ERROR: " << msg2 << sync_endl;
sync_cout << "info string ERROR: " << msg3 << sync_endl;
sync_cout << "info string ERROR: " << msg4 << sync_endl;
sync_cout << "info string ERROR: " << msg5 << sync_endl;
sync_cout << "info string ERROR: " << msg1 << sync_endl;
sync_cout << "info string ERROR: " << msg2 << sync_endl;
sync_cout << "info string ERROR: " << msg3 << sync_endl;
sync_cout << "info string ERROR: " << msg4 << sync_endl;
sync_cout << "info string ERROR: " << msg5 << sync_endl;
exit(EXIT_FAILURE);
exit(EXIT_FAILURE);
}
sync_cout << "info string NNUE evaluation using " << eval_file << sync_endl;
}
sync_cout << "info string NNUE evaluation using " << eval_file << sync_endl;
}
}
// Returns a static, purely materialistic evaluation of the position from
// the point of view of the given color. It can be divided by PawnValue to get
// an approximation of the material advantage on the board in terms of pawns.
@@ -163,18 +183,19 @@ Value Eval::evaluate(const Position& pos) {
int v;
Color stm = pos.side_to_move();
int shuffling = pos.rule50_count();
int simpleEval = simple_eval(pos, stm) + (int(pos.key() & 7) - 3);
bool lazy = std::abs(simpleEval) >= RookValue + KnightValue + 16 * shuffling * shuffling
+ std::abs(pos.this_thread()->bestValue)
+ std::abs(pos.this_thread()->rootSimpleEval);
int simpleEval = simple_eval(pos, stm);
bool lazy = std::abs(simpleEval) > 2300;
if (lazy)
v = simpleEval;
else
{
int nnueComplexity;
Value nnue = NNUE::evaluate(pos, true, &nnueComplexity);
bool smallNet = std::abs(simpleEval) > 1100;
int nnueComplexity;
Value nnue = smallNet ? NNUE::evaluate<NNUE::Small>(pos, true, &nnueComplexity)
: NNUE::evaluate<NNUE::Big>(pos, true, &nnueComplexity);
int optimism = pos.this_thread()->optimism[stm];
@@ -217,7 +238,7 @@ std::string Eval::trace(Position& pos) {
ss << std::showpoint << std::showpos << std::fixed << std::setprecision(2) << std::setw(15);
Value v;
v = NNUE::evaluate(pos, false);
v = NNUE::evaluate<NNUE::Big>(pos, false);
v = pos.side_to_move() == WHITE ? v : -v;
ss << "NNUE evaluation " << 0.01 * UCI::to_cp(v) << " (white side)\n";