Commit aa47c1c5 authored by xuxzh1's avatar xuxzh1 🎱
Browse files

update

parent 0cb78a2f
<?xml version="1.0" encoding="utf-8"?>
<resources>
<color name="purple_200">#FFBB86FC</color>
<color name="purple_500">#FF6200EE</color>
<color name="purple_700">#FF3700B3</color>
<color name="teal_200">#FF03DAC5</color>
<color name="teal_700">#FF018786</color>
<color name="black">#FF000000</color>
<color name="white">#FFFFFFFF</color>
</resources>
<resources>
<string name="app_name">LlamaAndroid</string>
</resources>
<?xml version="1.0" encoding="utf-8"?>
<resources>
<style name="Theme.LlamaAndroid" parent="android:Theme.Material.Light.NoActionBar" />
</resources>
<?xml version="1.0" encoding="utf-8"?><!--
Sample backup rules file; uncomment and customize as necessary.
See https://developer.android.com/guide/topics/data/autobackup
for details.
Note: This file is ignored for devices older that API 31
See https://developer.android.com/about/versions/12/backup-restore
-->
<full-backup-content>
<!--
<include domain="sharedpref" path="."/>
<exclude domain="sharedpref" path="device.xml"/>
-->
</full-backup-content>
<?xml version="1.0" encoding="utf-8"?><!--
Sample data extraction rules file; uncomment and customize as necessary.
See https://developer.android.com/about/versions/12/backup-restore#xml-changes
for details.
-->
<data-extraction-rules>
<cloud-backup>
<!-- TODO: Use <include> and <exclude> to control what is backed up.
<include .../>
<exclude .../>
-->
</cloud-backup>
<!--
<device-transfer>
<include .../>
<exclude .../>
</device-transfer>
-->
</data-extraction-rules>
// Top-level build file where you can add configuration options common to all sub-projects/modules.
plugins {
id("com.android.application") version "8.2.0" apply false
id("org.jetbrains.kotlin.android") version "1.9.0" apply false
id("com.android.library") version "8.2.0" apply false
}
# Project-wide Gradle settings.
# IDE (e.g. Android Studio) users:
# Gradle settings configured through the IDE *will override*
# any settings specified in this file.
# For more details on how to configure your build environment visit
# http://www.gradle.org/docs/current/userguide/build_environment.html
# Specifies the JVM arguments used for the daemon process.
# The setting is particularly useful for tweaking memory settings.
org.gradle.jvmargs=-Xmx2048m -Dfile.encoding=UTF-8
# When configured, Gradle will run in incubating parallel mode.
# This option should only be used with decoupled projects. More details, visit
# http://www.gradle.org/docs/current/userguide/multi_project_builds.html#sec:decoupled_projects
# org.gradle.parallel=true
# AndroidX package structure to make it clearer which packages are bundled with the
# Android operating system, and which are packaged with your app's APK
# https://developer.android.com/topic/libraries/support-library/androidx-rn
android.useAndroidX=true
# Kotlin code style for this project: "official" or "obsolete":
kotlin.code.style=official
# Enables namespacing of each library's R class so that its R class includes only the
# resources declared in the library itself and none from the library's dependencies,
# thereby reducing the size of the R class for that library
android.nonTransitiveRClass=true
#Thu Dec 21 14:31:09 AEDT 2023
distributionBase=GRADLE_USER_HOME
distributionPath=wrapper/dists
distributionUrl=https\://services.gradle.org/distributions/gradle-8.2-bin.zip
zipStoreBase=GRADLE_USER_HOME
zipStorePath=wrapper/dists
#!/usr/bin/env sh
#
# Copyright 2015 the original author or authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
##############################################################################
##
## Gradle start up script for UN*X
##
##############################################################################
# Attempt to set APP_HOME
# Resolve links: $0 may be a link
PRG="$0"
# Need this for relative symlinks.
while [ -h "$PRG" ] ; do
ls=`ls -ld "$PRG"`
link=`expr "$ls" : '.*-> \(.*\)$'`
if expr "$link" : '/.*' > /dev/null; then
PRG="$link"
else
PRG=`dirname "$PRG"`"/$link"
fi
done
SAVED="`pwd`"
cd "`dirname \"$PRG\"`/" >/dev/null
APP_HOME="`pwd -P`"
cd "$SAVED" >/dev/null
APP_NAME="Gradle"
APP_BASE_NAME=`basename "$0"`
# Add default JVM options here. You can also use JAVA_OPTS and GRADLE_OPTS to pass JVM options to this script.
DEFAULT_JVM_OPTS='"-Xmx64m" "-Xms64m"'
# Use the maximum available, or set MAX_FD != -1 to use that value.
MAX_FD="maximum"
warn () {
echo "$*"
}
die () {
echo
echo "$*"
echo
exit 1
}
# OS specific support (must be 'true' or 'false').
cygwin=false
msys=false
darwin=false
nonstop=false
case "`uname`" in
CYGWIN* )
cygwin=true
;;
Darwin* )
darwin=true
;;
MINGW* )
msys=true
;;
NONSTOP* )
nonstop=true
;;
esac
CLASSPATH=$APP_HOME/gradle/wrapper/gradle-wrapper.jar
# Determine the Java command to use to start the JVM.
if [ -n "$JAVA_HOME" ] ; then
if [ -x "$JAVA_HOME/jre/sh/java" ] ; then
# IBM's JDK on AIX uses strange locations for the executables
JAVACMD="$JAVA_HOME/jre/sh/java"
else
JAVACMD="$JAVA_HOME/bin/java"
fi
if [ ! -x "$JAVACMD" ] ; then
die "ERROR: JAVA_HOME is set to an invalid directory: $JAVA_HOME
Please set the JAVA_HOME variable in your environment to match the
location of your Java installation."
fi
else
JAVACMD="java"
which java >/dev/null 2>&1 || die "ERROR: JAVA_HOME is not set and no 'java' command could be found in your PATH.
Please set the JAVA_HOME variable in your environment to match the
location of your Java installation."
fi
# Increase the maximum file descriptors if we can.
if [ "$cygwin" = "false" -a "$darwin" = "false" -a "$nonstop" = "false" ] ; then
MAX_FD_LIMIT=`ulimit -H -n`
if [ $? -eq 0 ] ; then
if [ "$MAX_FD" = "maximum" -o "$MAX_FD" = "max" ] ; then
MAX_FD="$MAX_FD_LIMIT"
fi
ulimit -n $MAX_FD
if [ $? -ne 0 ] ; then
warn "Could not set maximum file descriptor limit: $MAX_FD"
fi
else
warn "Could not query maximum file descriptor limit: $MAX_FD_LIMIT"
fi
fi
# For Darwin, add options to specify how the application appears in the dock
if $darwin; then
GRADLE_OPTS="$GRADLE_OPTS \"-Xdock:name=$APP_NAME\" \"-Xdock:icon=$APP_HOME/media/gradle.icns\""
fi
# For Cygwin or MSYS, switch paths to Windows format before running java
if [ "$cygwin" = "true" -o "$msys" = "true" ] ; then
APP_HOME=`cygpath --path --mixed "$APP_HOME"`
CLASSPATH=`cygpath --path --mixed "$CLASSPATH"`
JAVACMD=`cygpath --unix "$JAVACMD"`
# We build the pattern for arguments to be converted via cygpath
ROOTDIRSRAW=`find -L / -maxdepth 1 -mindepth 1 -type d 2>/dev/null`
SEP=""
for dir in $ROOTDIRSRAW ; do
ROOTDIRS="$ROOTDIRS$SEP$dir"
SEP="|"
done
OURCYGPATTERN="(^($ROOTDIRS))"
# Add a user-defined pattern to the cygpath arguments
if [ "$GRADLE_CYGPATTERN" != "" ] ; then
OURCYGPATTERN="$OURCYGPATTERN|($GRADLE_CYGPATTERN)"
fi
# Now convert the arguments - kludge to limit ourselves to /bin/sh
i=0
for arg in "$@" ; do
CHECK=`echo "$arg"|egrep -c "$OURCYGPATTERN" -`
CHECK2=`echo "$arg"|egrep -c "^-"` ### Determine if an option
if [ $CHECK -ne 0 ] && [ $CHECK2 -eq 0 ] ; then ### Added a condition
eval `echo args$i`=`cygpath --path --ignore --mixed "$arg"`
else
eval `echo args$i`="\"$arg\""
fi
i=`expr $i + 1`
done
case $i in
0) set -- ;;
1) set -- "$args0" ;;
2) set -- "$args0" "$args1" ;;
3) set -- "$args0" "$args1" "$args2" ;;
4) set -- "$args0" "$args1" "$args2" "$args3" ;;
5) set -- "$args0" "$args1" "$args2" "$args3" "$args4" ;;
6) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" ;;
7) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" ;;
8) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" ;;
9) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" "$args8" ;;
esac
fi
# Escape application args
save () {
for i do printf %s\\n "$i" | sed "s/'/'\\\\''/g;1s/^/'/;\$s/\$/' \\\\/" ; done
echo " "
}
APP_ARGS=`save "$@"`
# Collect all arguments for the java command, following the shell quoting and substitution rules
eval set -- $DEFAULT_JVM_OPTS $JAVA_OPTS $GRADLE_OPTS "\"-Dorg.gradle.appname=$APP_BASE_NAME\"" -classpath "\"$CLASSPATH\"" org.gradle.wrapper.GradleWrapperMain "$APP_ARGS"
exec "$JAVACMD" "$@"
plugins {
id("com.android.library")
id("org.jetbrains.kotlin.android")
}
android {
namespace = "android.llama.cpp"
compileSdk = 34
defaultConfig {
minSdk = 33
testInstrumentationRunner = "androidx.test.runner.AndroidJUnitRunner"
consumerProguardFiles("consumer-rules.pro")
ndk {
// Add NDK properties if wanted, e.g.
// abiFilters += listOf("arm64-v8a")
}
externalNativeBuild {
cmake {
arguments += "-DCMAKE_BUILD_TYPE=Release"
cppFlags += listOf()
arguments += listOf()
cppFlags("")
}
}
}
buildTypes {
release {
isMinifyEnabled = false
proguardFiles(
getDefaultProguardFile("proguard-android-optimize.txt"),
"proguard-rules.pro"
)
}
}
externalNativeBuild {
cmake {
path("src/main/cpp/CMakeLists.txt")
version = "3.22.1"
}
}
compileOptions {
sourceCompatibility = JavaVersion.VERSION_1_8
targetCompatibility = JavaVersion.VERSION_1_8
}
kotlinOptions {
jvmTarget = "1.8"
}
packaging {
resources {
excludes += "/META-INF/{AL2.0,LGPL2.1}"
}
}
}
dependencies {
implementation("androidx.core:core-ktx:1.12.0")
implementation("androidx.appcompat:appcompat:1.6.1")
implementation("com.google.android.material:material:1.11.0")
testImplementation("junit:junit:4.13.2")
androidTestImplementation("androidx.test.ext:junit:1.1.5")
androidTestImplementation("androidx.test.espresso:espresso-core:3.5.1")
}
# Add project specific ProGuard rules here.
# You can control the set of applied configuration files using the
# proguardFiles setting in build.gradle.
#
# For more details, see
# http://developer.android.com/guide/developing/tools/proguard.html
# If your project uses WebView with JS, uncomment the following
# and specify the fully qualified class name to the JavaScript interface
# class:
#-keepclassmembers class fqcn.of.javascript.interface.for.webview {
# public *;
#}
# Uncomment this to preserve the line number information for
# debugging stack traces.
#-keepattributes SourceFile,LineNumberTable
# If you keep the line number information, uncomment this to
# hide the original source file name.
#-renamesourcefileattribute SourceFile
package android.llama.cpp
import androidx.test.platform.app.InstrumentationRegistry
import androidx.test.ext.junit.runners.AndroidJUnit4
import org.junit.Test
import org.junit.runner.RunWith
import org.junit.Assert.*
/**
* Instrumented test, which will execute on an Android device.
*
* See [testing documentation](http://d.android.com/tools/testing).
*/
@RunWith(AndroidJUnit4::class)
class ExampleInstrumentedTest {
@Test
fun useAppContext() {
// Context of the app under test.
val appContext = InstrumentationRegistry.getInstrumentation().targetContext
assertEquals("android.llama.cpp.test", appContext.packageName)
}
}
<?xml version="1.0" encoding="utf-8"?>
<manifest xmlns:android="http://schemas.android.com/apk/res/android">
</manifest>
# For more information about using CMake with Android Studio, read the
# documentation: https://d.android.com/studio/projects/add-native-code.html.
# For more examples on how to use CMake, see https://github.com/android/ndk-samples.
# Sets the minimum CMake version required for this project.
cmake_minimum_required(VERSION 3.22.1)
# Declares the project name. The project name can be accessed via ${ PROJECT_NAME},
# Since this is the top level CMakeLists.txt, the project name is also accessible
# with ${CMAKE_PROJECT_NAME} (both CMake variables are in-sync within the top level
# build script scope).
project("llama-android")
#include(FetchContent)
#FetchContent_Declare(
# llama
# GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
# GIT_TAG master
#)
# Also provides "common"
#FetchContent_MakeAvailable(llama)
# Creates and names a library, sets it as either STATIC
# or SHARED, and provides the relative paths to its source code.
# You can define multiple libraries, and CMake builds them for you.
# Gradle automatically packages shared libraries with your APK.
#
# In this top level CMakeLists.txt, ${CMAKE_PROJECT_NAME} is used to define
# the target library name; in the sub-module's CMakeLists.txt, ${PROJECT_NAME}
# is preferred for the same purpose.
#
#load local llama.cpp
add_subdirectory(../../../../../../ build-llama)
# In order to load a library into your app from Java/Kotlin, you must call
# System.loadLibrary() and pass the name of the library defined here;
# for GameActivity/NativeActivity derived applications, the same library name must be
# used in the AndroidManifest.xml file.
add_library(${CMAKE_PROJECT_NAME} SHARED
# List C/C++ source files with relative paths to this CMakeLists.txt.
llama-android.cpp)
# Specifies libraries CMake should link to your target library. You
# can link libraries from various origins, such as libraries defined in this
# build script, prebuilt third-party libraries, or Android system libraries.
target_link_libraries(${CMAKE_PROJECT_NAME}
# List libraries link to the target library
llama
common
android
log)
#include <android/log.h>
#include <jni.h>
#include <iomanip>
#include <math.h>
#include <string>
#include <unistd.h>
#include "llama.h"
#include "common.h"
// Write C++ code here.
//
// Do not forget to dynamically load the C++ library into your application.
//
// For instance,
//
// In MainActivity.java:
// static {
// System.loadLibrary("llama-android");
// }
//
// Or, in MainActivity.kt:
// companion object {
// init {
// System.loadLibrary("llama-android")
// }
// }
#define TAG "llama-android.cpp"
#define LOGi(...) __android_log_print(ANDROID_LOG_INFO, TAG, __VA_ARGS__)
#define LOGe(...) __android_log_print(ANDROID_LOG_ERROR, TAG, __VA_ARGS__)
jclass la_int_var;
jmethodID la_int_var_value;
jmethodID la_int_var_inc;
std::string cached_token_chars;
bool is_valid_utf8(const char * string) {
if (!string) {
return true;
}
const unsigned char * bytes = (const unsigned char *)string;
int num;
while (*bytes != 0x00) {
if ((*bytes & 0x80) == 0x00) {
// U+0000 to U+007F
num = 1;
} else if ((*bytes & 0xE0) == 0xC0) {
// U+0080 to U+07FF
num = 2;
} else if ((*bytes & 0xF0) == 0xE0) {
// U+0800 to U+FFFF
num = 3;
} else if ((*bytes & 0xF8) == 0xF0) {
// U+10000 to U+10FFFF
num = 4;
} else {
return false;
}
bytes += 1;
for (int i = 1; i < num; ++i) {
if ((*bytes & 0xC0) != 0x80) {
return false;
}
bytes += 1;
}
}
return true;
}
static void log_callback(ggml_log_level level, const char * fmt, void * data) {
if (level == GGML_LOG_LEVEL_ERROR) __android_log_print(ANDROID_LOG_ERROR, TAG, fmt, data);
else if (level == GGML_LOG_LEVEL_INFO) __android_log_print(ANDROID_LOG_INFO, TAG, fmt, data);
else if (level == GGML_LOG_LEVEL_WARN) __android_log_print(ANDROID_LOG_WARN, TAG, fmt, data);
else __android_log_print(ANDROID_LOG_DEFAULT, TAG, fmt, data);
}
extern "C"
JNIEXPORT jlong JNICALL
Java_android_llama_cpp_LLamaAndroid_load_1model(JNIEnv *env, jobject, jstring filename) {
llama_model_params model_params = llama_model_default_params();
auto path_to_model = env->GetStringUTFChars(filename, 0);
LOGi("Loading model from %s", path_to_model);
auto model = llama_load_model_from_file(path_to_model, model_params);
env->ReleaseStringUTFChars(filename, path_to_model);
if (!model) {
LOGe("load_model() failed");
env->ThrowNew(env->FindClass("java/lang/IllegalStateException"), "load_model() failed");
return 0;
}
return reinterpret_cast<jlong>(model);
}
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_free_1model(JNIEnv *, jobject, jlong model) {
llama_free_model(reinterpret_cast<llama_model *>(model));
}
extern "C"
JNIEXPORT jlong JNICALL
Java_android_llama_cpp_LLamaAndroid_new_1context(JNIEnv *env, jobject, jlong jmodel) {
auto model = reinterpret_cast<llama_model *>(jmodel);
if (!model) {
LOGe("new_context(): model cannot be null");
env->ThrowNew(env->FindClass("java/lang/IllegalArgumentException"), "Model cannot be null");
return 0;
}
int n_threads = std::max(1, std::min(8, (int) sysconf(_SC_NPROCESSORS_ONLN) - 2));
LOGi("Using %d threads", n_threads);
llama_context_params ctx_params = llama_context_default_params();
ctx_params.seed = 1234;
ctx_params.n_ctx = 2048;
ctx_params.n_threads = n_threads;
ctx_params.n_threads_batch = n_threads;
llama_context * context = llama_new_context_with_model(model, ctx_params);
if (!context) {
LOGe("llama_new_context_with_model() returned null)");
env->ThrowNew(env->FindClass("java/lang/IllegalStateException"),
"llama_new_context_with_model() returned null)");
return 0;
}
return reinterpret_cast<jlong>(context);
}
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_free_1context(JNIEnv *, jobject, jlong context) {
llama_free(reinterpret_cast<llama_context *>(context));
}
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_backend_1free(JNIEnv *, jobject) {
llama_backend_free();
}
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_log_1to_1android(JNIEnv *, jobject) {
llama_log_set(log_callback, NULL);
}
extern "C"
JNIEXPORT jstring JNICALL
Java_android_llama_cpp_LLamaAndroid_bench_1model(
JNIEnv *env,
jobject,
jlong context_pointer,
jlong model_pointer,
jlong batch_pointer,
jint pp,
jint tg,
jint pl,
jint nr
) {
auto pp_avg = 0.0;
auto tg_avg = 0.0;
auto pp_std = 0.0;
auto tg_std = 0.0;
const auto context = reinterpret_cast<llama_context *>(context_pointer);
const auto model = reinterpret_cast<llama_model *>(model_pointer);
const auto batch = reinterpret_cast<llama_batch *>(batch_pointer);
const int n_ctx = llama_n_ctx(context);
LOGi("n_ctx = %d", n_ctx);
int i, j;
int nri;
for (nri = 0; nri < nr; nri++) {
LOGi("Benchmark prompt processing (pp)");
llama_batch_clear(*batch);
const int n_tokens = pp;
for (i = 0; i < n_tokens; i++) {
llama_batch_add(*batch, 0, i, { 0 }, false);
}
batch->logits[batch->n_tokens - 1] = true;
llama_kv_cache_clear(context);
const auto t_pp_start = ggml_time_us();
if (llama_decode(context, *batch) != 0) {
LOGi("llama_decode() failed during prompt processing");
}
const auto t_pp_end = ggml_time_us();
// bench text generation
LOGi("Benchmark text generation (tg)");
llama_kv_cache_clear(context);
const auto t_tg_start = ggml_time_us();
for (i = 0; i < tg; i++) {
llama_batch_clear(*batch);
for (j = 0; j < pl; j++) {
llama_batch_add(*batch, 0, i, { j }, true);
}
LOGi("llama_decode() text generation: %d", i);
if (llama_decode(context, *batch) != 0) {
LOGi("llama_decode() failed during text generation");
}
}
const auto t_tg_end = ggml_time_us();
llama_kv_cache_clear(context);
const auto t_pp = double(t_pp_end - t_pp_start) / 1000000.0;
const auto t_tg = double(t_tg_end - t_tg_start) / 1000000.0;
const auto speed_pp = double(pp) / t_pp;
const auto speed_tg = double(pl * tg) / t_tg;
pp_avg += speed_pp;
tg_avg += speed_tg;
pp_std += speed_pp * speed_pp;
tg_std += speed_tg * speed_tg;
LOGi("pp %f t/s, tg %f t/s", speed_pp, speed_tg);
}
pp_avg /= double(nr);
tg_avg /= double(nr);
if (nr > 1) {
pp_std = sqrt(pp_std / double(nr - 1) - pp_avg * pp_avg * double(nr) / double(nr - 1));
tg_std = sqrt(tg_std / double(nr - 1) - tg_avg * tg_avg * double(nr) / double(nr - 1));
} else {
pp_std = 0;
tg_std = 0;
}
char model_desc[128];
llama_model_desc(model, model_desc, sizeof(model_desc));
const auto model_size = double(llama_model_size(model)) / 1024.0 / 1024.0 / 1024.0;
const auto model_n_params = double(llama_model_n_params(model)) / 1e9;
const auto backend = "(Android)"; // TODO: What should this be?
std::stringstream result;
result << std::setprecision(2);
result << "| model | size | params | backend | test | t/s |\n";
result << "| --- | --- | --- | --- | --- | --- |\n";
result << "| " << model_desc << " | " << model_size << "GiB | " << model_n_params << "B | " << backend << " | pp " << pp << " | " << pp_avg << " ± " << pp_std << " |\n";
result << "| " << model_desc << " | " << model_size << "GiB | " << model_n_params << "B | " << backend << " | tg " << tg << " | " << tg_avg << " ± " << tg_std << " |\n";
return env->NewStringUTF(result.str().c_str());
}
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_free_1batch(JNIEnv *, jobject, jlong batch_pointer) {
llama_batch_free(*reinterpret_cast<llama_batch *>(batch_pointer));
}
extern "C"
JNIEXPORT jlong JNICALL
Java_android_llama_cpp_LLamaAndroid_new_1batch(JNIEnv *, jobject, jint n_tokens, jint embd, jint n_seq_max) {
// Source: Copy of llama.cpp:llama_batch_init but heap-allocated.
llama_batch *batch = new llama_batch {
0,
nullptr,
nullptr,
nullptr,
nullptr,
nullptr,
nullptr,
0,
0,
0,
};
if (embd) {
batch->embd = (float *) malloc(sizeof(float) * n_tokens * embd);
} else {
batch->token = (llama_token *) malloc(sizeof(llama_token) * n_tokens);
}
batch->pos = (llama_pos *) malloc(sizeof(llama_pos) * n_tokens);
batch->n_seq_id = (int32_t *) malloc(sizeof(int32_t) * n_tokens);
batch->seq_id = (llama_seq_id **) malloc(sizeof(llama_seq_id *) * n_tokens);
for (int i = 0; i < n_tokens; ++i) {
batch->seq_id[i] = (llama_seq_id *) malloc(sizeof(llama_seq_id) * n_seq_max);
}
batch->logits = (int8_t *) malloc(sizeof(int8_t) * n_tokens);
return reinterpret_cast<jlong>(batch);
}
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_backend_1init(JNIEnv *, jobject) {
llama_backend_init();
}
extern "C"
JNIEXPORT jstring JNICALL
Java_android_llama_cpp_LLamaAndroid_system_1info(JNIEnv *env, jobject) {
return env->NewStringUTF(llama_print_system_info());
}
extern "C"
JNIEXPORT jint JNICALL
Java_android_llama_cpp_LLamaAndroid_completion_1init(
JNIEnv *env,
jobject,
jlong context_pointer,
jlong batch_pointer,
jstring jtext,
jint n_len
) {
cached_token_chars.clear();
const auto text = env->GetStringUTFChars(jtext, 0);
const auto context = reinterpret_cast<llama_context *>(context_pointer);
const auto batch = reinterpret_cast<llama_batch *>(batch_pointer);
const auto tokens_list = llama_tokenize(context, text, 1);
auto n_ctx = llama_n_ctx(context);
auto n_kv_req = tokens_list.size() + (n_len - tokens_list.size());
LOGi("n_len = %d, n_ctx = %d, n_kv_req = %d", n_len, n_ctx, n_kv_req);
if (n_kv_req > n_ctx) {
LOGe("error: n_kv_req > n_ctx, the required KV cache size is not big enough");
}
for (auto id : tokens_list) {
LOGi("%s", llama_token_to_piece(context, id).c_str());
}
llama_batch_clear(*batch);
// evaluate the initial prompt
for (auto i = 0; i < tokens_list.size(); i++) {
llama_batch_add(*batch, tokens_list[i], i, { 0 }, false);
}
// llama_decode will output logits only for the last token of the prompt
batch->logits[batch->n_tokens - 1] = true;
if (llama_decode(context, *batch) != 0) {
LOGe("llama_decode() failed");
}
env->ReleaseStringUTFChars(jtext, text);
return batch->n_tokens;
}
extern "C"
JNIEXPORT jstring JNICALL
Java_android_llama_cpp_LLamaAndroid_completion_1loop(
JNIEnv * env,
jobject,
jlong context_pointer,
jlong batch_pointer,
jint n_len,
jobject intvar_ncur
) {
const auto context = reinterpret_cast<llama_context *>(context_pointer);
const auto batch = reinterpret_cast<llama_batch *>(batch_pointer);
const auto model = llama_get_model(context);
if (!la_int_var) la_int_var = env->GetObjectClass(intvar_ncur);
if (!la_int_var_value) la_int_var_value = env->GetMethodID(la_int_var, "getValue", "()I");
if (!la_int_var_inc) la_int_var_inc = env->GetMethodID(la_int_var, "inc", "()V");
auto n_vocab = llama_n_vocab(model);
auto logits = llama_get_logits_ith(context, batch->n_tokens - 1);
std::vector<llama_token_data> candidates;
candidates.reserve(n_vocab);
for (llama_token token_id = 0; token_id < n_vocab; token_id++) {
candidates.emplace_back(llama_token_data{ token_id, logits[token_id], 0.0f });
}
llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false };
// sample the most likely token
const auto new_token_id = llama_sample_token_greedy(context, &candidates_p);
const auto n_cur = env->CallIntMethod(intvar_ncur, la_int_var_value);
if (llama_token_is_eog(model, new_token_id) || n_cur == n_len) {
return nullptr;
}
auto new_token_chars = llama_token_to_piece(context, new_token_id);
cached_token_chars += new_token_chars;
jstring new_token = nullptr;
if (is_valid_utf8(cached_token_chars.c_str())) {
new_token = env->NewStringUTF(cached_token_chars.c_str());
LOGi("cached: %s, new_token_chars: `%s`, id: %d", cached_token_chars.c_str(), new_token_chars.c_str(), new_token_id);
cached_token_chars.clear();
} else {
new_token = env->NewStringUTF("");
}
llama_batch_clear(*batch);
llama_batch_add(*batch, new_token_id, n_cur, { 0 }, true);
env->CallVoidMethod(intvar_ncur, la_int_var_inc);
if (llama_decode(context, *batch) != 0) {
LOGe("llama_decode() returned null");
}
return new_token;
}
extern "C"
JNIEXPORT void JNICALL
Java_android_llama_cpp_LLamaAndroid_kv_1cache_1clear(JNIEnv *, jobject, jlong context) {
llama_kv_cache_clear(reinterpret_cast<llama_context *>(context));
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment