Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: sync llama.cpp #110

Merged
merged 6 commits into from
Jan 24, 2025
Merged

feat: sync llama.cpp #110

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
e711996
feat: sync llama.cpp to b4518
a-ghorbani Jan 20, 2025
d4a10c4
fix: apply patches
a-ghorbani Jan 20, 2025
223aa74
fix: apply api changes
a-ghorbani Jan 21, 2025
fc0b73a
Merge branch 'main' into feat/sync-llama.cpp
jhen0409 Jan 23, 2025
4017a2d
fix: removing +fp16 flag from compiler
a-ghorbani Jan 23, 2025
1f4c1d3
Merge branch 'main' into feat/sync-llama.cpp
jhen0409 Jan 24, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 6 additions & 6 deletions android/src/main/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,13 +87,13 @@ build_library("rnllama" "")

if (${ANDROID_ABI} STREQUAL "arm64-v8a")
# ARM64 targets
build_library("rnllama_v8_4_fp16_dotprod_sve" "-march=armv8.4-a+fp16+dotprod+sve")
build_library("rnllama_v8_4_fp16_dotprod_i8mm_sve" "-march=armv8.4-a+fp16+dotprod+i8mm+sve")
build_library("rnllama_v8_4_fp16_dotprod_i8mm" "-march=armv8.4-a+fp16+dotprod+i8mm")
build_library("rnllama_v8_4_fp16_dotprod" "-march=armv8.4-a+fp16+dotprod")
build_library("rnllama_v8_2_fp16_dotprod" "-march=armv8.2-a+fp16+dotprod")
build_library("rnllama_v8_2_fp16" "-march=armv8.2-a+fp16")
# Removing fp16 for now as it leads to issues with some models like deepseek r1 distills
# https://github.com/mybigday/llama.rn/pull/110#issuecomment-2609918310
build_library("rnllama_v8" "-march=armv8-a")
build_library("rnllama_v8_2" "-march=armv8.2-a")
build_library("rnllama_v8_2_dotprod" "-march=armv8.2-a+dotprod")
build_library("rnllama_v8_2_i8mm" "-march=armv8.2-a+i8mm")
build_library("rnllama_v8_2_dotprod_i8mm" "-march=armv8.2-a+dotprod+i8mm")

# https://github.com/ggerganov/llama.cpp/blob/master/docs/android.md#cross-compile-using-android-ndk
# llama.cpp will deal with the cpu features
Expand Down
42 changes: 17 additions & 25 deletions android/src/main/java/com/rnllama/LlamaContext.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -349,32 +349,24 @@ public void release() {

// TODO: Add runtime check for cpu features
if (LlamaContext.isArm64V8a()) {
if (isAtLeastArmV84 && hasSve && hasI8mm && hasFp16 && hasDotProd) {
Log.d(NAME, "Loading librnllama_v8_4_fp16_dotprod_i8mm_sve.so");
System.loadLibrary("rnllama_v8_4_fp16_dotprod_i8mm_sve");
loadedLibrary = "rnllama_v8_4_fp16_dotprod_i8mm_sve";
} else if (isAtLeastArmV84 && hasSve && hasFp16 && hasDotProd) {
Log.d(NAME, "Loading librnllama_v8_4_fp16_dotprod_sve.so");
System.loadLibrary("rnllama_v8_4_fp16_dotprod_sve");
loadedLibrary = "rnllama_v8_4_fp16_dotprod_sve";
} else if (isAtLeastArmV84 && hasI8mm && hasFp16 && hasDotProd) {
Log.d(NAME, "Loading librnllama_v8_4_fp16_dotprod_i8mm.so");
System.loadLibrary("rnllama_v8_4_fp16_dotprod_i8mm");
loadedLibrary = "rnllama_v8_4_fp16_dotprod_i8mm";
} else if (isAtLeastArmV84 && hasFp16 && hasDotProd) {
Log.d(NAME, "Loading librnllama_v8_4_fp16_dotprod.so");
System.loadLibrary("rnllama_v8_4_fp16_dotprod");
loadedLibrary = "rnllama_v8_4_fp16_dotprod";
} else if (isAtLeastArmV82 && hasFp16 && hasDotProd) {
Log.d(NAME, "Loading librnllama_v8_2_fp16_dotprod.so");
System.loadLibrary("rnllama_v8_2_fp16_dotprod");
loadedLibrary = "rnllama_v8_2_fp16_dotprod";
} else if (isAtLeastArmV82 && hasFp16) {
Log.d(NAME, "Loading librnllama_v8_2_fp16.so");
System.loadLibrary("rnllama_v8_2_fp16");
loadedLibrary = "rnllama_v8_2_fp16";
if (hasDotProd && hasI8mm) {
Log.d(NAME, "Loading librnllama_v8_2_dotprod_i8mm.so");
System.loadLibrary("rnllama_v8_2_dotprod_i8mm");
loadedLibrary = "rnllama_v8_2_dotprod_i8mm";
} else if (hasDotProd) {
Log.d(NAME, "Loading librnllama_v8_2_dotprod.so");
System.loadLibrary("rnllama_v8_2_dotprod");
loadedLibrary = "rnllama_v8_2_dotprod";
} else if (hasI8mm) {
Log.d(NAME, "Loading librnllama_v8_2_i8mm.so");
System.loadLibrary("rnllama_v8_2_i8mm");
loadedLibrary = "rnllama_v8_2_i8mm";
} else if (hasFp16) {
Log.d(NAME, "Loading librnllama_v8_2.so");
System.loadLibrary("rnllama_v8_2");
loadedLibrary = "rnllama_v8_2";
} else {
Log.d(NAME, "Loading librnllama_v8.so");
Log.d(NAME, "Loading default librnllama_v8.so");
System.loadLibrary("rnllama_v8");
loadedLibrary = "rnllama_v8";
}
Expand Down
21 changes: 12 additions & 9 deletions android/src/main/jni.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -336,10 +336,10 @@ Java_com_rnllama_LlamaContext_initContext(
llama_free(llama->ctx);
}

std::vector<common_lora_adapter_info> lora;
std::vector<common_adapter_lora_info> lora;
const char *lora_chars = env->GetStringUTFChars(lora_str, nullptr);
if (lora_chars != nullptr && lora_chars[0] != '\0') {
common_lora_adapter_info la;
common_adapter_lora_info la;
la.path = lora_chars;
la.scale = lora_scaled;
lora.push_back(la);
Expand All @@ -353,7 +353,7 @@ Java_com_rnllama_LlamaContext_initContext(
jstring path = readablemap::getString(env, lora_adapter, "path", nullptr);
if (path != nullptr) {
const char *path_chars = env->GetStringUTFChars(path, nullptr);
common_lora_adapter_info la;
common_adapter_lora_info la;
la.path = path_chars;
la.scale = readablemap::getFloat(env, lora_adapter, "scaled", 1.0f);
lora.push_back(la);
Expand Down Expand Up @@ -400,7 +400,7 @@ Java_com_rnllama_LlamaContext_loadModelDetails(
for (int i = 0; i < count; i++) {
char key[256];
llama_model_meta_key_by_index(llama->model, i, key, sizeof(key));
char val[2048];
char val[4096];
llama_model_meta_val_str_by_index(llama->model, i, val, sizeof(val));

putString(env, meta, key, val);
Expand Down Expand Up @@ -613,9 +613,12 @@ Java_com_rnllama_LlamaContext_doCompletion(
sparams.dry_allowed_length = dry_allowed_length;
sparams.dry_penalty_last_n = dry_penalty_last_n;

const llama_model * model = llama_get_model(llama->ctx);
const llama_vocab * vocab = llama_model_get_vocab(model);

sparams.logit_bias.clear();
if (ignore_eos) {
sparams.logit_bias[llama_token_eos(llama->model)].bias = -INFINITY;
sparams.logit_bias[llama_vocab_eos(vocab)].bias = -INFINITY;
}

// dry break seq
Expand All @@ -634,7 +637,7 @@ Java_com_rnllama_LlamaContext_doCompletion(
sparams.dry_sequence_breakers = dry_sequence_breakers_vector;

// logit bias
const int n_vocab = llama_n_vocab(llama_get_model(llama->ctx));
const int n_vocab = llama_vocab_n_tokens(vocab);
jsize logit_bias_len = env->GetArrayLength(logit_bias);

for (jsize i = 0; i < logit_bias_len; i++) {
Expand Down Expand Up @@ -913,7 +916,7 @@ Java_com_rnllama_LlamaContext_applyLoraAdapters(
auto llama = context_map[(long) context_ptr];

// lora_adapters: ReadableArray<ReadableMap>
std::vector<common_lora_adapter_info> lora_adapters;
std::vector<common_adapter_lora_info> lora_adapters;
int lora_adapters_size = readablearray::size(env, loraAdapters);
for (int i = 0; i < lora_adapters_size; i++) {
jobject lora_adapter = readablearray::getMap(env, loraAdapters, i);
Expand All @@ -922,7 +925,7 @@ Java_com_rnllama_LlamaContext_applyLoraAdapters(
const char *path_chars = env->GetStringUTFChars(path, nullptr);
env->ReleaseStringUTFChars(path, path_chars);
float scaled = readablemap::getFloat(env, lora_adapter, "scaled", 1.0f);
common_lora_adapter_info la;
common_adapter_lora_info la;
la.path = path_chars;
la.scale = scaled;
lora_adapters.push_back(la);
Expand All @@ -947,7 +950,7 @@ Java_com_rnllama_LlamaContext_getLoadedLoraAdapters(
auto llama = context_map[(long) context_ptr];
auto loaded_lora_adapters = llama->getLoadedLoraAdapters();
auto result = createWritableArray(env);
for (common_lora_adapter_info &la : loaded_lora_adapters) {
for (common_adapter_lora_info &la : loaded_lora_adapters) {
auto map = createWriteableMap(env);
putString(env, map, "path", la.path.c_str());
putDouble(env, map, "scaled", la.scale);
Expand Down
Loading
Loading