Merge latest master branch

.github/workflows/build.yml

@@ -60,8 +60,7 @@ jobs:
             -DLLAMA_CURL=ON \
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
-            -DGGML_RPC=ON \
-            -DBUILD_SHARED_LIBS=OFF
+            -DGGML_RPC=ON
           cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
 
       - name: Test
@@ -123,8 +122,7 @@ jobs:
             -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_CURL=ON \
             -DGGML_METAL=OFF \
-            -DGGML_RPC=ON \
-            -DBUILD_SHARED_LIBS=OFF
+            -DGGML_RPC=ON
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
 
       - name: Test
@@ -181,7 +179,7 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON -DGGML_RPC=ON -DBUILD_SHARED_LIBS=OFF
+          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON -DGGML_RPC=ON
           cmake --build . --config Release -j $(nproc)
 
       - name: Test
@@ -651,23 +649,23 @@ jobs:
       matrix:
         include:
           - build: 'noavx-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DBUILD_SHARED_LIBS=ON'
+            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF'
           - build: 'avx2-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON'
           - build: 'avx-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_AVX2=OFF -DBUILD_SHARED_LIBS=ON'
+            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_AVX2=OFF'
           - build: 'avx512-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_AVX512=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_AVX512=ON'
           - build: 'openblas-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_BLAS=ON -DBUILD_SHARED_LIBS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
+            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
           - build: 'kompute-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_KOMPUTE=ON -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_KOMPUTE=ON -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON'
           - build: 'vulkan-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_VULKAN=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_VULKAN=ON'
           - build: 'llvm-arm64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON'
           - build: 'msvc-arm64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=O'
           - build: 'llvm-arm64-opencl-adreno'
             defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/opencl-arm64-release" -DGGML_OPENCL=ON -DGGML_OPENCL_USE_ADRENO_KERNELS=ON'
 
@@ -914,7 +912,7 @@ jobs:
         shell: cmd
         run: |
           call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
-          cmake -S . -B build -G "Ninja Multi-Config" -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=ON -DGGML_RPC=ON
+          cmake -S . -B build -G "Ninja Multi-Config" -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DGGML_RPC=ON
           set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1
           cmake --build build --config Release -j %NINJA_JOBS% -t ggml
           cmake --build build --config Release

README.md

@@ -201,6 +201,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [Paddler](https://github.com/distantmagic/paddler) - Stateful load balancer custom-tailored for llama.cpp
 - [GPUStack](https://github.com/gpustack/gpustack) - Manage GPU clusters for running LLMs
 - [llama_cpp_canister](https://github.com/onicai/llama_cpp_canister) - llama.cpp as a smart contract on the Internet Computer, using WebAssembly
+- [llama-swap](https://github.com/mostlygeek/llama-swap) - transparent proxy that adds automatic model switching with llama-server
 
 </details>
 

common/arg.cpp

@@ -1512,15 +1512,15 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         {"--lora"}, "FNAME",
         "path to LoRA adapter (can be repeated to use multiple adapters)",
         [](common_params & params, const std::string & value) {
-            params.lora_adapters.push_back({ std::string(value), 1.0 });
+            params.lora_adapters.push_back({ std::string(value), 1.0, nullptr });
         }
         // we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
     ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
     add_opt(common_arg(
         {"--lora-scaled"}, "FNAME", "SCALE",
         "path to LoRA adapter with user defined scaling (can be repeated to use multiple adapters)",
         [](common_params & params, const std::string & fname, const std::string & scale) {
-            params.lora_adapters.push_back({ fname, std::stof(scale) });
+            params.lora_adapters.push_back({ fname, std::stof(scale), nullptr });
         }
         // we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
     ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));

common/common.cpp

@@ -18,6 +18,7 @@
 #include <cstdarg>
 #include <cstring>
 #include <ctime>
+#include <filesystem>
 #include <fstream>
 #include <iostream>
 #include <iterator>
@@ -62,7 +63,9 @@
 #ifdef __linux__
 #include <linux/limits.h>
 #elif defined(_WIN32)
-#define PATH_MAX MAX_PATH
+#   if !defined(PATH_MAX)
+#   define PATH_MAX MAX_PATH
+#   endif
 #else
 #include <sys/syslimits.h>
 #endif
@@ -843,7 +846,7 @@ struct common_init_result common_init_from_params(common_params & params) {
     } else if (!params.model_url.empty()) {
         model = common_load_model_from_url(params.model_url, params.model, params.hf_token, mparams);
     } else {
-        model = llama_load_model_from_file(params.model.c_str(), mparams);
+        model = llama_model_load_from_file(params.model.c_str(), mparams);
     }
 
     if (model == NULL) {
@@ -870,7 +873,7 @@ struct common_init_result common_init_from_params(common_params & params) {
         }
 
         if (!ok) {
-            llama_free_model(model);
+            llama_model_free(model);
 
             return iparams;
         }
@@ -881,14 +884,13 @@ struct common_init_result common_init_from_params(common_params & params) {
     llama_context * lctx = llama_new_context_with_model(model, cparams);
     if (lctx == NULL) {
         LOG_ERR("%s: failed to create context with model '%s'\n", __func__, params.model.c_str());
-        llama_free_model(model);
+        llama_model_free(model);
         return iparams;
     }
 
     if (params.ctx_shift && !llama_kv_cache_can_shift(lctx)) {
-        LOG_ERR("%s: KV cache shifting is not supported for this model (--no-context-shift to disable)'\n", __func__);
-        llama_free_model(model);
-        return iparams;
+        LOG_WRN("%s: KV cache shifting is not supported for this model, disabling KV cache shifting\n", __func__);
+        params.ctx_shift = false;
     }
 
     if (!params.control_vectors.empty()) {
@@ -898,7 +900,7 @@ struct common_init_result common_init_from_params(common_params & params) {
         const auto cvec = common_control_vector_load(params.control_vectors);
         if (cvec.n_embd == -1) {
             llama_free(lctx);
-            llama_free_model(model);
+            llama_model_free(model);
 
             return iparams;
         }
@@ -911,28 +913,29 @@ struct common_init_result common_init_from_params(common_params & params) {
                                              params.control_vector_layer_end);
         if (err) {
             llama_free(lctx);
-            llama_free_model(model);
+            llama_model_free(model);
 
             return iparams;
         }
     }
 
     // load and optionally apply lora adapters
     for (auto & la : params.lora_adapters) {
-        common_lora_adapter_container loaded_la;
-        loaded_la.path = la.path;
-        loaded_la.scale = la.scale;
-        loaded_la.adapter = llama_lora_adapter_init(model, la.path.c_str());
-        if (loaded_la.adapter == nullptr) {
+        llama_lora_adapter_ptr lora;
+        lora.reset(llama_lora_adapter_init(model, la.path.c_str()));
+        if (lora == nullptr) {
             LOG_ERR("%s: failed to apply lora adapter '%s'\n", __func__, la.path.c_str());
             llama_free(lctx);
-            llama_free_model(model);
+            llama_model_free(model);
             return iparams;
         }
-        iparams.lora_adapters.push_back(loaded_la); // copy to list of loaded adapters
+
+        la.ptr = lora.get();
+        iparams.lora.emplace_back(std::move(lora)); // copy to list of loaded adapters
     }
+
     if (!params.lora_init_without_apply) {
-        common_lora_adapters_apply(lctx, iparams.lora_adapters);
+        common_lora_adapters_apply(lctx, params.lora_adapters);
     }
 
     if (params.sampling.ignore_eos && llama_token_eos(model) == LLAMA_TOKEN_NULL) {
@@ -979,7 +982,7 @@ struct common_init_result common_init_from_params(common_params & params) {
         if (llama_model_has_encoder(model)) {
             llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size()));
             llama_token decoder_start_token_id = llama_model_decoder_start_token(model);
-            if (decoder_start_token_id == -1) {
+            if (decoder_start_token_id == LLAMA_TOKEN_NULL) {
                 decoder_start_token_id = bos;
             }
             tmp.clear();
@@ -993,17 +996,17 @@ struct common_init_result common_init_from_params(common_params & params) {
         llama_perf_context_reset(lctx);
     }
 
-    iparams.model   = model;
-    iparams.context = lctx;
+    iparams.model.reset(model);
+    iparams.context.reset(lctx);
 
     return iparams;
 }
 
-void common_lora_adapters_apply(struct llama_context * ctx, std::vector<common_lora_adapter_container> & lora_adapters) {
+void common_lora_adapters_apply(struct llama_context * ctx, std::vector<common_lora_adapter_info> & lora) {
     llama_lora_adapter_clear(ctx);
-    for (auto & la : lora_adapters) {
+    for (auto & la : lora) {
         if (la.scale != 0.0f) {
-            llama_lora_adapter_set(ctx, la.adapter, la.scale);
+            llama_lora_adapter_set(ctx, la.ptr, la.scale);
         }
     }
 }
@@ -1148,8 +1151,7 @@ static bool common_download_file(const std::string & url, const std::string & pa
 #endif
 
     // Check if the file already exists locally
-    struct stat model_file_info;
-    auto file_exists = (stat(path.c_str(), &model_file_info) == 0);
+    auto file_exists = std::filesystem::exists(path);
 
     // If the file exists, check its JSON metadata companion file.
     std::string metadata_path = path + ".json";
@@ -1409,7 +1411,7 @@ struct llama_model * common_load_model_from_url(
         }
     }
 
-    return llama_load_model_from_file(local_path.c_str(), params);
+    return llama_model_load_from_file(local_path.c_str(), params);
 }
 
 struct llama_model * common_load_model_from_hf(
@@ -1612,6 +1614,18 @@ std::string common_detokenize(llama_context * ctx, const std::vector<llama_token
 // Chat template utils
 //
 
+std::string common_get_builtin_chat_template(const struct llama_model * model) {
+    static const char * template_key = "tokenizer.chat_template";
+    // call with NULL buffer to get the total size of the string
+    int32_t res = llama_model_meta_val_str(model, template_key, NULL, 0);
+    if (res > 0) {
+        std::vector<char> model_template(res + 1, 0);
+        llama_model_meta_val_str(model, template_key, model_template.data(), model_template.size());
+        return std::string(model_template.data(), model_template.size() - 1);
+    }
+    return "";
+}
+
 bool common_chat_verify_template(const std::string & tmpl) {
     llama_chat_message chat[] = {{"user", "test"}};
     int res = llama_chat_apply_template(nullptr, tmpl.c_str(), chat, 1, true, nullptr, 0);

common/common.h

@@ -2,7 +2,7 @@
 
 #pragma once
 
-#include "llama.h"
+#include "llama-cpp.h"
 
 #include <string>
 #include <vector>
@@ -27,10 +27,8 @@
 struct common_lora_adapter_info {
     std::string path;
     float scale;
-};
 
-struct common_lora_adapter_container : common_lora_adapter_info {
-    struct llama_lora_adapter * adapter;
+    struct llama_lora_adapter * ptr;
 };
 
 using llama_tokens = std::vector<llama_token>;
@@ -478,10 +476,12 @@ std::string fs_get_cache_file(const std::string & filename);
 // Model utils
 //
 
+// note: defines object's lifetime
 struct common_init_result {
-    struct llama_model   * model   = nullptr;
-    struct llama_context * context = nullptr;
-    std::vector<common_lora_adapter_container> lora_adapters;
+    llama_model_ptr   model;
+    llama_context_ptr context;
+
+    std::vector<llama_lora_adapter_ptr> lora;
 };
 
 struct common_init_result     common_init_from_params(common_params & params);
@@ -503,7 +503,7 @@ struct llama_model * common_load_model_from_hf(
     const struct llama_model_params & params);
 
 // clear LoRA adapters from context, then apply new list of adapters
-void common_lora_adapters_apply(struct llama_context * ctx, std::vector<common_lora_adapter_container> & lora_adapters);
+void common_lora_adapters_apply(struct llama_context * ctx, std::vector<common_lora_adapter_info> & lora);
 
 //
 // Batch utils
@@ -571,6 +571,9 @@ struct common_chat_msg {
     std::string content;
 };
 
+// Get the built-in chat template for the model. Return empty string if not present.
+std::string common_get_builtin_chat_template(const struct llama_model * model);
+
 // Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid
 bool common_chat_verify_template(const std::string & tmpl);
 
@@ -637,6 +640,10 @@ common_control_vector_data common_control_vector_load(const std::vector<common_c
 // Split utils
 //
 
-static const char * const LLM_KV_SPLIT_NO            = "split.no";
-static const char * const LLM_KV_SPLIT_COUNT         = "split.count";
-static const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
+namespace {
+
+const char * const LLM_KV_SPLIT_NO            = "split.no";
+const char * const LLM_KV_SPLIT_COUNT         = "split.count";
+const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
+
+}