@fugood/llama.node 0.3.17 → 0.4.0

package/src/llama.cpp/tools/mtmd/mtmd.h

@@ -0,0 +1,331 @@
+#ifndef MTMD_H
+#define MTMD_H
+
+#include "ggml.h"
+#include "llama.h"
+#include "clip.h"
+
+#include <stddef.h>
+#include <stdint.h>
+#include <stdbool.h>
+
+#ifdef __cplusplus
+#include <string>
+#include <vector>
+#include <cinttypes>
+#include <memory>
+#endif
+
+/**
+ * libmtmd: A library for multimodal support in llama.cpp.
+ *
+ * WARNING: This API is experimental and subject to many BREAKING CHANGES.
+ *          Issues related to API usage may receive lower priority support.
+ *
+ * For the usage, see an example in mtmd-cli.cpp
+ */
+
+#ifdef LLAMA_SHARED
+#    if defined(_WIN32) && !defined(__MINGW32__)
+#        ifdef LLAMA_BUILD
+#            define MTMD_API __declspec(dllexport)
+#        else
+#            define MTMD_API __declspec(dllimport)
+#        endif
+#    else
+#        define MTMD_API __attribute__ ((visibility ("default")))
+#    endif
+#else
+#    define MTMD_API
+#endif
+
+#define MTMD_DEFAULT_IMAGE_MARKER "<__image__>"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+enum mtmd_input_chunk_type {
+    MTMD_INPUT_CHUNK_TYPE_TEXT,
+    MTMD_INPUT_CHUNK_TYPE_IMAGE,
+};
+
+// opaque types
+struct mtmd_context;
+struct mtmd_bitmap;
+struct mtmd_image_tokens;
+struct mtmd_input_chunk;
+struct mtmd_input_chunks;
+
+struct mtmd_input_text {
+    const char * text;
+    bool add_special;
+    bool parse_special;
+};
+
+//
+// C API
+//
+
+typedef struct mtmd_context      mtmd_context;
+typedef struct mtmd_bitmap       mtmd_bitmap;
+typedef struct mtmd_image_tokens mtmd_image_tokens;
+typedef struct mtmd_input_chunk  mtmd_input_chunk;
+typedef struct mtmd_input_chunks mtmd_input_chunks;
+typedef struct mtmd_input_text   mtmd_input_text;
+
+struct mtmd_context_params {
+    bool use_gpu;
+    bool print_timings;
+    int n_threads;
+    enum ggml_log_level verbosity;
+    const char * image_marker;
+};
+
+MTMD_API struct mtmd_context_params mtmd_context_params_default(void);
+
+// initialize the mtmd context
+// return nullptr on failure
+MTMD_API mtmd_context * mtmd_init_from_file(const char * mmproj_fname,
+                                            const struct llama_model * text_model,
+                                            const struct mtmd_context_params ctx_params);
+
+MTMD_API void mtmd_free(mtmd_context * ctx);
+
+// whether we need to set non-causal mask before llama_decode
+MTMD_API bool mtmd_decode_use_non_causal(mtmd_context * ctx);
+
+// whether the current model use M-RoPE for llama_decode
+MTMD_API bool mtmd_decode_use_mrope(mtmd_context * ctx);
+
+
+// mtmd_bitmap
+//
+// length of data must be nx * ny * 3
+// the data is in RGBRGBRGB... format
+MTMD_API mtmd_bitmap *         mtmd_bitmap_init    (uint32_t nx,
+                                                    uint32_t ny,
+                                                    const unsigned char * data);
+MTMD_API uint32_t              mtmd_bitmap_get_nx  (const mtmd_bitmap * bitmap);
+MTMD_API uint32_t              mtmd_bitmap_get_ny  (const mtmd_bitmap * bitmap);
+MTMD_API const unsigned char * mtmd_bitmap_get_data(const mtmd_bitmap * bitmap);
+MTMD_API void                  mtmd_bitmap_free    (mtmd_bitmap * bitmap);
+// bitmap ID is optional, but useful for KV cache tracking
+// these getters/setters are dedicated functions, so you can for example calculate the hash of the image based on mtmd_bitmap_get_data()
+MTMD_API const char * mtmd_bitmap_get_id(const mtmd_bitmap * bitmap);
+MTMD_API void         mtmd_bitmap_set_id(mtmd_bitmap * bitmap, const char * id);
+
+
+// mtmd_input_chunks
+//
+// this is simply a list of mtmd_input_chunk
+// the elements can only be populated via mtmd_tokenize()
+MTMD_API mtmd_input_chunks *      mtmd_input_chunks_init(void);
+MTMD_API size_t                   mtmd_input_chunks_size(const mtmd_input_chunks * chunks);
+MTMD_API const mtmd_input_chunk * mtmd_input_chunks_get (const mtmd_input_chunks * chunks, size_t idx);
+MTMD_API void                     mtmd_input_chunks_free(mtmd_input_chunks * chunks);
+
+// mtmd_input_chunk
+//
+// the instance will be constructed via mtmd_tokenize()
+// it will be freed along with mtmd_input_chunks
+MTMD_API enum mtmd_input_chunk_type mtmd_input_chunk_get_type        (const mtmd_input_chunk * chunk);
+MTMD_API const llama_token *        mtmd_input_chunk_get_tokens_text (const mtmd_input_chunk * chunk, size_t * n_tokens_output);
+MTMD_API const mtmd_image_tokens *  mtmd_input_chunk_get_tokens_image(const mtmd_input_chunk * chunk);
+
+// in case you want to use custom logic to handle the chunk (i.e. KV cache management)
+// you can move the chunk ownership to your own code by copying it
+// remember to free the chunk when you are done with it
+MTMD_API mtmd_input_chunk * mtmd_input_chunk_copy(const mtmd_input_chunk * chunk);
+MTMD_API void               mtmd_input_chunk_free(mtmd_input_chunk * chunk);
+
+
+// mtmd_image_tokens
+//
+// the instance will be constructed via mtmd_tokenize()
+// it will be freed along with mtmd_input_chunk
+MTMD_API size_t       mtmd_image_tokens_get_n_tokens(const mtmd_image_tokens * image_tokens);
+MTMD_API size_t       mtmd_image_tokens_get_nx      (const mtmd_image_tokens * image_tokens);
+MTMD_API size_t       mtmd_image_tokens_get_ny      (const mtmd_image_tokens * image_tokens);
+MTMD_API const char * mtmd_image_tokens_get_id      (const mtmd_image_tokens * image_tokens);
+// number of temporal positions (always 1 for M-RoPE, n_tokens otherwise)
+MTMD_API llama_pos    mtmd_image_tokens_get_n_pos   (const mtmd_image_tokens * image_tokens);
+
+// tokenize an input text prompt and an image
+// the prompt must have the input image marker (default: "<__image__>") in it
+// the marker will be replaced with the image tokens
+// for example:
+//   "here is an image: <__image__>\ndescribe it in detail."
+//   this will gives 3 chunks:
+//   1. "here is an image: <start_of_image>"
+//   2. (image tokens)
+//   3. "<end_of_image>\ndescribe it in detail."
+// number of bitmaps must be equal to the number of image markers in the prompt
+// this function is thread-safe (shared ctx)
+// return values:
+//   0 on success
+//   1 on number of images not matching the number of markers
+//   2 on image preprocessing error
+MTMD_API int32_t mtmd_tokenize(mtmd_context * ctx,
+                               mtmd_input_chunks * output,
+                               const mtmd_input_text * text,
+                               const mtmd_bitmap ** bitmaps,
+                               size_t n_bitmaps);
+
+// returns 0 on success
+MTMD_API int32_t mtmd_encode(mtmd_context * ctx,
+                             const mtmd_image_tokens * image_tokens);
+
+// get output embeddings from the last encode pass
+MTMD_API float * mtmd_get_output_embd(mtmd_context * ctx);
+
+/////////////////////////////////////////
+
+//
+// Helper functions (can be implemented based on other functions)
+//
+// Please note that these helpers are not guaranteed to be stable.
+// BREAKING CHANGES are expected.
+//
+
+// helper function to construct a mtmd_bitmap from a file
+// returns nullptr on failure
+// this function is thread-safe
+MTMD_API mtmd_bitmap * mtmd_helper_bitmap_init_from_file(const char * fname);
+
+// helper function to construct a mtmd_bitmap from a buffer containing a file
+// the file content must be an image in format supported by stb_image (jpg, png, bmp, gif, etc.)
+// returns nullptr on failure
+// this function is thread-safe
+MTMD_API mtmd_bitmap * mtmd_helper_bitmap_init_from_buf(const unsigned char * buf, size_t len);
+
+// helper to count the total number of tokens from a list of chunks, useful to keep track of KV cache
+MTMD_API size_t mtmd_helper_get_n_tokens(const mtmd_input_chunks * chunks);
+
+// helper to count the total position of tokens from a list of chunks, useful to keep track of n_past
+// normally, n_pos is equal to n_tokens, but for M-RoPE it is different
+MTMD_API llama_pos mtmd_helper_get_n_pos(const mtmd_input_chunks * chunks);
+
+// helper function that automatically:
+// 1. run llama_decode() on text chunks
+// 2. run mtmd_encode() on image chunks, then mtmd_get_output_embd() and then llama_decode()
+// if any of the mtmd_encode() or llama_decode() calls return non-zero, stop and forward the error
+// otherwise, returns 0 on success
+// this function is NOT thread-safe
+MTMD_API int32_t mtmd_helper_eval_chunks(mtmd_context * ctx,
+                                         struct llama_context * lctx,
+                                         const mtmd_input_chunks * chunks,
+                                         llama_pos n_past,
+                                         llama_seq_id seq_id,
+                                         int32_t n_batch,
+                                         bool logits_last,
+                                         llama_pos * new_n_past);
+
+// works like mtmd_helper_eval_chunks(), but only for a single chunk
+// this function is NOT thread-safe
+MTMD_API int32_t mtmd_helper_eval_chunk_single(mtmd_context * ctx,
+                                               struct llama_context * lctx,
+                                               const mtmd_input_chunk * chunk,
+                                               llama_pos n_past,
+                                               llama_seq_id seq_id,
+                                               int32_t n_batch,
+                                               bool logits_last,
+                                               llama_pos * new_n_past);
+
+// helper function to decode an image whose embeddings have already been calculated
+// this helper will handle batching and pre/post decoding setup (for ex. gemma 3 requires non-causal attention)
+// ret 0 on success, -1 on chunk not being a valid image chunk, 1 on decode failure
+MTMD_API int32_t mtmd_helper_decode_image_chunk(mtmd_context * ctx,
+                                                struct llama_context * lctx,
+                                                const mtmd_input_chunk * chunk,
+                                                float * encoded_embd,
+                                                llama_pos n_past,
+                                                llama_seq_id seq_id,
+                                                int32_t n_batch,
+                                                llama_pos * new_n_past);
+
+/////////////////////////////////////////
+
+// test function, to be used in test-mtmd-c-api.c
+MTMD_API mtmd_input_chunks * mtmd_test_create_input_chunks(void);
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
+//
+// C++ wrappers
+//
+
+#ifdef __cplusplus
+
+namespace mtmd {
+
+struct mtmd_context_deleter {
+    void operator()(mtmd_context * val) { mtmd_free(val); }
+};
+using context_ptr = std::unique_ptr<mtmd_context, mtmd_context_deleter>;
+
+struct mtmd_bitmap_deleter {
+    void operator()(mtmd_bitmap * val) { mtmd_bitmap_free(val); }
+};
+using bitmap_ptr = std::unique_ptr<mtmd_bitmap, mtmd_bitmap_deleter>;
+
+struct mtmd_input_chunks_deleter {
+    void operator()(mtmd_input_chunks * val) { mtmd_input_chunks_free(val); }
+};
+using input_chunks_ptr = std::unique_ptr<mtmd_input_chunks, mtmd_input_chunks_deleter>;
+
+struct mtmd_input_chunk_deleter {
+    void operator()(mtmd_input_chunk * val) { mtmd_input_chunk_free(val); }
+};
+using input_chunk_ptr = std::unique_ptr<mtmd_input_chunk, mtmd_input_chunk_deleter>;
+
+struct bitmap {
+    bitmap_ptr ptr;
+    bitmap() : ptr(nullptr) {}
+    bitmap(mtmd_bitmap * bitmap) : ptr(bitmap) {}
+    bitmap(bitmap && other) noexcept : ptr(std::move(other.ptr)) {}
+    bitmap(uint32_t nx, uint32_t ny, const unsigned char * data) {
+        ptr.reset(mtmd_bitmap_init(nx, ny, data));
+    }
+    ~bitmap() = default;
+    uint32_t nx() { return mtmd_bitmap_get_nx(ptr.get()); }
+    uint32_t ny() { return mtmd_bitmap_get_ny(ptr.get()); }
+    const unsigned char * data() { return mtmd_bitmap_get_data(ptr.get()); }
+    std::string id() { return mtmd_bitmap_get_id(ptr.get()); }
+    void set_id(const char * id) { mtmd_bitmap_set_id(ptr.get(), id); }
+};
+
+struct bitmaps {
+    std::vector<bitmap> entries;
+    ~bitmaps() = default;
+    // return list of pointers to mtmd_bitmap
+    // example:
+    //   auto bitmaps_c_ptr = bitmaps.c_ptr();
+    //   int32_t res = mtmd_tokenize(... bitmaps_c_ptr.data(), bitmaps_c_ptr.size());
+    std::vector<const mtmd_bitmap *> c_ptr() {
+        std::vector<const mtmd_bitmap *> res(entries.size());
+        for (size_t i = 0; i < entries.size(); i++) {
+            res[i] = entries[i].ptr.get();
+        }
+        return res;
+    }
+};
+
+struct input_chunks {
+    input_chunks_ptr ptr;
+    input_chunks() = default;
+    input_chunks(mtmd_input_chunks * chunks) : ptr(chunks) {}
+    ~input_chunks() = default;
+    size_t size() { return mtmd_input_chunks_size(ptr.get()); }
+    const mtmd_input_chunk * operator[](size_t idx) {
+        return mtmd_input_chunks_get(ptr.get(), idx);
+    }
+};
+
+} // namespace mtmd
+
+#endif
+
+#endif

package/src/llama.cpp/{examples → tools}/perplexity/perplexity.cpp

@@ -1554,7 +1554,10 @@ static void multiple_choice_score(llama_context * ctx, const common_params & par
             if (int(batch_indeces.size()) != num_answers) {
                 batch_indeces.resize(num_answers);
             }
-            for (int s = 0; s < num_answers; ++s) batch_indeces[s] = s0 + s;
+
+            for (int s = 0; s < num_answers; ++s) {
+                batch_indeces[s] = s0 + s;
+            }
 
             for (size_t i = 0; i < cur_task.common_prefix; ++i) {
                 //llama_batch_add(batch, cur_task.seq_tokens[0][i], i, { s0 + 0, s0 + 1, s0 + 2, s0 + 3}, false);
@@ -1970,7 +1973,6 @@ int main(int argc, char ** argv) {
     common_params params;
 
     params.n_ctx = 512;
-    params.logits_all = true;
     params.escape = false;
 
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_PERPLEXITY)) {

package/src/llama.cpp/{examples → tools}/quantize/quantize.cpp

@@ -57,6 +57,12 @@ static const std::vector<quant_option> QUANT_OPTIONS = {
     { "COPY",     LLAMA_FTYPE_ALL_F32,         "only copy tensors, no quantizing",  },
 };
 
+// Quantization types. Changes to this struct must be replicated in llama-quantize.cpp
+struct tensor_quantization {
+    std::string name;
+    ggml_type quant = GGML_TYPE_COUNT;
+};
+
 static const char * const LLM_KV_QUANTIZE_IMATRIX_FILE       = "quantize.imatrix.file";
 static const char * const LLM_KV_QUANTIZE_IMATRIX_DATASET    = "quantize.imatrix.dataset";
 static const char * const LLM_KV_QUANTIZE_IMATRIX_N_ENTRIES  = "quantize.imatrix.entries_count";
@@ -244,56 +250,10 @@ static ggml_type parse_ggml_type(const char * arg) {
             return type;
         }
     }
-    fprintf(stderr, "%s: invalid ggml_type '%s'\n", __func__, arg);
+    fprintf(stderr, "\n%s: invalid ggml_type '%s'\n\n", __func__, arg);
     return GGML_TYPE_COUNT;
 }
 
-// Allowed tensors for arbitrary quantization with --tensor-type option
-static const std::vector<std::string> ALLOWED_TENSOR_TYPE = {
-    "attn_k",
-    "attn_kv_a_mqa",
-    "attn_kv_b",
-    "attn_o",
-    "attn_output",
-    "attn_q",
-    "attn_q_a",
-    "attn_q_b",
-    "attn_qkv",
-    "attn_v",
-    "channel_mix_key",
-    "channel_mix_receptance",
-    "channel_mix_value",
-    "cls",
-    "cls.output",
-    "cross_attn_k",
-    "cross_attn_o",
-    "cross_attn_q",
-    "cross_attn_v",
-    "ffn_act",
-    "ffn_down",
-    "ffn_down_exps",
-    "ffn_down_shexp",
-    "ffn_gate",
-    "ffn_gate_exps",
-    "ffn_gate_shexp",
-    "ffn_up",
-    "ffn_up_exps",
-    "ffn_up_shexp",
-    "ssm_in",
-    "ssm_out",
-    "time_mix_gate",
-    "time_mix_key",
-    "time_mix_output",
-    "time_mix_receptance",
-    "time_mix_value",
-};
-
-// changes to this struct must be replicated in llama-quant.cpp
-struct tensor_quantization {
-    std::string name;
-    ggml_type quant = GGML_TYPE_COUNT;
-};
-
 static bool parse_tensor_type(const char * data, std::vector<tensor_quantization> & tensor_type) {
     const char * sep = strchr(data, '=');
     if (sep == nullptr) {
@@ -306,7 +266,6 @@ static bool parse_tensor_type(const char * data, std::vector<tensor_quantization
         printf("\n%s: missing tensor name\n\n", __func__);
         return false;
     }
-
     if (const size_t qt_len = strlen(sep); qt_len == 1) {
         printf("\n%s: missing quantization type\n\n", __func__);
         return false;
@@ -315,37 +274,15 @@ static bool parse_tensor_type(const char * data, std::vector<tensor_quantization
     std::string tn(data, tn_len);
     std::transform(tn.begin(), tn.end(), tn.begin(), tolower);
     sep++;
-    const std::string qt(sep);
-
-    bool found = false;
-    for (const auto & allowed : ALLOWED_TENSOR_TYPE) {
-        std::string tensor;
-        tensor = tn.rfind('.') != std::string::npos ? tn.substr(tn.rfind('.') + 1) : tn;
-        // handle special case of cls.output
-        std::string cls_output = "cls.output";
-        if (tn.find(cls_output) != std::string::npos) {
-            tensor = "cls.output";
-        }
-        // check if an allowed tensor exists and it's at the end of the kv string
-        if (tensor == allowed) {
-            found = true;
-            break;
-        }
-    }
-    if (!found) {
-        printf("\n%s: invalid tensor name '%s'\n\n", __func__, tn.c_str());
-        return false;
-    }
-
-    if (parse_ggml_type(qt.c_str()) == GGML_TYPE_COUNT) {
-        printf("\n%s: invalid quantization type '%s'\n\n", __func__, qt.c_str());
-        return false;
-    }
-
     tensor_quantization tqz;
     tqz.name = tn;
-    tqz.quant = parse_ggml_type(qt.c_str());
+    tqz.quant = parse_ggml_type(sep);
     tensor_type.emplace_back(std::move(tqz));
+    if (tqz.quant == GGML_TYPE_COUNT) {
+        printf("\n%s: invalid quantization type '%s'\n\n", __func__, sep);
+        return false;
+    }
+
     return true;
 }
 

package/src/llama.cpp/{examples → tools}/rpc/rpc-server.cpp

@@ -2,24 +2,6 @@
 #define _SILENCE_CXX17_CODECVT_HEADER_DEPRECATION_WARNING
 #endif
 
-#include "ggml-cpu.h"
-
-#ifdef GGML_USE_CUDA
-#include "ggml-cuda.h"
-#endif
-
-#ifdef GGML_USE_METAL
-#include "ggml-metal.h"
-#endif
-
-#ifdef GGML_USE_VULKAN
-#include "ggml-vulkan.h"
-#endif
-
-#ifdef GGML_USE_SYCL
-#include "ggml-sycl.h"
-#endif
-
 #include "ggml-rpc.h"
 #ifdef _WIN32
 #  define NOMINMAX
@@ -154,6 +136,7 @@ struct rpc_server_params {
     size_t      backend_mem = 0;
     bool        use_cache   = false;
     int         n_threads   = std::max(1U, std::thread::hardware_concurrency()/2);
+    std::string device;
 };
 
 static void print_usage(int /*argc*/, char ** argv, rpc_server_params params) {
@@ -161,6 +144,7 @@ static void print_usage(int /*argc*/, char ** argv, rpc_server_params params) {
     fprintf(stderr, "options:\n");
     fprintf(stderr, "  -h, --help                show this help message and exit\n");
     fprintf(stderr, "  -t,      --threads        number of threads for the CPU backend (default: %d)\n", params.n_threads);
+    fprintf(stderr, "  -d DEV,  --device         device to use\n");
     fprintf(stderr, "  -H HOST, --host HOST      host to bind to (default: %s)\n", params.host.c_str());
     fprintf(stderr, "  -p PORT, --port PORT      port to bind to (default: %d)\n", params.port);
     fprintf(stderr, "  -m MEM,  --mem MEM        backend memory size (in MB)\n");
@@ -186,6 +170,22 @@ static bool rpc_server_params_parse(int argc, char ** argv, rpc_server_params &
                 fprintf(stderr, "error: invalid number of threads: %d\n", params.n_threads);
                 return false;
             }
+        } else if (arg == "-d" || arg == "--device") {
+            if (++i >= argc) {
+                return false;
+            }
+            params.device = argv[i];
+            if (ggml_backend_dev_by_name(params.device.c_str()) == nullptr) {
+                fprintf(stderr, "error: unknown device: %s\n", params.device.c_str());
+                fprintf(stderr, "available devices:\n");
+                for (size_t i = 0; i < ggml_backend_dev_count(); i++) {
+                    auto * dev = ggml_backend_dev_get(i);
+                    size_t free, total;
+                    ggml_backend_dev_memory(dev, &free, &total);
+                    printf("  %s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024);
+                }
+                return false;
+            }
         } else if (arg == "-p" || arg == "--port") {
             if (++i >= argc) {
                 return false;
@@ -214,66 +214,55 @@ static bool rpc_server_params_parse(int argc, char ** argv, rpc_server_params &
 }
 
 static ggml_backend_t create_backend(const rpc_server_params & params) {
-    ggml_backend_t backend = NULL;
-#ifdef GGML_USE_CUDA
-    fprintf(stderr, "%s: using CUDA backend\n", __func__);
-    backend = ggml_backend_cuda_init(0); // init device 0
-    if (!backend) {
-        fprintf(stderr, "%s: ggml_backend_cuda_init() failed\n", __func__);
-    }
-#elif GGML_USE_METAL
-    fprintf(stderr, "%s: using Metal backend\n", __func__);
-    backend = ggml_backend_metal_init();
-    if (!backend) {
-        fprintf(stderr, "%s: ggml_backend_metal_init() failed\n", __func__);
+    ggml_backend_t backend = nullptr;
+
+    if (!params.device.empty()) {
+        ggml_backend_dev_t dev = ggml_backend_dev_by_name(params.device.c_str());
+        if (dev) {
+            backend = ggml_backend_dev_init(dev, nullptr);
+            if (!backend) {
+                fprintf(stderr, "Failed to create backend for device %s\n", params.device.c_str());
+                return nullptr;
+            }
+        }
     }
-#elif GGML_USE_VULKAN
-    fprintf(stderr, "%s: using Vulkan backend\n", __func__);
-    backend = ggml_backend_vk_init(0); // init device 0
+
+    // try to initialize a GPU backend first
     if (!backend) {
-        fprintf(stderr, "%s: ggml_backend_vulkan_init() failed\n", __func__);
+        backend = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_GPU, nullptr);
     }
-#elif GGML_USE_SYCL
-    fprintf(stderr, "%s: using SYCL backend\n", __func__);
-    backend = ggml_backend_sycl_init(0); // init device 0
+
+    // if there aren't GPU backends fallback to CPU backend
     if (!backend) {
-        fprintf(stderr, "%s: ggml_backend_sycl_init() failed\n", __func__);
+        backend = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_CPU, nullptr);
     }
-#endif
 
-    // if there aren't GPU Backends fallback to CPU backend
-    if (!backend) {
-        fprintf(stderr, "%s: using CPU backend\n", __func__);
-        backend = ggml_backend_cpu_init();
-        ggml_backend_cpu_set_n_threads(backend, params.n_threads);
+    if (backend) {
+        fprintf(stderr, "%s: using %s backend\n", __func__, ggml_backend_name(backend));
+
+        // set the number of threads
+        ggml_backend_dev_t dev = ggml_backend_get_device(backend);
+        ggml_backend_reg_t reg = dev ? ggml_backend_dev_backend_reg(dev) : nullptr;
+        if (reg) {
+            auto ggml_backend_set_n_threads_fn = (ggml_backend_set_n_threads_t) ggml_backend_reg_get_proc_address(reg, "ggml_backend_set_n_threads");
+            if (ggml_backend_set_n_threads_fn) {
+                ggml_backend_set_n_threads_fn(backend, params.n_threads);
+            }
+        }
     }
+
     return backend;
 }
 
-static void get_backend_memory(size_t * free_mem, size_t * total_mem) {
-#ifdef GGML_USE_CUDA
-    ggml_backend_cuda_get_device_memory(0, free_mem, total_mem);
-#elif GGML_USE_VULKAN
-    ggml_backend_vk_get_device_memory(0, free_mem, total_mem);
-#elif GGML_USE_SYCL
-    ggml_backend_sycl_get_device_memory(0, free_mem, total_mem);
-#else
-    #ifdef _WIN32
-        MEMORYSTATUSEX status;
-        status.dwLength = sizeof(status);
-        GlobalMemoryStatusEx(&status);
-        *total_mem = status.ullTotalPhys;
-        *free_mem = status.ullAvailPhys;
-    #else
-        long pages = sysconf(_SC_PHYS_PAGES);
-        long page_size = sysconf(_SC_PAGE_SIZE);
-        *total_mem = pages * page_size;
-        *free_mem = *total_mem;
-    #endif
-#endif
+static void get_backend_memory(ggml_backend_t backend, size_t * free_mem, size_t * total_mem) {
+    ggml_backend_dev_t dev = ggml_backend_get_device(backend);
+    GGML_ASSERT(dev != nullptr);
+    ggml_backend_dev_memory(dev, free_mem, total_mem);
 }
 
 int main(int argc, char * argv[]) {
+    ggml_backend_load_all();
+
     rpc_server_params params;
     if (!rpc_server_params_parse(argc, argv, params)) {
         fprintf(stderr, "Invalid parameters\n");
@@ -301,7 +290,7 @@ int main(int argc, char * argv[]) {
         free_mem = params.backend_mem;
         total_mem = params.backend_mem;
     } else {
-        get_backend_memory(&free_mem, &total_mem);
+        get_backend_memory(backend, &free_mem, &total_mem);
     }
     const char * cache_dir = nullptr;
     std::string cache_dir_str;
@@ -313,14 +302,21 @@ int main(int argc, char * argv[]) {
         }
         cache_dir = cache_dir_str.c_str();
     }
-    printf("Starting RPC server v%d.%d.%d\n",
-           RPC_PROTO_MAJOR_VERSION,
-           RPC_PROTO_MINOR_VERSION,
-           RPC_PROTO_PATCH_VERSION);
-    printf("  endpoint       : %s\n", endpoint.c_str());
-    printf("  local cache    : %s\n", cache_dir ? cache_dir : "n/a");
-    printf("  backend memory : %zu MB\n", free_mem / (1024 * 1024));
-    ggml_backend_rpc_start_server(backend, endpoint.c_str(), cache_dir, free_mem, total_mem);
+
+    ggml_backend_reg_t reg = ggml_backend_reg_by_name("RPC");
+    if (!reg) {
+        fprintf(stderr, "Failed to find RPC backend\n");
+        return 1;
+    }
+
+    auto start_server_fn = (decltype(ggml_backend_rpc_start_server)*) ggml_backend_reg_get_proc_address(reg, "ggml_backend_rpc_start_server");
+    if (!start_server_fn) {
+        fprintf(stderr, "Failed to obtain RPC backend start server function\n");
+        return 1;
+    }
+
+    start_server_fn(backend, endpoint.c_str(), cache_dir, free_mem, total_mem);
+
     ggml_backend_free(backend);
     return 0;
 }