Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	README.md
#	examples/llama-bench/llama-bench.cpp
#	examples/llama.android/llama/src/main/cpp/llama-android.cpp
#	examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt
#	src/llama-vocab.cpp
#	tests/test-backend-ops.cpp

src/llama.cpp

@@ -49,7 +49,7 @@ static bool old_mixtral_warning_showed = false;
 #endif
 
 // Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback
-static int llama_model_load(const std::string & fname, llama_model & model, llama_model_params & params) {
+static int llama_model_load(const std::string & fname, std::vector<std::string> & splits, llama_model & model, llama_model_params & params) {
     // loading time will be recalculated after the first eval, so
     // we take page faults deferred by mmap() into consideration
     model.t_load_us = 0;
@@ -58,7 +58,7 @@ static int llama_model_load(const std::string & fname, llama_model & model, llam
     model.t_start_us = tm.t_start_us;
 
     try {
-        llama_model_loader ml(fname, params.use_mmap, params.check_tensors, params.kv_overrides);
+        llama_model_loader ml(fname, splits, params.use_mmap, params.check_tensors, params.kv_overrides);
 
         ml.print_info();
 
@@ -9414,14 +9414,9 @@ int64_t llama_time_us(void) {
     return ggml_time_us();
 }
 
-struct llama_model * llama_load_model_from_file(
-        const char * path_model,
-        struct llama_model_params params) {
-    return llama_model_load_from_file(path_model, params);
-}
-
-struct llama_model * llama_model_load_from_file(
-        const char * path_model,
+static struct llama_model * llama_model_load_from_file_impl(
+        const std::string & path_model,
+        std::vector<std::string> & splits,
         struct llama_model_params params) {
     ggml_time_init();
 
@@ -9444,47 +9439,6 @@ struct llama_model * llama_model_load_from_file(
         };
     }
 
-    if (params.rpc_servers != nullptr && params.rpc_servers[0] != '\0') {
-        // split the servers set them into model->rpc_servers
-        std::string servers(params.rpc_servers);
-        size_t pos = 0;
-        while ((pos = servers.find(',')) != std::string::npos) {
-            std::string server = servers.substr(0, pos);
-            model->rpc_servers.push_back(server);
-            servers.erase(0, pos + 1);
-        }
-        model->rpc_servers.push_back(servers);
-    }
-
-    // add RPC devices
-    if (!model->rpc_servers.empty()) {
-        ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
-        if (!rpc_reg) {
-            LLAMA_LOG_ERROR("%s: failed to find RPC backend\n", __func__);
-            llama_model_free(model);
-            return nullptr;
-        }
-
-        typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint);
-        ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device");
-        if (!ggml_backend_rpc_add_device_fn) {
-            LLAMA_LOG_ERROR("%s: failed to find RPC device add function\n", __func__);
-            llama_model_free(model);
-            return nullptr;
-        }
-
-        for (const std::string & server : model->rpc_servers) {
-            ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str());
-            if (dev) {
-                model->devices.push_back(dev);
-            } else {
-                LLAMA_LOG_ERROR("%s: failed to add RPC device for server '%s'\n", __func__, server.c_str());
-                llama_model_free(model);
-                return nullptr;
-            }
-        }
-    }
-
     // create list of devices to use with this model
     if (params.devices) {
         for (ggml_backend_dev_t * dev = params.devices; *dev; ++dev) {
@@ -9525,7 +9479,7 @@ struct llama_model * llama_model_load_from_file(
         LLAMA_LOG_INFO("%s: using device %s (%s) - %zu MiB free\n", __func__, ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), free/1024/1024);
     }
 
-    const int status = llama_model_load(path_model, *model, params);
+    const int status = llama_model_load(path_model, splits, *model, params);
     GGML_ASSERT(status <= 0);
     if (status < 0) {
         if (status == -1) {
@@ -9541,6 +9495,35 @@ struct llama_model * llama_model_load_from_file(
     return model;
 }
 
+// deprecated
+struct llama_model * llama_load_model_from_file(
+        const char * path_model,
+        struct llama_model_params params) {
+    return llama_model_load_from_file(path_model, params);
+}
+
+struct llama_model * llama_model_load_from_file(
+        const char * path_model,
+        struct llama_model_params params) {
+    std::vector<std::string> splits = {};
+    return llama_model_load_from_file_impl(path_model, splits, params);
+}
+
+struct llama_model * llama_model_load_from_splits(
+        const char ** paths,
+        size_t n_paths,
+        struct llama_model_params params) {
+    std::vector<std::string> splits;
+    if (n_paths == 0) {
+        LLAMA_LOG_ERROR("%s: list of splits is empty\n", __func__);
+        return nullptr;
+    }
+    for (size_t i = 0; i < n_paths; ++i) {
+        splits.push_back(paths[i]);
+    }
+    return llama_model_load_from_file_impl(splits.front(), splits, params);
+}
+
 struct llama_context * llama_init_from_model(
                  struct llama_model * model,
         struct llama_context_params   params) {