Parse https://ollama.com/library/ syntax (#11480)

People search for ollama models using the web ui, this change
allows one to copy the url from the browser and for it to be
compatible with llama-run.

Signed-off-by: Eric Curtin <ecurtin@redhat.com>

examples/run/run.cpp

@@ -674,36 +674,27 @@ class LlamaData {
     }
 
     int github_dl(const std::string & model, const std::string & bn) {
-        std::string repository = model;
-        std::string branch     = "main";
-        size_t      at_pos     = model.find('@');
+        std::string  repository = model;
+        std::string  branch     = "main";
+        const size_t at_pos     = model.find('@');
         if (at_pos != std::string::npos) {
             repository = model.substr(0, at_pos);
             branch     = model.substr(at_pos + 1);
         }
 
-        std::vector<std::string> repo_parts;
-        size_t                   start = 0;
-        for (size_t end = 0; (end = repository.find('/', start)) != std::string::npos; start = end + 1) {
-            repo_parts.push_back(repository.substr(start, end - start));
-        }
-
-        repo_parts.push_back(repository.substr(start));
+        const std::vector<std::string> repo_parts = string_split(repository, "/");
         if (repo_parts.size() < 3) {
             printe("Invalid GitHub repository format\n");
             return 1;
         }
 
-        const std::string org          = repo_parts[0];
-        const std::string project      = repo_parts[1];
-        std::string       project_path = repo_parts[2];
-        for (size_t i = 3; i < repo_parts.size(); ++i) {
-            project_path += "/" + repo_parts[i];
+        const std::string & org          = repo_parts[0];
+        const std::string & project      = repo_parts[1];
+        std::string         url          = "https://raw.githubusercontent.com/" + org + "/" + project + "/" + branch;
+        for (size_t i = 2; i < repo_parts.size(); ++i) {
+            url += "/" + repo_parts[i];
         }
 
-        const std::string url =
-            "https://raw.githubusercontent.com/" + org + "/" + project + "/" + branch + "/" + project_path;
-
         return download(url, bn, true);
     }
 
@@ -735,19 +726,20 @@ class LlamaData {
         }
 
         const std::string bn = basename(model_);
-        if (string_starts_with(model_, "hf://") || string_starts_with(model_, "huggingface://")) {
+        if (string_starts_with(model_, "hf://") || string_starts_with(model_, "huggingface://") ||
+            string_starts_with(model_, "hf.co/")) {
+            rm_until_substring(model_, "hf.co/");
             rm_until_substring(model_, "://");
             ret = huggingface_dl(model_, bn);
-        } else if (string_starts_with(model_, "hf.co/")) {
-            rm_until_substring(model_, "hf.co/");
-            ret = huggingface_dl(model_, bn);
-        } else if (string_starts_with(model_, "https://") || string_starts_with(model_, "http://")) {
+        } else if ((string_starts_with(model_, "https://") || string_starts_with(model_, "http://")) &&
+                   !string_starts_with(model_, "https://ollama.com/library/")) {
             ret = download(model_, bn, true);
         } else if (string_starts_with(model_, "github:") || string_starts_with(model_, "github://")) {
-            rm_until_substring(model_, "github://");
             rm_until_substring(model_, "github:");
+            rm_until_substring(model_, "://");
             ret = github_dl(model_, bn);
         } else {  // ollama:// or nothing
+            rm_until_substring(model_, "ollama.com/library/");
             rm_until_substring(model_, "://");
             ret = ollama_dl(model_, bn);
         }

ggml/src/ggml-cpu/ggml-cpu.c

@@ -1302,7 +1302,7 @@ struct ggml_threadpool {
     // these are atomic as an annotation for thread-sanitizer
     atomic_bool stop;         // Used for stopping the threadpool altogether
     atomic_bool pause;        // Used for pausing the threadpool or individual threads
-    atomic_bool abort;        // Used for aborting processing of a graph
+    atomic_int abort;         // Used for aborting processing of a graph
 
     struct ggml_compute_state * workers;   // per thread state
     int          n_threads_max; // number of threads in the pool
@@ -13851,14 +13851,14 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
         /*.threadpool=*/ tp,
     };
 
-    for (int node_n = 0; node_n < cgraph->n_nodes && !tp->abort; node_n++) {
+    for (int node_n = 0; node_n < cgraph->n_nodes && atomic_load_explicit(&tp->abort, memory_order_relaxed) != node_n; node_n++) {
         struct ggml_tensor * node = cgraph->nodes[node_n];
 
         ggml_compute_forward(&params, node);
 
         if (state->ith == 0 && cplan->abort_callback &&
                 cplan->abort_callback(cplan->abort_callback_data)) {
-            tp->abort = true;
+            atomic_store_explicit(&tp->abort, node_n + 1, memory_order_relaxed);
             tp->ec    = GGML_STATUS_ABORTED;
         }
 
@@ -14031,7 +14031,7 @@ static struct ggml_threadpool * ggml_threadpool_new_impl(
         threadpool->current_chunk    = 0;
         threadpool->stop             = false;
         threadpool->pause            = tpp->paused;
-        threadpool->abort            = false;
+        threadpool->abort            = -1;
         threadpool->workers          = NULL;
         threadpool->n_threads_max    = tpp->n_threads;
         threadpool->n_threads_cur    = tpp->n_threads;
@@ -14110,7 +14110,7 @@ enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cpl
         threadpool->cgraph           = cgraph;
         threadpool->cplan            = cplan;
         threadpool->current_chunk    = 0;
-        threadpool->abort            = false;
+        threadpool->abort            = -1;
         threadpool->ec               = GGML_STATUS_SUCCESS;
     }
 

ggml/src/ggml.c

@@ -128,6 +128,10 @@ static void ggml_print_backtrace_symbols(void) {
 #endif
 
 static void ggml_print_backtrace(void) {
+    const char * GGML_NO_BACKTRACE = getenv("GGML_NO_BACKTRACE");
+    if (GGML_NO_BACKTRACE) {
+        return;
+    }
     char attach[32];
     snprintf(attach, sizeof(attach), "attach %d", getpid());
     int pid = fork();

scripts/sync-ggml.last

@@ -1 +1 @@
-d92321c0d151fe73a47d89738c7c3091ac904297
+32f0b85987396945afea2291d5f4c5862434292b