move labels to llama_model

Author: CISC

examples/embedding/embedding.cpp

@@ -241,7 +241,7 @@ int main(int argc, char ** argv) {
             std::vector<std::string> cls_out_labels;
 
             for (uint32_t i = 0; i < n_cls_out; i++) {
-                const char * label = llama_model_get_classifier_label_by_index(model, i);
+                const char * label = llama_model_cls_label(model, i);
                 const std::string label_i = label == nullptr || strlen(label) == 0 ? std::to_string(i) : label;
                 cls_out_labels.emplace_back(label_i);
             }

include/llama.h

@@ -510,7 +510,7 @@ extern "C" {
     LLAMA_API uint32_t llama_model_n_cls_out(const struct llama_model * model);
 
     // Returns label of classifier output by index (<n_cls_out). Returns nullptr if no label provided
-    LLAMA_API const char * llama_model_get_classifier_label_by_index(const struct llama_model * model, uint32_t i);
+    LLAMA_API const char * llama_model_cls_label(const struct llama_model * model, uint32_t i);
 
     LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_vocab * vocab);
 

src/llama-model.cpp

@@ -417,19 +417,6 @@ void llama_model::load_arch(llama_model_loader & ml) {
     }
 }
 
-struct LLM_KV_MATCH_WITHOUT_ARCH {
-    const LLM_KV kv_arch = LLM_KV(LLM_ARCH_UNKNOWN);
-    const std::string kv_arch_prefix = llm_arch_name(LLM_ARCH_UNKNOWN);
-
-    bool operator()(const llm_kv & kv, const std::string & kv_name) const
-    {
-        std::string kv_match = kv_arch(kv);
-        auto kv_arch_pos = kv_match.find(kv_arch_prefix);
-
-        return kv_name.find(kv_match.substr(kv_arch_pos == std::string::npos ? 0 : kv_arch_pos + kv_arch_prefix.size())) != std::string::npos;
-    }
-};
-
 void llama_model::load_hparams(llama_model_loader & ml) {
     const gguf_context * ctx = ml.meta.get();
 
@@ -439,13 +426,12 @@ void llama_model::load_hparams(llama_model_loader & ml) {
         gguf_type type = gguf_get_kv_type(ctx, i);
 
         if (type == GGUF_TYPE_ARRAY) {
-            if (LLM_KV_MATCH_WITHOUT_ARCH()(LLM_KV_CLASSIFIER_OUTPUT_LABELS, name)) {
+            if (LLM_KV(arch)(LLM_KV_CLASSIFIER_OUTPUT_LABELS) == name) {
                 const size_t n_items = gguf_get_arr_n(ctx, i);
 
                 for (size_t j = 0; j < n_items; j++) {
-                    const std::string name_i = format("%s.%zu", name, j);
                     const std::string value = gguf_get_arr_str(ctx, i, j);
-                    gguf_kv.emplace(name_i, value);
+                    classifier_labels.emplace_back(value);
                 }
             }
         } else {
@@ -13620,12 +13606,9 @@ uint32_t llama_model_n_cls_out(const struct llama_model * model) {
     return model->hparams.n_cls_out;
 }
 
-const char * llama_model_get_classifier_label_by_index(const struct llama_model * model, uint32_t i) {
-    const std::string key = format("%s.%u", LLM_KV(model->arch)(LLM_KV_CLASSIFIER_OUTPUT_LABELS).c_str(), i);
-    const auto & it = model->gguf_kv.find(key);
-
-    if (it != model->gguf_kv.end()) {
-        return it->second.c_str();
+const char * llama_model_cls_label(const struct llama_model * model, uint32_t i) {
+    if (i < model->classifier_labels.size()) {
+        return model->classifier_labels[i].c_str();
     }
 
     return nullptr;

src/llama-model.h

@@ -363,6 +363,9 @@ struct llama_model {
     // for quantize-stats only
     std::vector<std::pair<std::string, struct ggml_tensor *>> tensors_by_name;
 
+    // for classifier models
+    std::vector<std::string> classifier_labels;
+
     int64_t t_load_us  = 0;
     int64_t t_start_us = 0;