llmodel_c: improve quality of error messages (#1625)

gpt4all-backend/llamamodel.cpp

@@ -385,22 +385,35 @@ DLL_EXPORT const char *get_build_variant() {
 }
 
 DLL_EXPORT bool magic_match(const char * fname) {
-
     struct ggml_context * ctx_meta = NULL;
     struct gguf_init_params params = {
         /*.no_alloc = */ true,
         /*.ctx      = */ &ctx_meta,
     };
     gguf_context *ctx_gguf = gguf_init_from_file(fname, params);
-    if (!ctx_gguf)
+    if (!ctx_gguf) {
+        std::cerr << __func__ << ": gguf_init_from_file failed\n";
         return false;
+    }
+
+    bool valid = true;
+
+    int gguf_ver = gguf_get_version(ctx_gguf);
+    if (valid && gguf_ver > 3) {
+        std::cerr << __func__ << ": unsupported gguf version: " << gguf_ver << "\n";
+        valid = false;
+    }
 
-    bool isValid = gguf_get_version(ctx_gguf) <= 3;
     auto arch = get_arch_name(ctx_gguf);
-    isValid = isValid && (arch == "llama" || arch == "starcoder" || arch == "falcon" || arch == "mpt");
+    if (valid && !(arch == "llama" || arch == "starcoder" || arch == "falcon" || arch == "mpt")) {
+        if (!(arch == "gptj" || arch == "bert")) { // we support these via other modules
+            std::cerr << __func__ << ": unsupported model architecture: " << arch << "\n";
+        }
+        valid = false;
+    }
 
     gguf_free(ctx_gguf);
-    return isValid;
+    return valid;
 }
 
 DLL_EXPORT LLModel *construct() {

gpt4all-backend/llmodel.cpp

@@ -123,11 +123,18 @@ const std::vector<LLModel::Implementation> &LLModel::Implementation::implementat
 }
 
 const LLModel::Implementation* LLModel::Implementation::implementation(const char *fname, const std::string& buildVariant) {
+    bool buildVariantMatched = false;
     for (const auto& i : implementationList()) {
         if (buildVariant != i.m_buildVariant) continue;
+        buildVariantMatched = true;
+
         if (!i.m_magicMatch(fname)) continue;
         return &i;
     }
+
+    if (!buildVariantMatched) {
+        std::cerr << "LLModel ERROR: Could not find any implementations for build variant: " << buildVariant << "\n";
+    }
     return nullptr;
 }
 

gpt4all-backend/llmodel_c.cpp

@@ -11,45 +11,33 @@ struct LLModelWrapper {
     ~LLModelWrapper() { delete llModel; }
 };
 
-
 thread_local static std::string last_error_message;
 
-
 llmodel_model llmodel_model_create(const char *model_path) {
-    auto fres = llmodel_model_create2(model_path, "auto", nullptr);
+    const char *error;
+    auto fres = llmodel_model_create2(model_path, "auto", &error);
     if (!fres) {
-        fprintf(stderr, "Invalid model file\n");
+        fprintf(stderr, "Unable to instantiate model: %s\n", error);
     }
     return fres;
 }
 
-llmodel_model llmodel_model_create2(const char *model_path, const char *build_variant, llmodel_error *error) {
+llmodel_model llmodel_model_create2(const char *model_path, const char *build_variant, const char **error) {
     auto wrapper = new LLModelWrapper;
-    int error_code = 0;
 
     try {
         wrapper->llModel = LLModel::Implementation::construct(model_path, build_variant);
+        if (!wrapper->llModel) {
+            last_error_message = "Model format not supported (no matching implementation found)";
+        }
     } catch (const std::exception& e) {
-        error_code = EINVAL;
         last_error_message = e.what();
     }
 
     if (!wrapper->llModel) {
         delete std::exchange(wrapper, nullptr);
-        // Get errno and error message if none
-        if (error_code == 0) {
-            if (errno != 0) {
-                error_code = errno;
-                last_error_message = std::strerror(error_code);
-            } else {
-                error_code = ENOTSUP;
-                last_error_message = "Model format not supported (no matching implementation found)";
-            }
-        }
-        // Set error argument
         if (error) {
-            error->message = last_error_message.c_str();
-            error->code = error_code;
+            *error = last_error_message.c_str();
         }
     }
     return reinterpret_cast<llmodel_model*>(wrapper);

gpt4all-backend/llmodel_c.h

@@ -23,17 +23,6 @@ extern "C" {
  */
 typedef void *llmodel_model;
 
-/**
- * Structure containing any errors that may eventually occur
- */
-struct llmodel_error {
-    const char *message;  // Human readable error description; Thread-local; guaranteed to survive until next llmodel C API call
-    int code;             // errno; 0 if none
-};
-#ifndef __cplusplus
-typedef struct llmodel_error llmodel_error;
-#endif
-
 /**
  * llmodel_prompt_context structure for holding the prompt context.
  * NOTE: The implementation takes care of all the memory handling of the raw logits pointer and the
@@ -105,10 +94,10 @@ DEPRECATED llmodel_model llmodel_model_create(const char *model_path);
  * Recognises correct model type from file at model_path
  * @param model_path A string representing the path to the model file; will only be used to detect model type.
  * @param build_variant A string representing the implementation to use (auto, default, avxonly, ...),
- * @param error A pointer to a llmodel_error; will only be set on error.
+ * @param error A pointer to a string; will only be set on error.
  * @return A pointer to the llmodel_model instance; NULL on error.
  */
-llmodel_model llmodel_model_create2(const char *model_path, const char *build_variant, llmodel_error *error);
+llmodel_model llmodel_model_create2(const char *model_path, const char *build_variant, const char **error);
 
 /**
  * Destroy a llmodel instance.