Enhance ValidateBreweryJson to include reasoning output and update GenerateBrewery to use user_prompt

Add gemma parser

pipeline/includes/data_generation/llama_generator.h

@@ -15,6 +15,7 @@
 #include <string_view>
 
 #include "data_generation/data_generator.h"
+#include "data_generation/prompt_formatting/prompt_formatter.h"
 #include "data_model/application_options.h"
 
 struct llama_model;
@@ -31,9 +32,11 @@ class LlamaGenerator final : public DataGenerator {
    *
    * @param options Parsed application options.
    * @param model_path Filesystem path to GGUF model assets.
+  * @param prompt_formatter Formatter that produces model-specific prompts.
    */
   LlamaGenerator(const ApplicationOptions& options,
-                 const std::string& model_path);
+                 const std::string& model_path,
+                 std::shared_ptr<IPromptFormatter> prompt_formatter);
 
   ~LlamaGenerator() override;
 
@@ -132,6 +135,7 @@ class LlamaGenerator final : public DataGenerator {
   std::mt19937 rng_;
   uint32_t n_ctx_ = kDefaultContextSize;
   std::string brewery_system_prompt_;
+  std::shared_ptr<IPromptFormatter> prompt_formatter_;
 };
 
 #endif  // BIERGARTEN_PIPELINE_INCLUDES_DATA_GENERATION_LLAMA_GENERATOR_H_

pipeline/includes/data_generation/llama_generator_helpers.h

@@ -12,7 +12,6 @@
 #include <string>
 #include <string_view>
 
-struct llama_model;
 struct llama_vocab;
 using llama_token = int32_t;
 
@@ -26,18 +25,6 @@ using llama_token = int32_t;
 std::string PrepareRegionContext(std::string_view region_context,
                                  size_t max_chars = 2000);
 
-/**
- * @brief Applies model chat template to system and user prompts.
- *
- * @param model Loaded llama model.
- * @param system_prompt System prompt text.
- * @param user_prompt User prompt text.
- * @return Model-formatted prompt.
- */
-std::string ToChatPrompt(const llama_model* model,
-                         const std::string& system_prompt,
-                         const std::string& user_prompt);
-
 /**
  * @brief Decodes a sampled token and appends it to output text.
  *
@@ -58,6 +45,7 @@ void AppendTokenPiece(const llama_vocab* vocab, llama_token token,
  */
 std::optional<std::string> ValidateBreweryJson(const std::string& raw,
                                                std::string& name_out,
-                                               std::string& description_out);
+                                               std::string& description_out,
+                                               std::string& reasoning_out);
 
 #endif  // BIERGARTEN_PIPELINE_INCLUDES_DATA_GENERATION_LLAMA_GENERATOR_HELPERS_H_

pipeline/includes/data_generation/prompt_formatting/gemma4_jinja_prompt_formatter.h

@@ -0,0 +1,15 @@
+#pragma once
+
+#include <string>
+#include <string_view>
+
+#include "data_generation/prompt_formatting/prompt_formatter.h"
+
+class Gemma4JinjaPromptFormatter final : public IPromptFormatter {
+ public:
+  Gemma4JinjaPromptFormatter() = default;
+  ~Gemma4JinjaPromptFormatter() override = default;
+
+  [[nodiscard]] std::string Format(std::string_view system_prompt,
+                                   std::string_view user_prompt) const override;
+};

pipeline/includes/data_generation/prompt_formatting/prompt_formatter.h

@@ -0,0 +1,18 @@
+#pragma once
+
+#include <string>
+#include <string_view>
+
+class IPromptFormatter {
+ public:
+  IPromptFormatter() = default;
+  IPromptFormatter(const IPromptFormatter&) = delete;
+  IPromptFormatter& operator=(const IPromptFormatter&) = delete;
+  IPromptFormatter(IPromptFormatter&&) = delete;
+  IPromptFormatter& operator=(IPromptFormatter&&) = delete;
+  virtual ~IPromptFormatter() = default;
+
+  [[nodiscard]] virtual std::string Format(
+      std::string_view system_prompt,
+      std::string_view user_prompt) const = 0;
+};