update Gemini API : use thinking levels

copienator/commands/correction.py

@@ -20,6 +20,8 @@ from copienator import configuration as config
 from copienator.commands import grouping
 from copienator.answer_origins import save_origins
 from copienator import prompting
+from copienator.gemini_config import thinking_config
+from copienator import configuration as settings
 from copienator.feedback_boxes import valid_feedback_box
 from copienator.latex_json import repair_latex_json
 from copienator import (
@@ -279,6 +281,9 @@ def call_gemini_with_retries(model_id, contents, config,
         # Switch to fallback immediately if quota was exhausted by another thread
         if model_id == MODEL_ID_pro and pro_quota_exhausted and fallback_model_id:
             model_id = fallback_model_id
+            config = config.model_copy(update={"thinking_config": thinking_config(
+                settings.THINKING_LEVEL_FLASH
+            )})
 
         try:
             full_response_text = ""
@@ -312,6 +317,9 @@ def call_gemini_with_retries(model_id, contents, config,
             if is_quota_error and model_id == MODEL_ID_pro and fallback_model_id:
                 tprint(f"\tGemini Pro quota hit ({e}). \n\n\tFalling back to Flash permanently...")
                 model_id = fallback_model_id
+                config = config.model_copy(update={"thinking_config": thinking_config(
+                    settings.THINKING_LEVEL_FLASH
+                )})
                 pro_quota_exhausted = True
                 continue # Retry immediately with Flash
 
@@ -506,7 +514,11 @@ def process_single_task(task_tuple, precomputed_response=None):
                     flash_count += 1
 
         try:
-            contents, config = prompting.generate_request(INPUT_DIR, file_path, label)
+            contents, config = prompting.generate_request(
+                INPUT_DIR, file_path, label,
+                thinking_level=(settings.THINKING_LEVEL_FLASH if use_flash
+                                else settings.THINKING_LEVEL_PRO),
+            )
             model_to_use = MODEL_ID_flash if use_flash else MODEL_ID_pro
 
             if precomputed_response:
@@ -891,13 +903,10 @@ def run_configured(args: argparse.Namespace) -> ExitCode:
                                 ],
                             }
                         ],
-                        "generation_config": {
-                            "temperature": 1.0,
-                            "topP": 0.95,
-                            "maxOutputTokens": 65535,
-                            "responseMimeType": "application/json",
-                            "responseSchema": prompting.UNROLLED_SCHEMA,
-                        },
+                        "generation_config": prompting.batch_generation_config(
+                            settings.THINKING_LEVEL_FLASH if use_flash
+                            else settings.THINKING_LEVEL_PRO
+                        ),
                     },
                 }
                 line = json.dumps(request)