Merge pull request #227 from Cloud-Code-AI/226-add-double-parsing-for…

…-json

fix: added llm based json parsing handling

kaizen/generator/ui.py

@@ -81,11 +81,7 @@ def generate_module_tests(self, web_content: str, test_modules: dict, web_url: s
         This method generates UI testing points for all modules.
         """
         ui_tests = test_modules
-        total_usage = {
-            "prompt_tokens": 0,
-            "completion_tokens": 0,
-            "total_tokens": 0
-        }
+        total_usage = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
         for module in ui_tests:
             for test in module["tests"]:
                 test_description = test["test_description"]

kaizen/helpers/parser.py

@@ -24,6 +24,19 @@ def extract_json(text):
     return parsed_data
 
 
+def extract_json_with_llm_retry(provider, text, total_usage):
+    try:
+        json_data = extract_json(text)
+        return json_data, total_usage
+    except Exception as e:
+        print(f"Error parsing json: {e}")
+        prompt = f"Help me fix this json data: {text}"
+        resp, usage = provider.chat_completion(prompt)
+        total_usage = provider.update_usage(total_usage, usage)
+        json_data = extract_json(resp)
+        return json_data, total_usage
+
+
 def extract_multi_json(text):
     start_index = text.find("[")
     end_index = text.rfind("]") + 1

kaizen/llms/prompts/work_summary_prompts.py

@@ -39,4 +39,4 @@
 7. Maintain a consistent, readable tone.  
   
 PATCH DATA: {PATCH_DATA}  
-"""  
+"""

kaizen/reviewer/code_review.py

@@ -63,15 +63,13 @@ def review_pull_request(
             PULL_REQUEST_DESC=pull_request_desc,
             CODE_DIFF=diff_text,
         )
-        total_usage = {
-            "prompt_tokens": 0,
-            "completion_tokens": 0,
-            "total_tokens": 0
-        }
+        total_usage = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
         if self.provider.is_inside_token_limit(PROMPT=prompt):
             self.logger.debug("Processing Directly from Diff")
             resp, usage = self.provider.chat_completion(prompt, user=user)
-            review_json = parser.extract_json(resp)
+            review_json, total_usage = parser.extract_json_with_llm_retry(
+                self.provider, resp, total_usage
+            )
             reviews = review_json["review"]
             total_usage = self.provider.update_usage(total_usage, usage)
         else:
@@ -90,12 +88,16 @@ def review_pull_request(
                         PULL_REQUEST_DESC=pull_request_desc,
                         FILE_PATCH=patch_details,
                     )
-                    if not self.provider.is_inside_token_limit(PROMPT=prompt, percentage=85):
+                    if not self.provider.is_inside_token_limit(
+                        PROMPT=prompt, percentage=85
+                    ):
                         # TODO: Chunk this big files and process them
                         continue
                     resp, usage = self.provider.chat_completion(prompt, user=user)
                     total_usage = self.provider.update_usage(total_usage, usage)
-                    review_json = parser.extract_json(resp)
+                    review_json, total_usage = parser.extract_json_with_llm_retry(
+                        self.provider, resp, total_usage
+                    )
                     reviews.extend(review_json["review"])
 
         topics = self.merge_topics(reviews=reviews)
@@ -128,16 +130,15 @@ def generate_pull_request_desc(
             CODE_DIFF=diff_text,
         )
 
-        total_usage = {
-            "prompt_tokens": 0,
-            "completion_tokens": 0,
-            "total_tokens": 0
-        }
+        total_usage = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
         if self.provider.is_inside_token_limit(PROMPT=prompt):
             self.logger.debug("Processing Directly from Diff")
             resp, usage = self.provider.chat_completion(prompt, user=user)
             total_usage = self.provider.update_usage(total_usage, usage)
-            desc = parser.extract_json(resp)["desc"]
+            json_data = parser.extract_json_with_llm_retry(
+                self.provider, resp, total_usage
+            )
+            desc = json_data["desc"]
         else:
             self.logger.debug("Processing Based on files")
             # We recurrsively get feedback for files and then get basic summary
@@ -159,13 +160,18 @@ def generate_pull_request_desc(
                         continue
                     resp, usage = self.provider.chat_completion(prompt, user=user)
                     total_usage = self.provider.update_usage(total_usage, usage)
-                    desc_json = parser.extract_json(resp)
+                    desc_json, total_usage = parser.extract_json_with_llm_retry(
+                        self.provider, resp, total_usage
+                    )
                     descs.append(desc_json["desc"])
 
             prompt = MERGE_PR_DESCRIPTION_PROMPT.format(DESCS=json.dumps(descs))
             resp, usage = self.provider.chat_completion(prompt, user=user)
             total_usage = self.provider.update_usage(total_usage, usage)
-            desc = parser.extract_json(resp)["desc"]
+            resp, total_usage = parser.extract_json_with_llm_retry(
+                self.provider, resp, total_usage
+            )
+            desc = resp["desc"]
         body = output.create_pr_description(desc, pull_request_desc)
         prompt_cost, completion_cost = self.provider.get_usage_cost(
             total_usage=total_usage

kaizen/reviewer/work_summarizer.py

@@ -22,11 +22,7 @@ def generate_work_summaries(
         summaries = []
         # Try to merge the files untill LLM can process the response
         combined_diff_data = ""
-        total_usage = {
-            "prompt_tokens": 0,
-            "completion_tokens": 0,
-            "total_tokens": 0
-        }
+        total_usage = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
         for file_dict in diff_file_data:
             temp_prompt = combined_diff_data
             temp_prompt += f"""\n---->\nFile Name: {file_dict["file"]}\nPatch: {file_dict["patch"]}\n Status: {file_dict["status"]}"""