Add: Json Parsing to solve Hallucination Errors (#3013)

evaluation/metrics/llm_judge.py

@@ -4,6 +4,7 @@ from collections import defaultdict
 
 import numpy as np
 from openai import OpenAI
+from mem0.memory.utils import extract_json
 
 client = OpenAI()
 
@@ -22,7 +23,7 @@ The generated answer might be much longer, but you should be generous with your
 
 For time related questions, the gold answer will be a specific date, month, year, etc. The generated answer might be much longer or use relative time references (like "last Tuesday" or "next month"), but you should be generous with your grading - as long as it refers to the same date or time period as the gold answer, it should be counted as CORRECT. Even if the format differs (e.g., "May 7th" vs "7 May"), consider it CORRECT if it's the same date.
 
-Now it’s time for the real question:
+Now it's time for the real question:
 Question: {question}
 Gold answer: {gold_answer}
 Generated answer: {generated_answer}
@@ -49,7 +50,7 @@ def evaluate_llm_judge(question, gold_answer, generated_answer):
         response_format={"type": "json_object"},
         temperature=0.0,
     )
-    label = json.loads(response.choices[0].message.content)["label"]
+    label = json.loads(extract_json(response.choices[0].message.content))["label"]
     return 1 if label == "CORRECT" else 0
 
 

mem0/embeddings/aws_bedrock.py

@@ -11,6 +11,7 @@ import numpy as np
 
 from mem0.configs.embeddings.base import BaseEmbedderConfig
 from mem0.embeddings.base import EmbeddingBase
+from mem0.memory.utils import extract_json
 
 
 class AWSBedrockEmbedding(EmbeddingBase):
@@ -74,7 +75,7 @@ class AWSBedrockEmbedding(EmbeddingBase):
                 contentType="application/json",
             )
 
-            response_body = json.loads(response.get("body").read())
+            response_body = json.loads(extract_json(response.get("body").read()))
 
             if provider == "cohere":
                 embeddings = response_body.get("embeddings")[0]

mem0/llms/aws_bedrock.py

@@ -10,6 +10,7 @@ except ImportError:
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 PROVIDERS = ["ai21", "amazon", "anthropic", "cohere", "meta", "mistral", "stability", "writer"]
 
@@ -101,7 +102,7 @@ class AWSBedrockLLM(LLMBase):
             return processed_response
 
         response_body = response.get("body").read().decode()
-        response_json = json.loads(response_body)
+        response_json = json.loads(extract_json(response_body))
         return response_json.get("content", [{"text": ""}])[0].get("text", "")
 
     def _prepare_input(

mem0/llms/azure_openai.py

@@ -6,6 +6,7 @@ from openai import AzureOpenAI
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 
 class AzureOpenAILLM(LLMBase):
@@ -53,7 +54,7 @@ class AzureOpenAILLM(LLMBase):
                     processed_response["tool_calls"].append(
                         {
                             "name": tool_call.function.name,
-                            "arguments": json.loads(tool_call.function.arguments),
+                            "arguments": json.loads(extract_json(tool_call.function.arguments)),
                         }
                     )
 

mem0/llms/deepseek.py

@@ -6,6 +6,7 @@ from openai import OpenAI
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 
 class DeepSeekLLM(LLMBase):
@@ -41,7 +42,7 @@ class DeepSeekLLM(LLMBase):
                     processed_response["tool_calls"].append(
                         {
                             "name": tool_call.function.name,
-                            "arguments": json.loads(tool_call.function.arguments),
+                            "arguments": json.loads(extract_json(tool_call.function.arguments)),
                         }
                     )
 

mem0/llms/groq.py

@@ -9,6 +9,7 @@ except ImportError:
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 
 class GroqLLM(LLMBase):
@@ -43,7 +44,7 @@ class GroqLLM(LLMBase):
                     processed_response["tool_calls"].append(
                         {
                             "name": tool_call.function.name,
-                            "arguments": json.loads(tool_call.function.arguments),
+                            "arguments": json.loads(extract_json(tool_call.function.arguments)),
                         }
                     )
 

mem0/llms/litellm.py

@@ -8,6 +8,7 @@ except ImportError:
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 
 class LiteLLM(LLMBase):
@@ -39,7 +40,7 @@ class LiteLLM(LLMBase):
                     processed_response["tool_calls"].append(
                         {
                             "name": tool_call.function.name,
-                            "arguments": json.loads(tool_call.function.arguments),
+                            "arguments": json.loads(extract_json(tool_call.function.arguments)),
                         }
                     )
 

mem0/llms/openai.py

@@ -7,6 +7,7 @@ from openai import OpenAI
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 
 class OpenAILLM(LLMBase):
@@ -62,7 +63,7 @@ class OpenAILLM(LLMBase):
                     processed_response["tool_calls"].append(
                         {
                             "name": tool_call.function.name,
-                            "arguments": json.loads(tool_call.function.arguments),
+                            "arguments": json.loads(extract_json(tool_call.function.arguments)),
                         }
                     )
 

mem0/llms/together.py

@@ -9,6 +9,7 @@ except ImportError:
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 
 class TogetherLLM(LLMBase):
@@ -43,7 +44,7 @@ class TogetherLLM(LLMBase):
                     processed_response["tool_calls"].append(
                         {
                             "name": tool_call.function.name,
-                            "arguments": json.loads(tool_call.function.arguments),
+                            "arguments": json.loads(extract_json(tool_call.function.arguments)),
                         }
                     )
 

mem0/llms/vllm.py

@@ -4,6 +4,7 @@ from typing import Dict, List, Optional
 
 from mem0.configs.llms.base import BaseLlmConfig
 from mem0.llms.base import LLMBase
+from mem0.memory.utils import extract_json
 
 
 class VllmLLM(LLMBase):
@@ -39,7 +40,7 @@ class VllmLLM(LLMBase):
                 for tool_call in response.choices[0].message.tool_calls:
                     processed_response["tool_calls"].append({
                         "name": tool_call.function.name,
-                        "arguments": json.loads(tool_call.function.arguments),
+                        "arguments": json.loads(extract_json(tool_call.function.arguments)),
                     })
 
             return processed_response

mem0/memory/utils.py

@@ -46,6 +46,20 @@ def remove_code_blocks(content: str) -> str:
     return match.group(1).strip() if match else content.strip()
 
 
+def extract_json(text):
+    """
+    Extracts JSON content from a string, removing enclosing triple backticks and optional 'json' tag if present.
+    If no code block is found, returns the text as-is.
+    """
+    text = text.strip()
+    match = re.search(r"```(?:json)?\s*(.*?)\s*```", text, re.DOTALL)
+    if match:
+        json_str = match.group(1)
+    else:
+        json_str = text  # assume it's raw JSON
+    return json_str
+
+
 def get_image_description(image_obj, llm, vision_details):
     """
     Get the description of the image

mem0/vector_stores/azure_ai_search.py

@@ -6,6 +6,7 @@ from typing import List, Optional
 from pydantic import BaseModel
 
 from mem0.vector_stores.base import VectorStoreBase
+from mem0.memory.utils import extract_json
 
 try:
     from azure.core.credentials import AzureKeyCredential
@@ -233,7 +234,7 @@ class AzureAISearch(VectorStoreBase):
 
         results = []
         for result in search_results:
-            payload = json.loads(result["payload"])
+            payload = json.loads(extract_json(result["payload"]))
             results.append(OutputData(id=result["id"], score=result["@search.score"], payload=payload))
         return results
 
@@ -288,7 +289,8 @@ class AzureAISearch(VectorStoreBase):
             result = self.search_client.get_document(key=vector_id)
         except ResourceNotFoundError:
             return None
-        return OutputData(id=result["id"], score=None, payload=json.loads(result["payload"]))
+        payload = json.loads(extract_json(result["payload"]))
+        return OutputData(id=result["id"], score=None, payload=payload)
 
     def list_cols(self) -> List[str]:
         """
@@ -335,7 +337,7 @@ class AzureAISearch(VectorStoreBase):
         search_results = self.search_client.search(search_text="*", filter=filter_expression, top=limit)
         results = []
         for result in search_results:
-            payload = json.loads(result["payload"])
+            payload = json.loads(extract_json(result["payload"]))
             results.append(OutputData(id=result["id"], score=result["@search.score"], payload=payload))
         return [results]
 

mem0/vector_stores/redis.py

@@ -12,6 +12,7 @@ from redisvl.query import VectorQuery
 from redisvl.query.filter import Tag
 
 from mem0.vector_stores.base import VectorStoreBase
+from mem0.memory.utils import extract_json
 
 logger = logging.getLogger(__name__)
 
@@ -175,7 +176,7 @@ class RedisDB(VectorStoreBase):
                         else {}
                     ),
                     **{field: result[field] for field in ["agent_id", "run_id", "user_id"] if field in result},
-                    **{k: v for k, v in json.loads(result["metadata"]).items()},
+                    **{k: v for k, v in json.loads(extract_json(result["metadata"])).items()},
                 },
             )
             for result in results
@@ -219,7 +220,7 @@ class RedisDB(VectorStoreBase):
                 else {}
             ),
             **{field: result[field] for field in ["agent_id", "run_id", "user_id"] if field in result},
-            **{k: v for k, v in json.loads(result["metadata"]).items()},
+            **{k: v for k, v in json.loads(extract_json(result["metadata"])).items()},
         }
 
         return MemoryResult(id=result["memory_id"], payload=payload)
@@ -286,7 +287,7 @@ class RedisDB(VectorStoreBase):
                             for field in ["agent_id", "run_id", "user_id"]
                             if field in result.__dict__
                         },
-                        **{k: v for k, v in json.loads(result["metadata"]).items()},
+                        **{k: v for k, v in json.loads(extract_json(result["metadata"])).items()},
                     },
                 )
                 for result in results.docs