diff --git a/lightrag/llm.py b/lightrag/llm.py
index 6a191a0f..12a4d5a6 100644
--- a/lightrag/llm.py
+++ b/lightrag/llm.py
@@ -29,7 +29,11 @@
 from pydantic import BaseModel, Field
 from typing import List, Dict, Callable, Any
 from .base import BaseKVStorage
-from .utils import compute_args_hash, wrap_embedding_func_with_attrs
+from .utils import (
+    compute_args_hash,
+    wrap_embedding_func_with_attrs,
+    locate_json_string_body_from_string,
+)
 
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
@@ -66,9 +70,14 @@ async def openai_complete_if_cache(
         if if_cache_return is not None:
             return if_cache_return["return"]
 
-    response = await openai_async_client.chat.completions.create(
-        model=model, messages=messages, **kwargs
-    )
+    if "response_format" in kwargs:
+        response = await openai_async_client.beta.chat.completions.parse(
+            model=model, messages=messages, **kwargs
+        )
+    else:
+        response = await openai_async_client.chat.completions.create(
+            model=model, messages=messages, **kwargs
+        )
     content = response.choices[0].message.content
     if r"\u" in content:
         content = content.encode("utf-8").decode("unicode_escape")
@@ -301,7 +310,7 @@ async def ollama_model_if_cache(
     model, prompt, system_prompt=None, history_messages=[], **kwargs
 ) -> str:
     kwargs.pop("max_tokens", None)
-    kwargs.pop("response_format", None)
+    # kwargs.pop("response_format", None) # allow json
     host = kwargs.pop("host", None)
     timeout = kwargs.pop("timeout", None)
 
@@ -345,9 +354,9 @@ def initialize_lmdeploy_pipeline(
         backend_config=TurbomindEngineConfig(
             tp=tp, model_format=model_format, quant_policy=quant_policy
         ),
-        chat_template_config=ChatTemplateConfig(model_name=chat_template)
-        if chat_template
-        else None,
+        chat_template_config=(
+            ChatTemplateConfig(model_name=chat_template) if chat_template else None
+        ),
         log_level="WARNING",
     )
     return lmdeploy_pipe
@@ -458,9 +467,16 @@ async def lmdeploy_model_if_cache(
     return response
 
 
+class GPTKeywordExtractionFormat(BaseModel):
+    high_level_keywords: List[str]
+    low_level_keywords: List[str]
+
+
 async def gpt_4o_complete(
-    prompt, system_prompt=None, history_messages=[], **kwargs
+    prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs
 ) -> str:
+    if keyword_extraction:
+        kwargs["response_format"] = GPTKeywordExtractionFormat
     return await openai_complete_if_cache(
         "gpt-4o",
         prompt,
@@ -471,8 +487,10 @@ async def gpt_4o_complete(
 
 
 async def gpt_4o_mini_complete(
-    prompt, system_prompt=None, history_messages=[], **kwargs
+    prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs
 ) -> str:
+    if keyword_extraction:
+        kwargs["response_format"] = GPTKeywordExtractionFormat
     return await openai_complete_if_cache(
         "gpt-4o-mini",
         prompt,
@@ -483,45 +501,56 @@ async def gpt_4o_mini_complete(
 
 
 async def azure_openai_complete(
-    prompt, system_prompt=None, history_messages=[], **kwargs
+    prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs
 ) -> str:
-    return await azure_openai_complete_if_cache(
+    result = await azure_openai_complete_if_cache(
         "conversation-4o-mini",
         prompt,
         system_prompt=system_prompt,
         history_messages=history_messages,
         **kwargs,
     )
+    if keyword_extraction:  # TODO: use JSON API
+        return locate_json_string_body_from_string(result)
+    return result
 
 
 async def bedrock_complete(
-    prompt, system_prompt=None, history_messages=[], **kwargs
+    prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs
 ) -> str:
-    return await bedrock_complete_if_cache(
+    result = await bedrock_complete_if_cache(
         "anthropic.claude-3-haiku-20240307-v1:0",
         prompt,
         system_prompt=system_prompt,
         history_messages=history_messages,
         **kwargs,
     )
+    if keyword_extraction:  # TODO: use JSON API
+        return locate_json_string_body_from_string(result)
+    return result
 
 
 async def hf_model_complete(
-    prompt, system_prompt=None, history_messages=[], **kwargs
+    prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs
 ) -> str:
     model_name = kwargs["hashing_kv"].global_config["llm_model_name"]
-    return await hf_model_if_cache(
+    result = await hf_model_if_cache(
         model_name,
         prompt,
         system_prompt=system_prompt,
         history_messages=history_messages,
         **kwargs,
     )
+    if keyword_extraction:  # TODO: use JSON API
+        return locate_json_string_body_from_string(result)
+    return result
 
 
 async def ollama_model_complete(
-    prompt, system_prompt=None, history_messages=[], **kwargs
+    prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs
 ) -> str:
+    if keyword_extraction:
+        kwargs["format"] = "json"
     model_name = kwargs["hashing_kv"].global_config["llm_model_name"]
     return await ollama_model_if_cache(
         model_name,
diff --git a/lightrag/operate.py b/lightrag/operate.py
index 94cd412b..74c8e5f6 100644
--- a/lightrag/operate.py
+++ b/lightrag/operate.py
@@ -17,7 +17,6 @@
     split_string_by_multi_markers,
     truncate_list_by_token_size,
     process_combine_contexts,
-    locate_json_string_body_from_string,
 )
 from .base import (
     BaseGraphStorage,
@@ -461,12 +460,12 @@ async def kg_query(
     use_model_func = global_config["llm_model_func"]
     kw_prompt_temp = PROMPTS["keywords_extraction"]
     kw_prompt = kw_prompt_temp.format(query=query, examples=examples, language=language)
-    result = await use_model_func(kw_prompt)
+    result = await use_model_func(kw_prompt, keyword_extraction=True)
     logger.info("kw_prompt result:")
     print(result)
     try:
-        json_text = locate_json_string_body_from_string(result)
-        keywords_data = json.loads(json_text)
+        # json_text = locate_json_string_body_from_string(result) # handled in use_model_func
+        keywords_data = json.loads(result)
         hl_keywords = keywords_data.get("high_level_keywords", [])
         ll_keywords = keywords_data.get("low_level_keywords", [])
 
diff --git a/lightrag/utils.py b/lightrag/utils.py
index bdd1aa9e..8997b651 100644
--- a/lightrag/utils.py
+++ b/lightrag/utils.py
@@ -54,7 +54,7 @@ def locate_json_string_body_from_string(content: str) -> Union[str, None]:
             maybe_json_str = maybe_json_str.replace("\\n", "")
             maybe_json_str = maybe_json_str.replace("\n", "")
             maybe_json_str = maybe_json_str.replace("'", '"')
-            json.loads(maybe_json_str)
+            # json.loads(maybe_json_str) # don't check here, cannot validate schema after all
             return maybe_json_str
     except Exception:
         pass