MindsDB 25.7.2.0__py3-none-any.whl → 25.7.4.0__py3-none-any.whl

mindsdb/integrations/handlers/langchain_handler/langchain_handler.py

@@ -13,7 +13,10 @@ import pandas as pd
 
 from mindsdb.interfaces.agents.safe_output_parser import SafeOutputParser
 from mindsdb.interfaces.agents.langchain_agent import (
-    get_llm_provider, get_embedding_model_provider, create_chat_model, get_chat_model_params
+    get_llm_provider,
+    get_embedding_model_provider,
+    create_chat_model,
+    get_chat_model_params,
 )
 
 from mindsdb.interfaces.agents.constants import (
@@ -24,19 +27,21 @@ from mindsdb.interfaces.agents.constants import (
     DEFAULT_MAX_TOKENS,
     DEFAULT_MODEL_NAME,
     USER_COLUMN,
-    ASSISTANT_COLUMN
+    ASSISTANT_COLUMN,
 )
 from mindsdb.integrations.utilities.rag.settings import DEFAULT_RAG_PROMPT_TEMPLATE
 from mindsdb.integrations.handlers.langchain_handler.tools import setup_tools
 from mindsdb.integrations.libs.base import BaseMLEngine
 from mindsdb.interfaces.storage.model_fs import HandlerStorage, ModelStorage
-from mindsdb.integrations.handlers.langchain_embedding_handler.langchain_embedding_handler import construct_model_from_args
-from mindsdb.integrations.handlers.openai_handler.constants import CHAT_MODELS  # noqa, for dependency checker
+from mindsdb.integrations.handlers.langchain_embedding_handler.langchain_embedding_handler import (
+    construct_model_from_args,
+)
+from mindsdb.integrations.handlers.openai_handler.constants import CHAT_MODELS  # noqa: F401 - for dependency checker
 
 from mindsdb.utilities import log
 from mindsdb.utilities.context_executor import ContextThreadPoolExecutor
 
-_PARSING_ERROR_PREFIXES = ['An output parsing error occured', 'Could not parse LLM output']
+_PARSING_ERROR_PREFIXES = ["An output parsing error occured", "Could not parse LLM output"]
 
 logger = log.getLogger(__name__)
 
@@ -58,13 +63,10 @@ class LangChainHandler(BaseMLEngine):
         - python_repl
         - serper.dev search
     """
-    name = 'langchain'
 
-    def __init__(
-            self,
-            model_storage: ModelStorage,
-            engine_storage: HandlerStorage,
-            **kwargs):
+    name = "langchain"
+
+    def __init__(self, model_storage: ModelStorage, engine_storage: HandlerStorage, **kwargs):
         super().__init__(model_storage, engine_storage, **kwargs)
         # if True, the target column name does not have to be specified at creation time.
         self.generative = True
@@ -81,77 +83,78 @@ class LangChainHandler(BaseMLEngine):
                 #
                 # Ideally, in the future, we would write a parser that is more robust and flexible than the one Langchain uses.
                 # Response is wrapped in ``
-                logger.info('Handling parsing error, salvaging response...')
-                response_output = response.split('`')
+                logger.info("Handling parsing error, salvaging response...")
+                response_output = response.split("`")
                 if len(response_output) >= 2:
                     response = response_output[-2]
 
                 # Wrap response in Langchain conversational react format.
-                langchain_react_formatted_response = f'''Thought: Do I need to use a tool? No
-AI: {response}'''
+                langchain_react_formatted_response = f"""Thought: Do I need to use a tool? No
+AI: {response}"""
                 return langchain_react_formatted_response
-        return f'Agent failed with error:\n{str(error)}...'
+        return f"Agent failed with error:\n{str(error)}..."
 
     def create(self, target: str, args: Dict = None, **kwargs):
-        self.default_agent_tools = args.get('tools', self.default_agent_tools)
-
-        args = args['using']
-        args['target'] = target
-        args['model_name'] = args.get('model_name', DEFAULT_MODEL_NAME)
-        args['provider'] = args.get('provider', get_llm_provider(args))
-        args['embedding_model_provider'] = args.get('embedding_model', get_embedding_model_provider(args))
-        if args.get('mode') == 'retrieval':
+        self.default_agent_tools = args.get("tools", self.default_agent_tools)
+
+        args = args["using"]
+        args["target"] = target
+        args["model_name"] = args.get("model_name", DEFAULT_MODEL_NAME)
+        args["provider"] = args.get("provider", get_llm_provider(args))
+        args["embedding_model_provider"] = args.get("embedding_model", get_embedding_model_provider(args))
+        if args.get("mode") == "retrieval":
             # use default prompt template for retrieval i.e. RAG if not provided
             if "prompt_template" not in args:
                 args["prompt_template"] = DEFAULT_RAG_PROMPT_TEMPLATE
 
-        self.model_storage.json_set('args', args)
+        self.model_storage.json_set("args", args)
 
     @staticmethod
     def create_validation(_, args: Dict = None, **kwargs):
-        if 'using' not in args:
+        if "using" not in args:
             raise Exception("LangChain engine requires a USING clause! Refer to its documentation for more details.")
         else:
-            args = args['using']
-        if 'prompt_template' not in args:
-            if not args.get('mode') == 'retrieval':
-                raise ValueError('Please provide a `prompt_template` for this engine.')
+            args = args["using"]
+        if "prompt_template" not in args:
+            if not args.get("mode") == "retrieval":
+                raise ValueError("Please provide a `prompt_template` for this engine.")
 
     def predict(self, df: pd.DataFrame, args: Dict = None) -> pd.DataFrame:
         """
         Dispatch is performed depending on the underlying model type. Currently, only the default text completion
         is supported.
         """
-        pred_args = args['predict_params'] if args else {}
-        args = self.model_storage.json_get('args')
-        if 'prompt_template' not in args and 'prompt_template' not in pred_args:
+        pred_args = args["predict_params"] if args else {}
+        args = self.model_storage.json_get("args")
+        if "prompt_template" not in args and "prompt_template" not in pred_args:
             raise ValueError("This model expects a `prompt_template`, please provide one.")
         # Back compatibility for old models
-        args['provider'] = args.get('provider', get_llm_provider(args))
-        args['embedding_model_provider'] = args.get('embedding_model', get_embedding_model_provider(args))
+        args["provider"] = args.get("provider", get_llm_provider(args))
+        args["embedding_model_provider"] = args.get("embedding_model", get_embedding_model_provider(args))
 
         df = df.reset_index(drop=True)
 
-        if pred_args.get('mode') == 'chat_model':
+        if pred_args.get("mode") == "chat_model":
             return self.call_llm(df, args, pred_args)
 
         agent = self.create_agent(df, args, pred_args)
         # Use last message as prompt, remove other questions.
-        user_column = args.get('user_column', USER_COLUMN)
+        user_column = args.get("user_column", USER_COLUMN)
         if user_column not in df.columns:
             raise Exception(
-                f"Expected user input in column `{user_column}`, which is not found in the input data. Either provide the column, or redefine the expected column at model creation (`USING user_column = 'value'`)")  # noqa
+                f"Expected user input in column `{user_column}`, which is not found in the input data. Either provide the column, or redefine the expected column at model creation (`USING user_column = 'value'`)"
+            )  # noqa
         df.iloc[:-1, df.columns.get_loc(user_column)] = None
         return self.run_agent(df, agent, args, pred_args)
 
     def call_llm(self, df, args=None, pred_args=None):
         llm = create_chat_model({**args, **pred_args})
 
-        user_column = args.get('user_column', USER_COLUMN)
-        assistant_column = args.get('assistant_column', ASSISTANT_COLUMN)
+        user_column = args.get("user_column", USER_COLUMN)
+        assistant_column = args.get("assistant_column", ASSISTANT_COLUMN)
 
         question = df[user_column].iloc[-1]
-        resp = llm([HumanMessage(question)], stop=['\nObservation:', '\n\tObservation:'])
+        resp = llm([HumanMessage(question)], stop=["\nObservation:", "\n\tObservation:"])
 
         return pd.DataFrame([resp.content], columns=[assistant_column])
 
@@ -162,25 +165,22 @@ AI: {response}'''
         model_kwargs = get_chat_model_params({**args, **pred_args})
         llm = create_chat_model({**args, **pred_args})
 
-        tools = setup_tools(llm,
-                            model_kwargs,
-                            pred_args,
-                            self.default_agent_tools)
+        tools = setup_tools(llm, model_kwargs, pred_args, self.default_agent_tools)
 
         # Prefer prediction prompt template over original if provided.
-        prompt_template = pred_args.get('prompt_template', args['prompt_template'])
-        if 'context' in pred_args:
-            prompt_template += '\n\n' + 'Useful information:\n' + pred_args['context'] + '\n'
+        prompt_template = pred_args.get("prompt_template", args["prompt_template"])
+        if "context" in pred_args:
+            prompt_template += "\n\n" + "Useful information:\n" + pred_args["context"] + "\n"
 
         # Set up memory.
-        memory = ConversationSummaryBufferMemory(llm=llm,
-                                                 max_token_limit=model_kwargs.get('max_tokens', DEFAULT_MAX_TOKENS),
-                                                 memory_key='chat_history')
+        memory = ConversationSummaryBufferMemory(
+            llm=llm, max_token_limit=model_kwargs.get("max_tokens", DEFAULT_MAX_TOKENS), memory_key="chat_history"
+        )
         memory.chat_memory.messages.insert(0, SystemMessage(content=prompt_template))
         # User - Assistant conversation. All except the last message.
-        user_column = args.get('user_column', USER_COLUMN)
-        assistant_column = args.get('assistant_column', ASSISTANT_COLUMN)
-        for row in df[:-1].to_dict('records'):
+        user_column = args.get("user_column", USER_COLUMN)
+        assistant_column = args.get("assistant_column", ASSISTANT_COLUMN)
+        for row in df[:-1].to_dict("records"):
             question = row[user_column]
             answer = row[assistant_column]
             if question:
@@ -188,45 +188,47 @@ AI: {response}'''
             if answer:
                 memory.chat_memory.add_ai_message(answer)
 
-        agent_type = args.get('agent_type', DEFAULT_AGENT_TYPE)
+        agent_type = args.get("agent_type", DEFAULT_AGENT_TYPE)
         agent_executor = initialize_agent(
             tools,
             llm,
             agent=agent_type,
             # Use custom output parser to handle flaky LLMs that don't ALWAYS conform to output format.
-            agent_kwargs={'output_parser': SafeOutputParser()},
+            agent_kwargs={"output_parser": SafeOutputParser()},
             # Calls the agent’s LLM Chain one final time to generate a final answer based on the previous steps
-            early_stopping_method='generate',
+            early_stopping_method="generate",
             handle_parsing_errors=self._handle_parsing_errors,
             # Timeout per agent invocation.
-            max_execution_time=pred_args.get('timeout_seconds', args.get('timeout_seconds', DEFAULT_AGENT_TIMEOUT_SECONDS)),
-            max_iterations=pred_args.get('max_iterations', args.get('max_iterations', DEFAULT_MAX_ITERATIONS)),
+            max_execution_time=pred_args.get(
+                "timeout_seconds", args.get("timeout_seconds", DEFAULT_AGENT_TIMEOUT_SECONDS)
+            ),
+            max_iterations=pred_args.get("max_iterations", args.get("max_iterations", DEFAULT_MAX_ITERATIONS)),
             memory=memory,
-            verbose=pred_args.get('verbose', args.get('verbose', True))
+            verbose=pred_args.get("verbose", args.get("verbose", True)),
         )
         return agent_executor
 
     def run_agent(self, df: pd.DataFrame, agent: AgentExecutor, args: Dict, pred_args: Dict) -> pd.DataFrame:
         # Prefer prediction time prompt template, if available.
-        base_template = pred_args.get('prompt_template', args['prompt_template'])
+        base_template = pred_args.get("prompt_template", args["prompt_template"])
 
         input_variables = []
         matches = list(re.finditer("{{(.*?)}}", base_template))
 
         for m in matches:
-            input_variables.append(m[0].replace('{', '').replace('}', ''))
+            input_variables.append(m[0].replace("{", "").replace("}", ""))
         empty_prompt_ids = np.where(df[input_variables].isna().all(axis=1).values)[0]
 
-        base_template = base_template.replace('{{', '{').replace('}}', '}')
+        base_template = base_template.replace("{{", "{").replace("}}", "}")
         prompts = []
 
-        user_column = args.get('user_column', USER_COLUMN)
+        user_column = args.get("user_column", USER_COLUMN)
         for i, row in df.iterrows():
             if i not in empty_prompt_ids:
                 prompt = PromptTemplate(input_variables=input_variables, template=base_template)
                 kwargs = {}
                 for col in input_variables:
-                    kwargs[col] = row[col] if row[col] is not None else ''  # add empty quote if data is missing
+                    kwargs[col] = row[col] if row[col] is not None else ""  # add empty quote if data is missing
                 prompts.append(prompt.format(**kwargs))
             elif row.get(user_column):
                 # Just add prompt
@@ -234,32 +236,37 @@ AI: {response}'''
 
         def _invoke_agent_executor_with_prompt(agent_executor, prompt):
             if not prompt:
-                return ''
+                return ""
             try:
                 answer = agent_executor.invoke(prompt)
             except Exception as e:
                 answer = str(e)
                 if not answer.startswith("Could not parse LLM output: `"):
                     raise e
-                answer = {'output': answer.removeprefix("Could not parse LLM output: `").removesuffix("`")}
+                answer = {"output": answer.removeprefix("Could not parse LLM output: `").removesuffix("`")}
 
-            if 'output' not in answer:
+            if "output" not in answer:
                 # This should never happen unless Langchain changes invoke output format, but just in case.
                 return agent_executor.run(prompt)
-            return answer['output']
+            return answer["output"]
 
         completions = []
         # max_workers defaults to number of processors on the machine multiplied by 5.
         # https://docs.python.org/3/library/concurrent.futures.html#concurrent.futures.ThreadPoolExecutor
-        max_workers = args.get('max_workers', None)
-        agent_timeout_seconds = args.get('timeout', DEFAULT_AGENT_TIMEOUT_SECONDS)
+        max_workers = args.get("max_workers", None)
+        agent_timeout_seconds = args.get("timeout", DEFAULT_AGENT_TIMEOUT_SECONDS)
         executor = ContextThreadPoolExecutor(max_workers=max_workers)
         futures = [executor.submit(_invoke_agent_executor_with_prompt, agent, prompt) for prompt in prompts]
         try:
             for future in as_completed(futures, timeout=agent_timeout_seconds):
                 completions.append(future.result())
         except TimeoutError:
-            completions.append("I'm sorry! I couldn't come up with a response in time. Please try again.")
+            completions.append(
+                f"I'm sorry! I couldn't generate a response within the allotted time ({agent_timeout_seconds} seconds). "
+                "If you need more time for processing, you can adjust the timeout settings. "
+                "Please refer to the documentation for instructions on how to change the timeout value. "
+                "Feel free to try your request again."
+            )
         # Can't use ThreadPoolExecutor as context manager since we need wait=False.
         executor.shutdown(wait=False)
 
@@ -267,13 +274,13 @@ AI: {response}'''
         for i in sorted(empty_prompt_ids)[:-1]:
             completions.insert(i, None)
 
-        pred_df = pd.DataFrame(completions, columns=[args['target']])
+        pred_df = pd.DataFrame(completions, columns=[args["target"]])
 
         return pred_df
 
     def describe(self, attribute: Optional[str] = None) -> pd.DataFrame:
-        tables = ['info']
-        return pd.DataFrame(tables, columns=['tables'])
+        tables = ["info"]
+        return pd.DataFrame(tables, columns=["tables"])
 
     def finetune(self, df: Optional[pd.DataFrame] = None, args: Optional[Dict] = None) -> None:
-        raise NotImplementedError('Fine-tuning is not supported for LangChain models')
+        raise NotImplementedError("Fine-tuning is not supported for LangChain models")

mindsdb/integrations/handlers/lightwood_handler/requirements.txt

@@ -1,4 +1,4 @@
-lightwood>=25.5.2.2
-lightwood[extra]>=25.5.2.2
-lightwood[xai]>=25.5.2.2
-type_infer==0.0.22
+lightwood>=25.7.5.1
+lightwood[extra]>=25.7.5.1
+lightwood[xai]>=25.7.5.1
+type_infer==0.0.23

mindsdb/integrations/handlers/litellm_handler/litellm_handler.py

@@ -2,7 +2,8 @@ import ast
 from typing import Dict, Optional, List
 
 
-from litellm import completion, batch_completion, embedding, acompletion
+from litellm import completion, batch_completion, embedding, acompletion, supports_response_schema
+
 import pandas as pd
 
 from mindsdb.integrations.libs.base import BaseMLEngine
@@ -58,6 +59,15 @@ class LiteLLMHandler(BaseMLEngine):
     @classmethod
     def completion(cls, provider: str, model: str, messages: List[dict], args: dict):
         model, args = cls.prepare_arguments(provider, model, args)
+        json_output = args.pop("json_output", False)
+
+        supports_json_output = supports_response_schema(model=model, custom_llm_provider=provider)
+
+        if json_output and supports_json_output:
+            args["response_format"] = {"type": "json_object"}
+        else:
+            args["response_format"] = None
+
         return completion(model=model, messages=messages, stream=False, **args)
 
     def create(
@@ -77,6 +87,7 @@ class LiteLLMHandler(BaseMLEngine):
 
         # check engine_storage for api_key
         input_args.update({k: v for k, v in ml_engine_args.items()})
+        input_args["target"] = target
 
         # validate args
         export_args = CompletionParameters(**input_args).model_dump()
@@ -94,6 +105,8 @@ class LiteLLMHandler(BaseMLEngine):
         # validate args
         args = CompletionParameters(**input_args).model_dump()
 
+        target = args.pop("target")
+
         # build messages
         self._build_messages(args, df)
 
@@ -103,12 +116,12 @@ class LiteLLMHandler(BaseMLEngine):
         if len(args["messages"]) > 1:
             # if more than one message, use batch completion
             responses = batch_completion(**args)
-            return pd.DataFrame({"result": [response.choices[0].message.content for response in responses]})
+            return pd.DataFrame({target: [response.choices[0].message.content for response in responses]})
 
         # run completion
         response = completion(**args)
 
-        return pd.DataFrame({"result": [response.choices[0].message.content]})
+        return pd.DataFrame({target: [response.choices[0].message.content]})
 
     @staticmethod
     def _prompt_to_messages(prompt: str, **kwargs) -> List[Dict]:

mindsdb/integrations/handlers/litellm_handler/settings.py

@@ -31,7 +31,8 @@ class CompletionParameters(BaseModel):
     # set api_base, api_version, api_key
     base_url: Optional[str] = None  # Base URL of the API.
     api_version: Optional[str] = None  # Version of the API to be used.
-    api_key: str  # API key for authentication.
+    api_key: Optional[str] = None  # API key for authentication.
+    target: Optional[str] = None  # the name of output column
 
     class Config:
         extra = Extra.forbid

mindsdb/integrations/handlers/llama_index_handler/requirements.txt

@@ -1,4 +1,4 @@
-llama-index==0.12.28
+llama-index==0.12.41
 pydantic-settings >= 2.1.0
 llama-index-readers-web
 llama-index-embeddings-openai