cognee 0.5.0.dev1__py3-none-any.whl → 0.5.1.dev0__py3-none-any.whl

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/mistral/adapter.py

@@ -1,13 +1,13 @@
 import litellm
 import instructor
 from pydantic import BaseModel
-from typing import Type
+from typing import Type, Optional
 from litellm import JSONSchemaValidationError
-
+from cognee.infrastructure.files.utils.open_data_file import open_data_file
 from cognee.shared.logging_utils import get_logger
 from cognee.modules.observability.get_observe import get_observe
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
-    LLMInterface,
+from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.generic_llm_api.adapter import (
+    GenericAPIAdapter,
 )
 from cognee.infrastructure.llm.config import get_llm_config
 from cognee.shared.rate_limiting import llm_rate_limiter_context_manager
@@ -20,12 +20,14 @@ from tenacity import (
     retry_if_not_exception_type,
     before_sleep_log,
 )
+from ..types import TranscriptionReturnType
+from mistralai import Mistral
 
 logger = get_logger()
 observe = get_observe()
 
 
-class MistralAdapter(LLMInterface):
+class MistralAdapter(GenericAPIAdapter):
     """
     Adapter for Mistral AI API, for structured output generation and prompt display.
 
@@ -34,10 +36,6 @@ class MistralAdapter(LLMInterface):
     - show_prompt
     """
 
-    name = "Mistral"
-    model: str
-    api_key: str
-    max_completion_tokens: int
     default_instructor_mode = "mistral_tools"
 
     def __init__(
@@ -46,12 +44,19 @@ class MistralAdapter(LLMInterface):
         model: str,
         max_completion_tokens: int,
         endpoint: str = None,
+        transcription_model: str = None,
+        image_transcribe_model: str = None,
         instructor_mode: str = None,
     ):
-        from mistralai import Mistral
-
-        self.model = model
-        self.max_completion_tokens = max_completion_tokens
+        super().__init__(
+            api_key=api_key,
+            model=model,
+            max_completion_tokens=max_completion_tokens,
+            name="Mistral",
+            endpoint=endpoint,
+            transcription_model=transcription_model,
+            image_transcribe_model=image_transcribe_model,
+        )
 
         self.instructor_mode = instructor_mode if instructor_mode else self.default_instructor_mode
 
@@ -60,7 +65,9 @@ class MistralAdapter(LLMInterface):
             mode=instructor.Mode(self.instructor_mode),
             api_key=get_llm_config().llm_api_key,
         )
+        self.mistral_client = Mistral(api_key=self.api_key)
 
+    @observe(as_type="generation")
     @retry(
         stop=stop_after_delay(128),
         wait=wait_exponential_jitter(8, 128),
@@ -119,3 +126,41 @@ class MistralAdapter(LLMInterface):
             logger.error(f"Schema validation failed: {str(e)}")
             logger.debug(f"Raw response: {e.raw_response}")
             raise ValueError(f"Response failed schema validation: {str(e)}")
+
+    @observe(as_type="transcription")
+    @retry(
+        stop=stop_after_delay(128),
+        wait=wait_exponential_jitter(2, 128),
+        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
+        before_sleep=before_sleep_log(logger, logging.DEBUG),
+        reraise=True,
+    )
+    async def create_transcript(self, input) -> Optional[TranscriptionReturnType]:
+        """
+        Generate an audio transcript from a user query.
+
+        This method creates a transcript from the specified audio file.
+        The audio file is processed and the transcription is retrieved from the API.
+
+        Parameters:
+        -----------
+            - input: The path to the audio file that needs to be transcribed.
+
+        Returns:
+        --------
+            The generated transcription of the audio file.
+        """
+        transcription_model = self.transcription_model
+        if self.transcription_model.startswith("mistral"):
+            transcription_model = self.transcription_model.split("/")[-1]
+        file_name = input.split("/")[-1]
+        async with open_data_file(input, mode="rb") as f:
+            transcription_response = self.mistral_client.audio.transcriptions.complete(
+                model=transcription_model,
+                file={
+                    "content": f,
+                    "file_name": file_name,
+                },
+            )
+
+            return TranscriptionReturnType(transcription_response.text, transcription_response)

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/ollama/adapter.py

@@ -12,7 +12,6 @@ from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.ll
 from cognee.infrastructure.files.utils.open_data_file import open_data_file
 from cognee.shared.logging_utils import get_logger
 from cognee.shared.rate_limiting import llm_rate_limiter_context_manager
-
 from tenacity import (
     retry,
     stop_after_delay,

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/openai/adapter.py

@@ -1,4 +1,3 @@
-import base64
 import litellm
 import instructor
 from typing import Type
@@ -16,8 +15,8 @@ from tenacity import (
     before_sleep_log,
 )
 
-from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.llm_interface import (
-    LLMInterface,
+from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.generic_llm_api.adapter import (
+    GenericAPIAdapter,
 )
 from cognee.infrastructure.llm.exceptions import (
     ContentPolicyFilterError,
@@ -26,13 +25,16 @@ from cognee.shared.rate_limiting import llm_rate_limiter_context_manager
 from cognee.infrastructure.files.utils.open_data_file import open_data_file
 from cognee.modules.observability.get_observe import get_observe
 from cognee.shared.logging_utils import get_logger
+from cognee.infrastructure.llm.structured_output_framework.litellm_instructor.llm.types import (
+    TranscriptionReturnType,
+)
 
 logger = get_logger()
 
 observe = get_observe()
 
 
-class OpenAIAdapter(LLMInterface):
+class OpenAIAdapter(GenericAPIAdapter):
     """
     Adapter for OpenAI's GPT-3, GPT-4 API.
 
@@ -53,12 +55,7 @@ class OpenAIAdapter(LLMInterface):
     - MAX_RETRIES
     """
 
-    name = "OpenAI"
-    model: str
-    api_key: str
-    api_version: str
     default_instructor_mode = "json_schema_mode"
-
     MAX_RETRIES = 5
 
     """Adapter for OpenAI's GPT-3, GPT=4 API"""
@@ -66,17 +63,29 @@ class OpenAIAdapter(LLMInterface):
     def __init__(
         self,
         api_key: str,
-        endpoint: str,
-        api_version: str,
         model: str,
-        transcription_model: str,
         max_completion_tokens: int,
+        endpoint: str = None,
+        api_version: str = None,
+        transcription_model: str = None,
         instructor_mode: str = None,
         streaming: bool = False,
         fallback_model: str = None,
         fallback_api_key: str = None,
         fallback_endpoint: str = None,
     ):
+        super().__init__(
+            api_key=api_key,
+            model=model,
+            max_completion_tokens=max_completion_tokens,
+            name="OpenAI",
+            endpoint=endpoint,
+            api_version=api_version,
+            transcription_model=transcription_model,
+            fallback_model=fallback_model,
+            fallback_api_key=fallback_api_key,
+            fallback_endpoint=fallback_endpoint,
+        )
         self.instructor_mode = instructor_mode if instructor_mode else self.default_instructor_mode
         # TODO: With gpt5 series models OpenAI expects JSON_SCHEMA as a mode for structured outputs.
         #       Make sure all new gpt models will work with this mode as well.
@@ -91,18 +100,8 @@ class OpenAIAdapter(LLMInterface):
             self.aclient = instructor.from_litellm(litellm.acompletion)
             self.client = instructor.from_litellm(litellm.completion)
 
-        self.transcription_model = transcription_model
-        self.model = model
-        self.api_key = api_key
-        self.endpoint = endpoint
-        self.api_version = api_version
-        self.max_completion_tokens = max_completion_tokens
         self.streaming = streaming
 
-        self.fallback_model = fallback_model
-        self.fallback_api_key = fallback_api_key
-        self.fallback_endpoint = fallback_endpoint
-
     @observe(as_type="generation")
     @retry(
         stop=stop_after_delay(128),
@@ -198,7 +197,7 @@ class OpenAIAdapter(LLMInterface):
                         f"The provided input contains content that is not aligned with our content policy: {text_input}"
                     ) from error
 
-    @observe
+    @observe(as_type="transcription")
     @retry(
         stop=stop_after_delay(128),
         wait=wait_exponential_jitter(2, 128),
@@ -206,58 +205,7 @@ class OpenAIAdapter(LLMInterface):
         before_sleep=before_sleep_log(logger, logging.DEBUG),
         reraise=True,
     )
-    def create_structured_output(
-        self, text_input: str, system_prompt: str, response_model: Type[BaseModel], **kwargs
-    ) -> BaseModel:
-        """
-        Generate a response from a user query.
-
-        This method creates structured output by sending a synchronous request to the OpenAI API
-        using the provided parameters to generate a completion based on the user input and
-        system prompt.
-
-        Parameters:
-        -----------
-
-            - text_input (str): The input text provided by the user for generating a response.
-            - system_prompt (str): The system's prompt to guide the model's response.
-            - response_model (Type[BaseModel]): The expected model type for the response.
-
-        Returns:
-        --------
-
-            - BaseModel: A structured output generated by the model, returned as an instance of
-              BaseModel.
-        """
-
-        return self.client.chat.completions.create(
-            model=self.model,
-            messages=[
-                {
-                    "role": "user",
-                    "content": f"""{text_input}""",
-                },
-                {
-                    "role": "system",
-                    "content": system_prompt,
-                },
-            ],
-            api_key=self.api_key,
-            api_base=self.endpoint,
-            api_version=self.api_version,
-            response_model=response_model,
-            max_retries=self.MAX_RETRIES,
-            **kwargs,
-        )
-
-    @retry(
-        stop=stop_after_delay(128),
-        wait=wait_exponential_jitter(2, 128),
-        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
-        before_sleep=before_sleep_log(logger, logging.DEBUG),
-        reraise=True,
-    )
-    async def create_transcript(self, input, **kwargs):
+    async def create_transcript(self, input, **kwargs) -> TranscriptionReturnType:
         """
         Generate an audio transcript from a user query.
 
@@ -286,60 +234,6 @@ class OpenAIAdapter(LLMInterface):
                 max_retries=self.MAX_RETRIES,
                 **kwargs,
             )
+            return TranscriptionReturnType(transcription.text, transcription)
 
-        return transcription
-
-    @retry(
-        stop=stop_after_delay(128),
-        wait=wait_exponential_jitter(2, 128),
-        retry=retry_if_not_exception_type(litellm.exceptions.NotFoundError),
-        before_sleep=before_sleep_log(logger, logging.DEBUG),
-        reraise=True,
-    )
-    async def transcribe_image(self, input, **kwargs) -> BaseModel:
-        """
-        Generate a transcription of an image from a user query.
-
-        This method encodes the image and sends a request to the OpenAI API to obtain a
-        description of the contents of the image.
-
-        Parameters:
-        -----------
-
-            - input: The path to the image file that needs to be transcribed.
-
-        Returns:
-        --------
-
-            - BaseModel: A structured output generated by the model, returned as an instance of
-              BaseModel.
-        """
-        async with open_data_file(input, mode="rb") as image_file:
-            encoded_image = base64.b64encode(image_file.read()).decode("utf-8")
-
-        return litellm.completion(
-            model=self.model,
-            messages=[
-                {
-                    "role": "user",
-                    "content": [
-                        {
-                            "type": "text",
-                            "text": "What's in this image?",
-                        },
-                        {
-                            "type": "image_url",
-                            "image_url": {
-                                "url": f"data:image/jpeg;base64,{encoded_image}",
-                            },
-                        },
-                    ],
-                }
-            ],
-            api_key=self.api_key,
-            api_base=self.endpoint,
-            api_version=self.api_version,
-            max_completion_tokens=300,
-            max_retries=self.MAX_RETRIES,
-            **kwargs,
-        )
+    # transcribe_image is inherited from GenericAPIAdapter

cognee/infrastructure/llm/structured_output_framework/litellm_instructor/llm/types.py

@@ -0,0 +1,10 @@
+from pydantic import BaseModel
+
+
+class TranscriptionReturnType:
+    text: str
+    payload: BaseModel
+
+    def __init__(self, text: str, payload: BaseModel):
+        self.text = text
+        self.payload = payload

cognee/modules/data/models/Data.py

@@ -13,7 +13,7 @@ class Data(Base):
     __tablename__ = "data"
 
     id = Column(UUID, primary_key=True, default=uuid4)
-
+    label = Column(String, nullable=True)
     name = Column(String)
     extension = Column(String)
     mime_type = Column(String)
@@ -49,6 +49,7 @@ class Data(Base):
         return {
             "id": str(self.id),
             "name": self.name,
+            "label": self.label,
             "extension": self.extension,
             "mimeType": self.mime_type,
             "rawDataLocation": self.raw_data_location,

cognee/modules/retrieval/triplet_retriever.py

@@ -36,7 +36,7 @@ class TripletRetriever(BaseRetriever):
         """Initialize retriever with optional custom prompt paths."""
         self.user_prompt_path = user_prompt_path
         self.system_prompt_path = system_prompt_path
-        self.top_k = top_k if top_k is not None else 1
+        self.top_k = top_k if top_k is not None else 5
         self.system_prompt = system_prompt
 
     async def get_context(self, query: str) -> str:

cognee/modules/retrieval/utils/brute_force_triplet_search.py

@@ -16,24 +16,6 @@ logger = get_logger(level=ERROR)
 
 
 def format_triplets(edges):
-    print("\n\n\n")
-
-    def filter_attributes(obj, attributes):
-        """Helper function to filter out non-None properties, including nested dicts."""
-        result = {}
-        for attr in attributes:
-            value = getattr(obj, attr, None)
-            if value is not None:
-                # If the value is a dict, extract relevant keys from it
-                if isinstance(value, dict):
-                    nested_values = {
-                        k: v for k, v in value.items() if k in attributes and v is not None
-                    }
-                    result[attr] = nested_values
-                else:
-                    result[attr] = value
-        return result
-
     triplets = []
     for edge in edges:
         node1 = edge.node1

cognee/tasks/ingestion/data_item.py

@@ -0,0 +1,8 @@
+from dataclasses import dataclass
+from typing import Any, Optional
+
+
+@dataclass
+class DataItem:
+    data: Any
+    label: Optional[str] = None

cognee/tasks/ingestion/ingest_data.py

@@ -20,6 +20,7 @@ from cognee.modules.data.methods import (
 
 from .save_data_item_to_storage import save_data_item_to_storage
 from .data_item_to_text_file import data_item_to_text_file
+from .data_item import DataItem
 
 
 async def ingest_data(
@@ -78,8 +79,16 @@ async def ingest_data(
         dataset_data_map = {str(data.id): True for data in dataset_data}
 
         for data_item in data:
+            # Support for DataItem (custom label + data wrapper)
+            current_label = None
+            underlying_data = data_item
+
+            if isinstance(data_item, DataItem):
+                underlying_data = data_item.data
+                current_label = data_item.label
+
             # Get file path of data item or create a file if it doesn't exist
-            original_file_path = await save_data_item_to_storage(data_item)
+            original_file_path = await save_data_item_to_storage(underlying_data)
             # Transform file path to be OS usable
             actual_file_path = get_data_file_path(original_file_path)
 
@@ -139,6 +148,7 @@ async def ingest_data(
                 data_point.external_metadata = ext_metadata
                 data_point.node_set = json.dumps(node_set) if node_set else None
                 data_point.tenant_id = user.tenant_id if user.tenant_id else None
+                data_point.label = current_label
 
                 # Check if data is already in dataset
                 if str(data_point.id) in dataset_data_map:
@@ -169,6 +179,7 @@ async def ingest_data(
                     tenant_id=user.tenant_id if user.tenant_id else None,
                     pipeline_status={},
                     token_count=-1,
+                    label=current_label,
                 )
 
                 new_datapoints.append(data_point)

cognee/tasks/ingestion/save_data_item_to_storage.py

@@ -9,6 +9,7 @@ from cognee.shared.logging_utils import get_logger
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
 from cognee.tasks.web_scraper.utils import fetch_page_content
+from cognee.tasks.ingestion.data_item import DataItem
 
 
 logger = get_logger()
@@ -95,5 +96,9 @@ async def save_data_item_to_storage(data_item: Union[BinaryIO, str, Any]) -> str
         # data is text, save it to data storage and return the file path
         return await save_data_to_file(data_item)
 
+    if isinstance(data_item, DataItem):
+        # If instance is DataItem use the underlying data
+        return await save_data_item_to_storage(data_item.data)
+
     # data is not a supported type
     raise IngestionError(message=f"Data type not supported: {type(data_item)}")