rag-manager/ingest_pipeline/utils/vectorizer.py

"""Vectorizer utility for generating embeddings."""

from types import TracebackType
from typing import Final, Self, cast

import httpx

from typings import EmbeddingResponse

from ..core.exceptions import VectorizationError
from ..core.models import StorageConfig, VectorConfig
from ..config import get_settings

JSON_CONTENT_TYPE: Final[str] = "application/json"
AUTHORIZATION_HEADER: Final[str] = "Authorization"


class Vectorizer:
    """Handles text vectorization using LLM endpoints."""

    endpoint: str
    model: str
    dimension: int

    def __init__(self, config: StorageConfig | VectorConfig):
        """
        Initialize vectorizer.

        Args:
            config: Configuration with embedding details
        """
        settings = get_settings()
        if isinstance(config, StorageConfig):
            # Extract vector config from global settings when storage config is provided
            self.endpoint = str(settings.llm_endpoint).rstrip("/")
            self.model = settings.embedding_model
            self.dimension = settings.embedding_dimension
        else:
            self.endpoint = str(config.embedding_endpoint).rstrip("/")
            self.model = config.model
            self.dimension = config.dimension

        resolved_api_key = settings.get_llm_api_key() or ""
        headers: dict[str, str] = {"Content-Type": JSON_CONTENT_TYPE}
        if resolved_api_key:
            headers[AUTHORIZATION_HEADER] = f"Bearer {resolved_api_key}"

        timeout_seconds = float(settings.request_timeout)
        self.client = httpx.AsyncClient(timeout=timeout_seconds, headers=headers)

    async def vectorize(self, text: str) -> list[float]:
        """
        Generate embedding vector for text.

        Args:
            text: Text to vectorize

        Returns:
            Embedding vector
        """
        if not text:
            raise VectorizationError("Cannot vectorize empty text")

        try:
            return (
                await self._ollama_embed(text)
                if "ollama" in self.model
                else await self._openai_embed(text)
            )
        except Exception as e:
            raise VectorizationError(f"Vectorization failed: {e}") from e

    async def vectorize_batch(self, texts: list[str]) -> list[list[float]]:
        """
        Generate embeddings for multiple texts.

        Args:
            texts: List of texts to vectorize

        Returns:
            List of embedding vectors
        """
        vectors: list[list[float]] = []

        for text in texts:
            vector = await self.vectorize(text)
            vectors.append(vector)

        return vectors

    async def _ollama_embed(self, text: str) -> list[float]:
        """
        Generate embedding using Ollama via OpenAI-compatible endpoint.

        Args:
            text: Text to embed

        Returns:
            Embedding vector
        """
        # Use the full model name as it appears in the API
        model_name = self.model

        # Use OpenAI-compatible endpoint for ollama models
        response = await self.client.post(
            f"{self.endpoint}/v1/embeddings",
            json={
                "model": model_name,
                "input": text,
            },
        )
        _ = response.raise_for_status()

        response_json = response.json()
        # Response is expected to be dict[str, object] from our type stub

        response_data = cast(EmbeddingResponse, cast(object, response_json))

        # Parse OpenAI-compatible response format
        embeddings_list = response_data.get("data", [])
        if not embeddings_list:
            raise VectorizationError("No embeddings returned")

        first_embedding = embeddings_list[0]
        embedding_raw = first_embedding.get("embedding")
        if not embedding_raw:
            raise VectorizationError("Invalid embedding format")

        # Convert to float list and validate
        embedding: list[float] = []
        embedding.extend(float(item) for item in embedding_raw)
        # Ensure correct dimension
        if len(embedding) != self.dimension:
            raise VectorizationError(
                f"Embedding dimension mismatch: expected {self.dimension}, received {len(embedding)}"
            )

        return embedding

    async def _openai_embed(self, text: str) -> list[float]:
        """
        Generate embedding using OpenAI-compatible API.

        Args:
            text: Text to embed

        Returns:
            Embedding vector
        """
        response = await self.client.post(
            f"{self.endpoint}/v1/embeddings",
            json={
                "model": self.model,
                "input": text,
            },
        )
        _ = response.raise_for_status()

        response_json = response.json()
        # Response is expected to be dict[str, object] from our type stub

        response_data = cast(EmbeddingResponse, cast(object, response_json))

        embeddings_list = response_data.get("data", [])
        if not embeddings_list:
            raise VectorizationError("No embeddings returned")

        first_embedding = embeddings_list[0]
        embedding_raw = first_embedding.get("embedding")
        if not embedding_raw:
            raise VectorizationError("Invalid embedding format")

        # Convert to float list and validate
        embedding: list[float] = []
        embedding.extend(float(item) for item in embedding_raw)
        # Ensure correct dimension
        if len(embedding) != self.dimension:
            raise VectorizationError(
                f"Embedding dimension mismatch: expected {self.dimension}, received {len(embedding)}"
            )

        return embedding

    async def __aenter__(self) -> Self:
        """Async context manager entry."""
        return self

    async def __aexit__(
        self,
        exc_type: type[BaseException] | None,
        exc_val: BaseException | None,
        exc_tb: TracebackType | None,
    ) -> None:
        """Async context manager exit."""
        await self.client.aclose()