MicProtocol / app /services /embedding /embedder.py
Benard John
Initial commit
42a0d15
Raw History Blame Contribute Delete
2.64 kB
"""
Ukweli — BGE-M3 Embedding Service
Generates 1024-dimensional multilingual embeddings.
Architecture Section 3.3.1.
"""
from __future__ import annotations
import logging
import numpy as np
from sentence_transformers import SentenceTransformer
logger = logging.getLogger("ukweli.embedding.embedder")
class EmbeddingService:
"""
Wraps the BAAI/bge-m3 model for generating text embeddings.
Loaded once at startup and reused for all embedding requests.
"""
def __init__(
self,
model_name: str = "BAAI/bge-m3",
dimensions: int = 1024,
batch_size: int = 32,
normalize: bool = True,
max_seq_length: int = 8192,
) -> None:
self.model_name = model_name
self.dimensions = dimensions
self.batch_size = batch_size
self.normalize = normalize
self.max_seq_length = max_seq_length
logger.info("Loading embedding model: %s", model_name)
self.model = SentenceTransformer(
model_name,
trust_remote_code=True,
)
self.model.max_seq_length = max_seq_length
logger.info(
"Embedding model loaded: dim=%d, max_seq=%d",
dimensions,
max_seq_length,
)
def embed_texts(self, texts: list[str]) -> list[list[float]]:
"""
Generate embeddings for a list of texts.
Args:
texts: List of text strings to embed.
Returns:
List of embedding vectors (each a list of floats).
"""
if not texts:
return []
all_embeddings: list[list[float]] = []
for i in range(0, len(texts), self.batch_size):
batch = texts[i: i + self.batch_size]
embeddings = self.model.encode(
batch,
normalize_embeddings=self.normalize,
show_progress_bar=False,
batch_size=self.batch_size,
)
# Convert numpy arrays to plain lists
for emb in embeddings:
if isinstance(emb, np.ndarray):
all_embeddings.append(emb.tolist())
else:
all_embeddings.append(list(emb))
logger.debug("Embedded %d texts in %d batches.", len(texts), -(-len(texts) // self.batch_size))
return all_embeddings
def embed_query(self, query: str) -> list[float]:
"""
Generate an embedding for a single query string.
Uses the same model but semantically this is for retrieval queries.
"""
result = self.embed_texts([query])
return result[0]