mirror of
https://github.com/hwchase17/langchain
synced 2024-11-08 07:10:35 +00:00
69344a0661
- **Description:** Added Integration with Meta AI's LASER Language-Agnostic SEntence Representations embedding library, which supports multilingual embedding for any of the languages listed here: https://github.com/facebookresearch/flores/blob/main/flores200/README.md#languages-in-flores-200, including several low resource languages - **Dependencies:** laser_encoders
90 lines
3.0 KiB
Python
90 lines
3.0 KiB
Python
from typing import Any, Dict, List, Optional
|
|
|
|
import numpy as np
|
|
from langchain_core.embeddings import Embeddings
|
|
from langchain_core.pydantic_v1 import BaseModel, Extra, root_validator
|
|
|
|
LASER_MULTILINGUAL_MODEL: str = "laser2"
|
|
|
|
|
|
class LaserEmbeddings(BaseModel, Embeddings):
|
|
"""LASER Language-Agnostic SEntence Representations.
|
|
LASER is a Python library developed by the Meta AI Research team
|
|
and used for creating multilingual sentence embeddings for over 147 languages
|
|
as of 2/25/2024
|
|
See more documentation at:
|
|
* https://github.com/facebookresearch/LASER/
|
|
* https://github.com/facebookresearch/LASER/tree/main/laser_encoders
|
|
* https://arxiv.org/abs/2205.12654
|
|
|
|
To use this class, you must install the `laser_encoders` Python package.
|
|
|
|
`pip install laser_encoders`
|
|
Example:
|
|
from laser_encoders import LaserEncoderPipeline
|
|
encoder = LaserEncoderPipeline(lang="eng_Latn")
|
|
embeddings = encoder.encode_sentences(["Hello", "World"])
|
|
"""
|
|
|
|
lang: Optional[str]
|
|
"""The language or language code you'd like to use
|
|
If empty, this implementation will default
|
|
to using a multilingual earlier LASER encoder model (called laser2)
|
|
Find the list of supported languages at
|
|
https://github.com/facebookresearch/flores/blob/main/flores200/README.md#languages-in-flores-200
|
|
"""
|
|
|
|
_encoder_pipeline: Any # : :meta private:
|
|
|
|
class Config:
|
|
"""Configuration for this pydantic object."""
|
|
|
|
extra = Extra.forbid
|
|
|
|
@root_validator()
|
|
def validate_environment(cls, values: Dict) -> Dict:
|
|
"""Validate that laser_encoders has been installed."""
|
|
try:
|
|
from laser_encoders import LaserEncoderPipeline
|
|
|
|
lang = values.get("lang")
|
|
if lang:
|
|
encoder_pipeline = LaserEncoderPipeline(lang=lang)
|
|
else:
|
|
encoder_pipeline = LaserEncoderPipeline(laser=LASER_MULTILINGUAL_MODEL)
|
|
values["_encoder_pipeline"] = encoder_pipeline
|
|
|
|
except ImportError as e:
|
|
raise ImportError(
|
|
"Could not import 'laser_encoders' Python package. "
|
|
"Please install it with `pip install laser_encoders`."
|
|
) from e
|
|
return values
|
|
|
|
def embed_documents(self, texts: List[str]) -> List[List[float]]:
|
|
"""Generate embeddings for documents using LASER.
|
|
|
|
Args:
|
|
texts: The list of texts to embed.
|
|
|
|
Returns:
|
|
List of embeddings, one for each text.
|
|
"""
|
|
embeddings: np.ndarray
|
|
embeddings = self._encoder_pipeline.encode_sentences(texts)
|
|
|
|
return embeddings.tolist()
|
|
|
|
def embed_query(self, text: str) -> List[float]:
|
|
"""Generate single query text embeddings using LASER.
|
|
|
|
Args:
|
|
text: The text to embed.
|
|
|
|
Returns:
|
|
Embeddings for the text.
|
|
"""
|
|
query_embeddings: np.ndarray
|
|
query_embeddings = self._encoder_pipeline.encode_sentences([text])
|
|
return query_embeddings.tolist()[0]
|